University of Nebraska - Lincoln

DigitalCommons@University of Nebraska - Lincoln

Mechanical (and Materials) Engineering -- Mechanical & Materials Engineering, Department
Dissertations, Theses, and Student Research of

11-2008

Real-Time Vertical Track Deflection Measurement

System
Sheng Lu
University of Nebraska - Lincoln, [email protected]

Follow this and additional works at: http://digitalcommons.unl.edu/mechengdiss

Part of the Mechanical Engineering Commons

Lu, Sheng, "Real-Time Vertical Track Deflection Measurement System" (2008). Mechanical (and Materials) Engineering -- Dissertations,
Theses, and Student Research. 2.
http://digitalcommons.unl.edu/mechengdiss/2

This Article is brought to you for free and open access by the Mechanical & Materials Engineering, Department of at DigitalCommons@University of
Nebraska - Lincoln. It has been accepted for inclusion in Mechanical (and Materials) Engineering -- Dissertations, Theses, and Student Research by an
authorized administrator of DigitalCommons@University of Nebraska - Lincoln.
 REAL-TIME VERTICAL TRACK DEFLECTION

MEASUREMENT SYSTEM

by

Sheng Lu

A DISSERTATION

Presented to the Faculty of

The Graduate College at the University of Nebraska
In Partial Fulfillment of Requirements
For the Degree of Doctor of Philosophy

Major: Engineering
(Mechanical Engineering)

Under the Supervision of Professor Shane M. Farritor

Lincoln, Nebraska

December, 2008
 REAL-TIME VERTICAL TRACK DEFLECTION MEASUREMENT SYSTEM

Sheng Lu, Ph.D.

University of Nebraska, 2008

Adviser: Shane M. Farritor

Track quality is a major factor in railroad safety, and one accepted indicator of

track quality is the vertical track deflection. Measuring track deflection from a moving

railcar is difficult because there is no stable reference for the measurements.

A system developed by researchers at the University of Nebraska to measure track

deflection from a moving railcar in real-time is described in detail. The system consists

of a loaded hopper with a camera/laser sensor system to detect the vertical deflection of

the rail relative to the wheel/rail contact point. Modeling and simulation of the system

is also presented along with the mathematical models which can be used to estimate track

modulus.

The measurement system has been used to conduct revenue service tests over

three thousand miles of track. A special validation test was also performed. The

results from these tests have shown that the system’s measurement is repeatable and

accurate; the system has notable ability to indicate track support problems.
 I

ACKNOWLEDGEMENT

I would like to thank Dr. Shane Farritor, who has been a wonderful advisor and

mentor. Without his guidance, none of these will be possible. Thanks to Dr. Richard

Arnold who has also been a great teacher. I would also like to thank my colleagues Curt

Greisen, Haoliang Duan and Cory Hogan for the help and supports during the research.

Thanks to my parents and my sister, for supporting me throughout my life.

Their supports and love have been the major motivation and inspiration of my life.

This project is funded by Federal Railway Administration (FRA). Thanks to

UPRR and BNSF for operational support and track access for testing.
 II

Table of Contents
ACKNOWLEDGEMENTS…………………………………………...….……………… I

TABLE OF CONTENTS………………………………………………………………....II

LIST OF FIGURES………………………………………………………………...….....V

1  INTRODUCTION ...................................................................................1 
2  BACKGROUND ......................................................................................5 
2.1  Problem Definition ................................................................................................... 5 
2.2  Beam on Elastic Foundation Model ......................................................................... 7 
2.3  Nonlinear Cubic Model ............................................................................................ 9 
2.4  Methods for Determining Track Modulus ............................................................. 15 
2.4.1  Beam On Elastic Foundation (BOEF) Method ................................................ 16 
2.4.2  Deflection Basin Method ................................................................................. 17 
2.4.3  Heavy-Light Load Method............................................................................... 18 
2.4.4  Track Modulus at Characteristic Load ............................................................. 20 
2.5  Factors Influencing Track Modulus ....................................................................... 22 
3  SYSTEM MODELING AND SIMULATION ....................................24 
3.1  Measurement Principle and Methodology ............................................................. 24 
3.2  The Relation between Yrel and Modulus (Winkler Model) .................................. 28 
3.3  Effects of Track Geometry ..................................................................................... 33 
3.4  Eliminating the Effects of Track Geometry Variation ........................................... 38 
3.4.1  10-ft ECO (End-Chord Offset) Calculation from Rail Profile ......................... 38 
3.4.2  Subtracting -ECO from Yrel ............................................................................ 39 
3.5  Stress and Strain on the Rail .................................................................................. 41 
3.6  Different Loads ...................................................................................................... 45 
4  MEASUREMENT SYSTEM ................................................................47 
4.1  Instrumentation ...................................................................................................... 47 
4.1.1  Sensor System .................................................................................................. 48 
4.1.2  GPS and Encoder ............................................................................................. 54 
 III

4.1.3  Power Supply and Management System .......................................................... 56 

4.1.4  Remote Supervision and Data Downloading ................................................... 57 
4.1.5  Testing Results Database ................................................................................. 58 
4.2  Real-time Image Processing ................................................................................... 60 
4.2.1  Basic Algorithm ............................................................................................... 60 
4.2.2  Verification Using Neural Networks ............................................................... 61 
4.3  Calibration Approach and Procedure ..................................................................... 65 
4.3.1  Finding the Ratio of Δd and ΔYrel (R) ............................................................. 67 
4.3.2  Verifying the Ratio of Δd and ΔYrel (R) .......................................................... 68 
4.3.3  Finding Yrel* and d* ........................................................................................ 71 
4.3.4  Verifying Calibration Results .......................................................................... 75 
4.4  Error Analysis ........................................................................................................ 77 
4.4.1  Errors Caused by Limited Sampling Rate........................................................ 77 
4.4.2  Measurement Resolution.................................................................................. 78 
4.4.3  Laser Line Width ............................................................................................. 79 
4.4.4  Laser Beam Drifting......................................................................................... 80 
5  FIELD TESTING AND DATA ANALYSIS .......................................82 
5.1  Revenue Service Testing ........................................................................................ 82 
5.2  Measurement Repeatability.................................................................................... 86 
5.3  System Measurement under Various Conditions ................................................... 87 
5.3.1  Results From Different Train Speeds............................................................... 88 
5.3.2  Results From Different Seasons ....................................................................... 89 
5.3.3  Results from Different Rail Sizes .................................................................... 91 
5.4  Measurement Validation ........................................................................................ 92 
5.4.1  String Measurements........................................................................................ 92 
5.4.2  Survey Measurements ...................................................................................... 94 
5.4.3  Absolute Deflection Measurements by Cameras ............................................. 98 
5.5  Trending Analysis ................................................................................................ 101 
5.5.1  Data Pre-processing ....................................................................................... 101 
5.5.2  A Trending Example ...................................................................................... 104 
5.5.3  Trending Results for Bridge Approaches....................................................... 108 
 IV

5.6  Implementation of Track Geometry Data ............................................................ 110 

5.7  Exception Criteria ................................................................................................ 113 
5.7.1  Mathematical Formulation of Exception Criteria .......................................... 113 
5.7.2  Prioritized Exception Results ......................................................................... 115 
5.7.3  Consequences of the Exception Criteria ........................................................ 116 
5.7.4  Using “Yrel+ECO” as an Exception Criterion .............................................. 118 
5.8  Comparison between Different Measurement Systems ....................................... 119 
5.8.1  Comparison of Vertical Track Deflection (VTD) with Vehicle-Track
Interaction (VTI) ............................................................................................ 119 
5.8.2  Comparison of Revenue Service Results from VTD and VTI ....................... 120 
5.8.3  Comparison of Revenue Service Results of Yrel and ECO ........................... 122 
5.9  Field Investigations .............................................................................................. 126 
5.9.1  Site 1: A Crushed Rail Head .......................................................................... 126 
5.9.2  Site 2: A Muddy Crossing .............................................................................. 128 
5.9.3  Site 3: A Failing Joint .................................................................................... 130 
5.9.4  Site 4: A Series of Broken Ties ...................................................................... 132 
5.10 Modulus Estimation ............................................................................................. 135 
6  CONCLUSION ....................................................................................137 
7  APPENDIX Program Flowchart........................................................139 
8  References .............................................................................................140 

 
 V

LIST OF FIGURES

Figure 2-1: Free-body Diagram of the Rail ........................................................................ 6 

Figure 2-2: Boundary Conditions of the Rail ..................................................................... 7 
Figure 2-3: Relative Rail Displacement for Different Loads under a Railcar .................... 9 
Figure 2-4: Deflection of Track under Three Loads ......................................................... 10 
Figure 2-5: Experimental Data (Zarembski and Choros, 1980) and Curve Fitting .......... 12 
Figure 2-6: Comparison of Cubic and Winkler Models ................................................... 15 
Figure 2-7: Deflection Basin (Selig and Li, 1994) ........................................................... 17 
Figure 2-8: Piece-wise Linear Approximation for Track Load-Deflection Behavior ...... 19 
Figure 2-9: Modulus Calculations in Winkler and Cubic Model ..................................... 21 
Figure 2-10: Track Structure (Chang et al. 1980) ............................................................. 22 
Figure 3-1: Diagram of Measurement Principle ............................................................... 25 
Figure 3-2: Camera/Laser System .................................................................................... 25 
Figure 3-3: Sensor (Lasers and Camera) Geometry ......................................................... 27 
Figure 3-4: Superposition of the Deflections from Two Loads ........................................ 29 
Figure 3-5: Relation between Yrel and Modulus (Winkler model) .................................. 32 
Figure 3-6: Relation between the Total Deflection and Yrel (Winkler model) ................ 33 
Figure 3-7: An Example Site with Both Significant Unloaded Geometry and Low Track
Stiffness ......................................................................................................... 34 
Figure 3-8: Simulation on the Effects of Track Geometry ............................................... 36 
Figure 3-9: Effects of Unloaded Geometry of Various Length (L) and Depth (d) ........... 37 
Figure 3-10: 10-ft ECO Calculation ................................................................................. 38 
Figure 3-11: Deflection Calculation ................................................................................. 40 
Figure 3-12: Relation between Rail Bending Moment and Yrel ...................................... 42 
Figure 3-13: Relation between Rail Bending Stress (RE132, bottom) and Yrel .............. 43 
Figure 3-14: Relation between Rail Strain (RE132, bottom) and Yrel............................. 44 
Figure 3-15: Effect of Different Loads on Yrel-Modulus Relation .................................. 46 
Figure 4-1: System Instrumentation.................................................................................. 48 
Figure 4-2: Hopper Car Truck and Rigid Beams Assembly ............................................. 49 
Figure 4-3: The Rigid Beam on the Side Frame ............................................................... 50 
 VI

Figure 4-4: Sensor Head Assembly .................................................................................. 51 

Figure 4-5: Shrouds Attached to the Rigid Beams ........................................................... 52 
Figure 4-6: Typical Test Image......................................................................................... 52 
Figure 4-7: Enclosed Box for Computers ......................................................................... 54 
Figure 4-8: Power Supply System .................................................................................... 57 
Figure 4-9: Power Supply System Monitoring Information in the April 2008 Test ......... 58 
Figure 4-10: Database Website Screenshot ...................................................................... 59 
Figure 4-11: Exception Locations List from the Website ................................................. 59 
Figure 4-12: Reproduced Laser Curves ............................................................................ 60 
Figure 4-13: An Imperfect Image Example ...................................................................... 61 
Figure 4-14: Various Images Captured by the Camera .................................................... 62 
Figure 4-15: Neural Network Structure ............................................................................ 64 
Figure 4-16: Processing Results without Using a Neural Network .................................. 64 
Figure 4-17: Processing Results Using Neural Network Verification .............................. 65 
Figure 4-18: Converting Number of Pixels into Distance in Inches ................................. 66 
Figure 4-19: Calibration (Geometry of Sensor Head and Rail) ........................................ 68 
Figure 4-20: Calibration Plate on the Top of the Rail ...................................................... 69 
Figure 4-21: Calibration Plate on Top of the Rail (Side View) ........................................ 69 
Figure 4-22: Captured Image of the Calibration Plate ...................................................... 69 
Figure 4-23: Calibration Results ....................................................................................... 70 
Figure 4-24: Capturing the Rail Deflection with Video Camera ...................................... 72 
Figure 4-25: Captured Video Showing the Rail Deflection.............................................. 73 
Figure 4-26: The Deflection Curve of the Rail from Calibration ..................................... 74 
Figure 4-27: Captured Image when Sensor Head Passes by the Marker .......................... 75 
Figure 4-28: Yrel Data from the Mechanical Shop .......................................................... 76 
Figure 4-29: Limited Sampling Rate Causing Measurement Errors ................................ 78 
Figure 4-30: Laser Line Width ......................................................................................... 80 
Figure 4-31: Laser Beam Drifting..................................................................................... 81 
Figure 5-1: System in Revenue Service Testing ............................................................... 83 
Figure 5-2: Yrel Data Overlaid on a Satellite Map ........................................................... 84 
Figure 5-3: A Rough High-Speed Crossover .................................................................... 85 
 VII

Figure 5-4: Track with Consistent Modulus ..................................................................... 86 

Figure 5-5: Measurements from Multiple Tests ............................................................... 87 
Figure 5-6: Measurements from Different Testing Speeds............................................... 88 
Figure 5-7: Dynamic Loads Effecting Measurements ...................................................... 89 
Figure 5-8: Variations of the Measurements .................................................................... 90 
Figure 5-9: Variations in Some Sections of Track............................................................ 91 
Figure 5-10: Average Yrel from Three Tests on Different Size Rail ............................... 92 
Figure 5-11: String Measurement Diagram ...................................................................... 93 
Figure 5-12: Field String Measurement ............................................................................ 93 
Figure 5-13: Instruments Used in Survey Measurements ................................................. 95 
Figure 5-14: Measurement of Vertical Rail Position by Surveying ................................. 96 
Figure 5-15: Survey Measurement Results ....................................................................... 97 
Figure 5-16: Wayside Camera Measurements Setup ........................................................ 98 
Figure 5-17: Sample Data of Absolute Deflection from Wayside Cameras ..................... 99 
Figure 5-18: Deflection Data from Camera Measurement ............................................. 100 
Figure 5-19: The Original Data from Two Tests ............................................................ 102 
Figure 5-20: Cross Correlation ....................................................................................... 103 
Figure 5-21: The Shifted Data from Two Tests .............................................................. 104 
Figure 5-22: Data from 3 Tests at MP A.74 ................................................................... 105 
Figure 5-23: Trending at MP A.74 and A.76 .................................................................. 106 
Figure 5-24: Test Data at MP A.74 as a Function of Time............................................. 107 
Figure 5-25: Trending from MP A.70 to A.74................................................................ 108 
Figure 5-26: A Bridge at MP B.6.................................................................................... 109 
Figure 5-27: A Bridge at MP C.63.................................................................................. 110 
Figure 5-28: Data alignment for ECO and Yrel data ...................................................... 112 
Figure 5-29: Site of Broken Field Weld 14 Days after Test ........................................... 117 
Figure 5-30: Failed Non-Insulated Joint 30 days post-test ............................................. 118 
Figure 5-31: Data at the Crushed Rail Head Site ............................................................ 127 
Figure 5-32: A Crushed Rail Head ................................................................................. 128 
Figure 5-33: Data at the Muddy Crossing Site ............................................................... 129 
Figure 5-34: The Muddy Crossing .................................................................................. 130 
 VIII

Figure 5-35: Data at the Failing Joint Site ...................................................................... 131 

Figure 5-36: The Failing Insulated Joint ......................................................................... 132 
Figure 5-37: Data at the Broken Ties Site ...................................................................... 133 
Figure 5-38: Six Broken Ties in a Row .......................................................................... 134 
Figure 5-39: Track Taken Out of Service ....................................................................... 134 
Figure 5-40: Track Modulus Calculated from Relative Deflection Data ....................... 136 
 1

1 INTRODUCTION

The economic constraints of both passenger and freight railroad traffic are

moving the railroad industry to higher-speed vehicles and higher axle loads. The heavy

axle loads and high speeds of modern freight trains produce high track stresses leading to

quicker deterioration of track condition. As a result, the need for track maintenance

increases. Fast and reliable methods are needed to identify and prioritize track in need

of maintenance in order to minimize delays, avoid derailments, and reduce maintenance

costs.

The condition and performance of railroad track depends on a number of different

parameters. Some of the factors that influence track quality are track modulus, internal

rail defects, profile, cross-level, gage, and gage restraint. Monitoring these parameters

can improve safe train operation by identifying track locations that produce poor vehicle

performance or derailment potential. Track monitoring also provides information for

optimizing track maintenance activities by focusing activities where maintenance is

critical and by selecting more effective maintenance and repair methods.

Automated methods of inspection are available for most of the parameters that

are included in track geometry (Li et al, 2002). An example of an automated vehicle is

the Federal Railroad Administration’s (FRA) high-speed track geometry vehicle, referred

to as the T-18. The T-18 is capable of measuring rail head profile, gage restraint, and

other track geometry parameters (Research Results, 2001). Another example is an

ultrasonic and inductive test vehicle capable of 60 mph produced by Sperry (Wanek,

2004).
 2

However, at the present time, there is no vehicle available to measure one of the

most important parameters – track modulus at normal track speeds in real-time. Track

modulus is defined as the coefficient of proportionality between the rail deflection and the

vertical contact pressure between the rail base and track foundation (Cai et al, 1994). In

other words, track modulus is the supporting force per unit length of rail per unit rail

deflection (Selig and Li, 1994). Track modulus is a single parameter that represents the

effects of all of the track components under the rail (Cai et al, 1994). These components

include the subgrade, ballast, subballast, ties, and tie fasteners.

Track modulus is important because it significantly affects track performance

and maintenance requirements. Both low track modulus and large variations in track

modulus are undesirable. Low track modulus has been shown to cause differential

settlement that subsequently increases maintenance needs (Read et al, 1994; Ebersohn et

al, 1993). Large variations in track modulus, such as those often found near bridges and

crossings, have been shown to increase dynamic loading (Zarembski and Palese, 2003;

Davis et al, 2003). Increased dynamic loading reduces the life of the track components,

resulting in shorter maintenance cycles (Davis et al; 2003). It has been shown that

reducing variations in track modulus at grade (i.e. road) crossings leads to better track

performance and less track maintenance (Zarembski and Palese, 2003). It has also been

suggested that track with a high and consistent modulus will allow for higher train speeds

and therefore increase both performance and revenue (Heelis et al, 1999). Ride quality,

as indicated by vertical acceleration, is also strongly dependent on track modulus.

Previous localized field testing has shown that it is possible to measure areas of

low track modulus, variable track modulus, void deflection, variable total deflection, and
 3

inconsistent rail deflection (Sussmann et al., 2001; Ebersohn and Selig, 1994). In the

past, such systems have been used to identify sections of track with poor performance.

These measurements have been useful. However, they are expensive and have only

been made over short distances (in the range of tens of meters). The ability to make

these measurements continuously over large sections of track is desirable (Ebersohn and

Selig, 1994; Read et al., 1994).

Previous onboard track modulus measurement systems are similar to systems

developed by the military and used in highway research (Carr, 1999). The systems use

a long rigid truss that rides on two unloaded wheels. This truss creates a straight line, or

cord, that is used as a reference for the measurement. A third wheel is then used to

apply a load at the midpoint of the cord (or truss) and the relative displacement between

the loaded wheel and the unloaded truss is measured. The truss must be long enough so

that the two endpoints are not affected by the load at the center of the truss. This

method requires two measurements, one with a light load, made with a similar truss, and

one with a heavy load, to distinguish between changes in geometry and changes in

modulus. The output of this approach is a measurement of the relative displacement of

the loaded wheel with respect to the unloaded wheel. Using this measurement, track

modulus is then estimated.

One vehicle, called the Track Loading Vehicle (TLV), uses this approach

(Thompson and Li, 2002). This vehicle is capable of measuring track modulus at speeds

up to 16.1 km/hr (10 mph). The TLV uses two cars, each with a center load bogie

capable of applying loads from 4.45 kN to 267 kN (1 to 60 kips). A light load (13.3 kN

or 3 kips) is applied by the first vehicle while a heavier load is applied by the second
 4

vehicle. A laser-based system on each vehicle measures the deflections of the rail

caused by the center load bogies. The test procedure involves two passes over a section

of track – first applying a 44.5 kN (10 kip) load and then a 178 kN (40 kip) load

(Thompson and Li, 2002).

Although the TLV is operational, it does have limitations. First, tests are often

performed at speeds below 16.1 km/hr (10 mph) so it is difficult to test long sections of

track (hundreds of miles). Second, significant expense in both equipment and personnel

is required for operation. For these reasons the TLV has not yet been widely

implemented.

This thesis presents a method to measure vertical track deflection from a moving

railcar. These deflection measurements can then be used to estimate track modulus. The

system uses a non-contact vision sensor system to make displacement measurements with

respect to the wheel/rail contact point. The system is inexpensive and does not require

significant support equipment and personnel. The system is capable of automated

testing and operation at higher speeds.

 5

2 BACKGROUND

The relationship between applied loads and track deformations is an important

parameter to be considered in proper track design and maintenance. A representative

mathematical model that accurately describes this relationship is desirable.

2.1 Problem Definition

Figure 2-1 shows a free-body diagram of the rail under a one-wheel load. The rail

is considered as a continuously supported beam where x represents the distance along the

beam and w(x) represents the vertical beam deflection. The approximation that the rail

is continuously supported improves as the cross-tie spacing decreases and as the rail

bending stiffness increases. The applied load, P, is assumed to be a point load and

0+
creates a vertical load on the rail, q(x), where P = ∫ − q( x)dx . The supporting structure
0

supports the bottom of the rail with a reaction distributed force, p(x). In real track, the

supporting structure consists of tie plates, fasteners, cross-ties, ballast, etc. In this model

the supporting structure is an infinite medium.

 6

Figure 2-1: Free-body Diagram of the Rail

The difference in the vertical distributed force applied to the beam (q(x) - p(x))

causes curvature in the beam as given by the following differential equation:

d 4w
EI 4 + p( x) = q( x) Equation 2-1
dx

where:

E is the modulus of elasticity of the rail

I is the moment of inertia of the rail

x is the longitudinal distance along the rail

The solution to the differential equation is dependent upon the boundary

conditions of the beam as well as the loading conditions. A free body diagram that shows

sections of the beam is shown in Figure 2-2. The figure shows that the concentrated
 7

applied load P must be supported by the foundation reaction distributed force p(x) on

each half of the infinite beam such that:

∞ P
∫0
p( x)dx =
2 Equation 2-2

In addition, symmetry and the stiffness of the beam demand that the slope of the

beam be zero at the point of loading.

dw
=0 Equation 2-3
dx x=0

Figure 2-2: Boundary Conditions of the Rail

The above differential equation and boundary conditions can now be set up and

solved in different ways to represent various track behaviors.

2.2 Beam on Elastic Foundation Model

The Beam on an Elastic Foundation (BOEF) model, proposed by Winkler (1867),

describes a point load applied to an infinite beam on an infinite elastic foundation. It

assumes the distributed supporting force of the track foundation is linearly proportional to
 8

the vertical rail deflection (i.e. p(x)=uw(x) ). Here, the coefficient u is defined as track

modulus. The differential Equation 2-1 then becomes

d 4 w( x)
EI + uw( x) = q( x) Equation 2-4
dx 4

This model has been shown to be an effective method for determining track

modulus (Raymond, 1985; Meyer, 2002) and derivations can be found in (Kerr, 1976;

Boresi and Schmidt, 2003). The vertical deflection of the rail, w, as a function of

longitudinal distance along the rail x (referenced from the position of the applied load) is

given by:

Pβ − β x
w( x) = − e [cos( β x ) + sin( β x )] Equation 2-5
2u

where: 1
⎛ u ⎞4
β =⎜ ⎟ Equation 2-6
⎜ 4 EI ⎟
⎝ ⎠

P is the load on the track

u is the track modulus

When multiple loads are applied, the rail deflections caused by each of the loads

are superposed (assuming small vertical deflections) (Boresi and Schmidt, 2003).

A plot of the rail deflection given by the Winkler model over the length of a

four-axle coal hopper is shown in Figure 2-3. The deflection is shown relative to the

wheel/rail contact point for five different reasonable values of track modulus (6.89, 13.8,

20.7, 27.6, and 34.5 MPa corresponding to 1000, 2000, 3000, 4000, and 5000 psi
 9

respectively). This example assumes 115 lb rail with an elastic modulus of 206.8 GPa

(30,000,000 psi) and an area moment of inertia of 2704 cm4 (64.97 in4).

Figure 2-3: Relative Rail Displacement for Different Loads under a Railcar

The limitations of the Winkler model are clear given the widely accepted

non-linearity of track structure. However, this model is often used because it does

provide a clear closed-form solution to the relationship between load and deflection in

track structure.

2.3 Nonlinear Cubic Model

Field tests conducted by the ASCE-AREA Special Committee on Stresses in

Railroad Track (1918) clearly showed that the vertical rail deflections were not linearly
 10

proportional to the wheel loads. An extensive experimental study conducted by

Zarembski and Choros (1980) also clearly documented this nonlinear response.

Figure 2-4 shows the experimental results of the track responses under various

applied loads. Rail deflection was measured at given locations using linear variable

differential transformers (LVDTs) as a short, slow moving train of known weight passed.

The axles of the train carried 150,600 N (33850 lbf), 60,230 N (13540 lbf), and 30650 N

(6890 lbf). The LVDTs were mounted to steel rods (about 1m (3ft)) driven into the

subgrade to provide a stable reference. The LVDTs then measured the vertical motion of

the flange relative to the steel rod. The results from four LVDTs are shown in Figure

2-4. Here the LVDTs were placed at 1m (3ft) increments along the track (x=1m, 2m,

3m, 4m).

Figure 2-4: Deflection of Track under Three Loads

 11

These measurements, along with many others dating back to the Talbot Report

(ASCE-AREA Special Committee, 1918) clearly indicate that the vertical rail deflections

are not linearly proportional to the wheel loads. It is also important to note that the

“degree” of non-linearity can change dramatically over very short distances along the

track. Note the deflection of the track under the 30650 N (6890 lbf) load increased

about 60% over a distance of one meter. This non-linearity and variability greatly

complicates determining and modeling track structure. Several methods have been

developed for calculating modulus with each method assuming a different definition of

track modulus that approximates the non-linear behavior of real track.

Here, a new model is proposed that represents the relationship between vertical

rail deflection and the distributed rail support force as a cubic polynomial. To define

this relationship the experimental results of (Zarembski and Choros, 1980) are plotted in

Figure 2-5 along with a cubic polynomial curve fit. The polynomial fits the

experimental results very well (R2=0.9987).

Using a cubic polynomial has several advantages. First, it clearly captures the

behavior of real track in that it provides for low stiffness at low loads and higher stiffness

at higher loads. Also, negative displacement of the track (track lift) does not result in

significant downward forces being applied to the rail. Unlike the previous models, the

cubic polynomial closely represents the fact that if the track rises slightly, the ballast does

not pull the track down.

 12

Figure 2-5: Experimental Data (Zarembski and Choros, 1980) and Curve Fitting

Here, the supporting distributed load p(x) has a cubic relationship between p(x)

and w(x):

p ( x) = u1 w( x) + u 3 w 3 ( x) Equation 2-7

Note that symmetry about the applied load requires the second order term to

vanish. Substitution into the BOEF model gives the following differential equation:

d 4w
EI 4 + u1 w + u 3 w 3 = q Equation 2-8
dx

Equation 2-8 is a nonlinear differential equation, and a closed form analytical

solution is not straightforward. One analytical approximation based on Cunningham’s

 13

method can be found in McVey (2006). However, a numerical solution for this

boundary value problem (BVP) can be obtained.

The BVP can be written in state space notation as:

⎡ w( x) ⎤
⎢ ⎥
∂ w′( x) ⎥
w′ = ⎢ = func( w, x) Equation 2-9
∂x ⎢ w′′( x) ⎥
⎢ ⎥
⎣ w′′′( x)⎦

Given Equation 2-9 the BVP becomes:

⎡ w( x) ⎤ ⎡ w′( x) ⎤
⎢ ⎥ ⎢ w′′( x) ⎥
∂ ⎢ w′( x ) ⎥ ⎢ ⎥
=⎢ w′′′( x ) ⎥ Equation 2-10
∂x ⎢ w′′( x) ⎥ ⎢
⎢ ⎥ − 1 u w( x) + u w 3 ( x) ⎥
( )
⎣ w′′
′ ( x ) ⎦ ⎢⎣ EI 1 3 ⎥
⎦

As the name implies, the fourth order BVP described above requires the values of

four boundary conditions which are displayed in the following equations:

w( x) | x =∞ = 0
w( x) | x = −∞ = 0
w′( x) | x =0 = 0 Equation 2-11
w( x) | x =0 = wo

Now, since the BVP can have more than one correct solution, an initial “guess”

for the last boundary condition is needed in order for the solution to converge to an

expected solution. In this case, the initial guess is provided by the Winkler model

evaluated at x=0 and u=u3.

 14

Pβ ⎛ u ⎞4
w(0) = wo = − where : β = ⎜ 3 ⎟ Equation 2-12
2u 3 ⎜ 4 EI ⎟
⎝ ⎠

The mechanics of this problem also require that the solution be found subject to

the additional constraint given by the free body diagram in Figure 2:

∫ (u w + u w )dx = 2
∞
3 P
0
1 3 Equation 2-13

The unique solution that satisfies each of these constraints will give the rail

deflection. Many numerical techniques can be used to solve this well-posed BVP. In

this work the “bvp4c” function in Matlab (Kierzenka J. and Shampine L. F., 2001) was

used.

While the cubic model closely represents the deflection test data over the entire

range of wheel loads, the accuracy of the linear analysis depends on the magnitude of the

test load.

Since the cubic spring is initially softer than the one in the Winkler model, the rail

must deflect more before the base can pick up the full load. This means that the

distributed load will be spread over a wider span than for the linear model as shown in

Figure 2-6. Meanwhile, the deflection at the contact point for the cubic model is slightly

larger than the one for the Winkler model when the applied load is relatively large.

Although the cubic model represent the real track response more accurately, it is

not easy to apply. In other words, for a given load and known maximum deflection, it is

not possible to depict the actual track response. However, modulus can be simply

calculated from the applied load and maximum deflection using Winkler model.
 15

Therefore, the application of cubic model is limited and the Winkler model is widely

used in the industry.

Figure 2-6: Comparison of Cubic and Winkler Models

2.4 Methods for Determining Track Modulus

Many methods and models have been developed and used to determine track

modulus. The methods include the Beam on Elastic Foundation method, Deflection

Basin method. In these methods, a static load must be applied to the rail, and rail

deflection measurements must be made before and after the load is applied. Focused on

the nonlinear cubic model, a new method is proposed to calculate modulus at

characteristic load.
 16

2.4.1 Beam On Elastic Foundation (BOEF) Method

The most straightforward method to estimate track modulus at a given track

location is to simply measure the vertical deflection at the point of an applied known

load, P. This produces the measurement of the track stiffness, k, which is the ratio

between applied load and maximum deflection under the wheel, but this measurement

can be related to track modulus, u, using the BOEF model and assuming that the

relationship between rail supporting load p(x) and deflection w(x) is linear and elastic (i.e.

p(x)=uw(x) as in Selig and Li, 1994; Cai et al, 1994). These assumptions lead to the

Winkler model. The resulting track modulus is given by:

1 4
1 ⎛ 1 ⎞3 ⎛ P ⎞ 3
u = ⎜ ⎟ ⎜⎜ ⎟⎟ Equation 2-14
4 ⎝ E1 ⎠ ⎝ w0 ⎠

where: u is the track modulus

E is the modulus of elasticity of the rail

I is the moment of inertia of the rail

P is the load applied to the track

w0 is the deflection of the rail at the loading point

This method only requires a single measurement and it has also been suggested to

be the best method for field measurement of track modulus (Zarembski and Choros,

1980). However, as shown in Figure 2-4, it is clear that this linear approximation has

large error for real track. Using a single applied load and a single measurement of

deflection does not capture the changes in the load-deflection curve present in real track.
 17

2.4.2 Deflection Basin Method

The Deflection Basin Method uses the vertical equilibrium of the loaded rail and

several deflection measurements to estimate track modulus more directly. In this

approach, rail deflection caused by point loads is measured at several (ideally infinite)

locations along the rail and the entire deflected “area” calculated. The deflection basin

for two applied loads is shown in Figure 2-7.

Figure 2-7: Deflection Basin (Selig and Li, 1994)

Using the concept of vertical equilibrium, the applied load can be shown to be

proportional to the integral of the rail deflection or the deflection basin area (Selig and Li,

1994; Cai et al, 1994). For a single applied load, the equation is:

Equation 2-15

where: P is the load on the track

q(x) is the vertical supporting force per unit length

 18

u is the track modulus

w(x) is the vertical rail deflection

Aw is the deflection basin area

x is the longitudinal distance along the track

This method requires several deflection measurements over the section of track

that supports the load(s), which makes it very time consuming (Selig and Li, 1994).

When using this method, the non-linearity caused by slack in the rail should be removed

by using a light load for the base measurements and a heavy load as deflected

measurements. This method is not recommended because it is more time consuming and

is based on some questionable assumptions (Kerr and Shenton, 1985).

2.4.3 Heavy-Light Load Method

Many have represented the load/defection curve as piece-wise linear with a low

stiffness at low loads and a much higher stiffness at higher loads (Kerr and Shenton,

1986). This is seen in real track as slack in the rail and can be caused by many things

such as the ties not contacting the ballast. As the rail is loaded, a low stiffness is

experienced until the tie contacts the ballast, resulting in a higher stiffness. This leads to

a measurement of track stiffness using two loads as shown in Figure 2-8, that are ideally

both in the high stiffness range (e.g. slack is removed) (Ebersohn and Selig, 1994; Read

et al, 1994, Kerr, 2003).

The Equation 2-16 demonstrated how to calculate track stiffness by using the two

different loads (seating load and full load). This calculated track stiffness can then be

related to track modulus.

 19

Figure 2-8: Piece-wise Linear Approximation for Track Load-Deflection Behavior

P2 − P1
k= Equation 2-16
w2 − w1

where: k is the track stiffness

Pi are the applied loads

wi are the corresponding deflections

Again, a linear assumption is used to transform the stiffness measurements of the

two loads to track modulus. The clear difficulty with this measurement is that the real

load/deflection relationship is not piecewise linear and the resulting stiffness varies with
 20

the selection of the two loads, P1 and P2. It is also clear that any two choices of loads

will give a different value of track modulus.

2.4.4 Track Modulus at Characteristic Load

It is proposed that a good definition of track modulus is the variation in

supporting distributed force relative to the variation in deflection near the characteristic

load for a given track (Lu et al., 2008). This characteristic load might be defined as the

nominal axle load for a given freight line (e.g. 160kN or 286,000/8=36kips). This can

be expressed mathematically as the derivative of the pressure-deflection curve evaluated

at the characteristic load P*:

∂p
u* = Equation 2-17
∂w
P*

where: u is the track modulus

p is the supporting force per unit length of rail

P* is the characteristic load corresponding to a given rail line

To evaluate the derivative at the characteristic load, the load must again be

transformed to a distributed load. This can be done with the linear assumptions as

described previously (the Winkler model). This definition of track modulus has been

used in field measurements (Arnold et al., 2006).

Finally, in the nonlinear cubic model described previously, the track modulus at

characteristic load can be calculated as:

 21

u* =
∂p
=
(
∂ u1 w + u 3 w 3 ) = u1 + 3u 3 w 2
∂w ∂w P* Equation 2-18
P* P*

This definition of track modulus is compared to the Winkler model as shown in

Figure 2-9. In this figure, the load-deflection curve is plotted from the experimental

data of (Zarembski and Choros, 1980) shown in Figure 2-5.

Figure 2-9: Modulus Calculations in Winkler and Cubic Model

It is clear that for single data points at higher loads the Winkler model will always

underestimate the actual track modulus (Figure 2-9). The Winkler model will also poorly

represent changes in deflection with respect to changes in load at these higher values.
 22

2.5 Factors Influencing Track Modulus

Railroad track structure is demonstrated in Figure 2-10. Railroad track has

several components that all contribute to track modulus, including the rail, subgrade,

ballast, subballast, ties, and fasteners. The rail directly supports the train wheels and is

supported on a tie pad and held in place with fasteners to ties. The crossties rest on a

layer of rock ballast and subballast used to provide drainage. The soil below the

subballast is the subgrade.

Figure 2-10: Track Structure (Chang et al. 1980)

The subgrade resilient modulus and subgrade thickness have the strongest

influence on track modulus. These parameters depend upon the physical state of the soil,

the stress state of the soil, and the soil type (Li and Selig, 1994; Selig and Li, 1994).

Track modulus increases with increasing subgrade resilient modulus, and decreases with

increasing subgrade layer thickness (Selig and Li, 1994). Ballast layer thickness and
 23

fastener stiffness are the next most important factors (Selig and Li, 1994; Li and Selig,

1998). Increasing the thickness of the ballast layer and/or increasing fastener stiffness

will increase track modulus (Stewart, 1985; Selig and Li, 1994). This effect is caused

by the load being spread over a larger area. The system presented in this dissertation

measures the net effective track modulus that includes all of these factors.
 24

3 SYSTEM MODELING AND SIMULATION

In this chapter, the measurement principle and methodology will be described.

Based on Winkler model, the relation between system’s measurement and track modulus

will be analyzed so that modulus can be estimated from the system’s measurement.

Simulations will show that track geometry variations will affect system’s measurement.

Therefore, a method is developed to eliminate this effect.

3.1 Measurement Principle and Methodology

The geometry of the measurement system is shown in Figure 3-1. An instrument

beam rigidly mounted on the side frame of the hopper car extends a few feet away from

the wheels. A sensor head which includes a laser/camera system is attached to the end

of the beam. The sensor system has two line lasers and a camera as shown in Figure

3-2. The line lasers intersect the rail surface at an acute angle to create curves across the

surface of the rail. Using line lasers allows the system to compensate for lateral

movement of the rail relative to the camera and for changes in rail profile. The camera

captures images showing two curved laser lines on the rail surface and the distance

between the lines d is obtained by an image processing program. This distance d is then

converted to the distance between the beam and the rail surface under the camera h. A

track model is then used to calculate the track modulus.

 25

Figure 3-1: Diagram of Measurement Principle

Figure 3-2: Camera/Laser System

 26

Figure 3-1 illustrates that the fixed distance between the wheel/rail contact point

and the sensor, H, relates the relative rail displacement, Yrel, to the measured height of

the sensor above the rail surface, h. Here ycamera is the deflection of the rail at the

location underneath the camera/lasers and ywheel is the deflection of the rail at the

wheel/rail contact point. The deflections are negative in value because the positive axis

is defined upwards.

The sensor system measures the distance between the camera image plane and the

rail surface, h. Then, the displacement of the rail surface with respect to the wheel/rail

contact plane, Yrel (Figure 3-1) can be found. The displacement, Yrel, can then be

related to the absolute rail deflection of the wheel/rail contact point (with respect to the

unloaded rail), ywheel (Figure 3-1) by using the Winkler model or the Cubic model.

The mathematical model relates the measured distance between the laser lines to

the track modulus. The rail deflection measured by the sensor is dependent on the four

wheel loads. The sensor will measure the relative rail displacement between the rail and

wheel/rail contact point. This measurement can be made if it is assumed that the

instrument beam, truck, and wheels are rigid. With this assumption, the distance

between the sensor system and wheel/rail contact point can be assumed constant (H is

constant). This is a reasonable assumption as the instrument beam, side frame, and

wheels are all massive, nearly rigid elements and these elements do not include the

suspension of the railcar. Rotation of the side frame could cause this distance (H) to

change, but this rotation has been experimentally shown to be insignificant (Norman,

2004).
 27

The sensor reading, which is the measured distance between the lasers, is

geometrically related to the height of the sensor above the rail. The sensor in effect

measures its height above the rail by measuring the distance between the lasers. As the

sensor moves closer or farther from the rail surface, the distance between the lasers

changes. A schematic of the sensor is shown in Figure 3-3.

Figure 3-3: Sensor (Lasers and Camera) Geometry

From the above figure, the following equations can be written:

(L1 + l1 ) tan θ 1 = h Equation 3-1

(L2 + l2 ) tan θ2 = h Equation 3-2

d = l1 + l 2 Equation 3-3
 28

where L1 and L2 are the horizontal displacement of the lasers from the camera, θ1

and θ2 are the angles between the lasers and the horizontal, l1 and l2 are the horizontal

distance between the center of the camera and laser/rail intersection, h is the vertical

distance between the camera/lasers and the surface of the rail, and d is the distance

between the lasers on the rail surface. Solving these equations results in:

h h
d= + − (L1 + L 2 )
tan θ 1 tan θ 2 Equation 3-4

Combining Equation 3-1 to Equation 3-4, a sensor reading can be calculated for a

value of track relative deflection.

Combining this information with the track model (e.g. Winkler model) the sensor

reading d can be related to the track modulus u. On softer track the rail will rise relative

to the wheel/rail contact point and the laser lines as observed by the camera will move

closer together. Conversely, the distance between the lasers will be large for stiffer

track.

3.2 The Relation between Yrel and Modulus (Winkler Model)

Figure 3-4 shows the rail deflection from multiple loaded axles.
 29

Figure 3-4: Superposition of the Deflections from Two Loads

In Figure 3-4, assuming the loads of wheel one (the left one) and wheel two (the

right one) are the same (P), w1 is the deflection of the rail attributed to wheel one and w2

is the deflection of the rail attributed to wheel two. The total rail deflection is the

superposition of w1 and w2.

From the Winkler model,

Pβ1 − β1⋅ x−c

w1 (x ) = − e [cos(β1 x − c ) + sin(β1 x − c )] Equation 3-5
2u (c)

where: 1

⎛ u (c ) ⎞ 4
β1 = ⎜ ⎟
⎝ 4 EI ⎠

E is the modulus of elasticity of the rail.

I is the moment of inertia of the rail.

 30

x is the longitudinal distance along the rail.

c is the position of wheel one in the x coordinate (see Figure 3-4).

Pβ 2
and w2 (x ) = −
− β ⋅ x − c −b
e 2 [cos(β 2 x − c − b ) + sin(β 2 x − c − b )] Equation 3-6
2u (c + b)

⎛ u (c + b ) ⎞ 4
where: β2 = ⎜ ⎟
⎝ 4 EI ⎠

b is the distance between the two wheel axles (72”).

(c+b) indicates the position of wheel two in the x coordinate frame.

The total deflection of the rail is the superposition of the two expressions:

wtotal ( x ) = w1 ( x ) + w2 ( x )

Then the total deflection at the wheel-rail contact point of wheel one is:

Pβ1 Pβ 2
wtotal ( x ) x=c = w1 (c) + w2 (c) = − − e −β 2 ⋅b [cos(β 2b ) + sin (β 2b )]
2u (c) 2u (c + b) Equation 3-7

and the deflection of the rail under the sensor head which is four feet away from

wheel one is:

wcamera ( x ) x=c−a = w1 (c − a) + w2 (c − a)
Pβ1 − β 2 ⋅a
=− e [cos(β 2 a ) + sin (β 2 a )] − Pβ 2 e −β2 ⋅( a+b) [cos(β 2 (a + b)) + sin(β 2 (a + b) )]
2u (c) 2u (c + b)

Equation 3-8

Then,
 31

Yrel = wtotal (c) − wcamera (c − a)

Assuming the track is absolutely uniform (i.e., u is a constant), then

⎛ u ⎞4 Equation 3-9
β1 = β 2 = β = ⎜ ⎟
⎝ 4 EI ⎠

Therefore,

Yrel = wtotal (c) − wcamera (c − a )

Pβ
=− {1 + e − β ⋅b [cos(β b ) + sin (β b )] Equation 3-10
2u
− e [cos(β a ) + sin (β a )] − e − β ⋅( a +b ) [cos(β ( a + b) ) + sin (β ( a + b) )]}
− β ⋅a

The result of Equation 3-10 is shown in Figure 3-5 where modulus is plotted as a

function of Yrel using a look-up table. In this model, 132 RE rail was chosen (I=87.9

in4); E is set to be 30,000,000 psi; the load on each wheel is 32500 lb; and the distance

between the two axles is six feet. The typical values of modulus for various main-line

track conditions (Kerr, 2003) are listed in Table 3-1 along with the corresponding Yrel

value.
 32

Figure 3-5: Re
elation betw
ween Yrel and
a Moduluss (Winkler model)
m

Table 3-1: Modulus and

a Yrel forr typical track conditions

Trrack Condition Descriptiion Moduulus (psi) Yrel (inch)

Wo
ood-tie trackk, after tampping 1000 0.2

Wood-tie track, coompacted byy traffic 3000 0.095

Concrette-tie track, compacted

c b traffic
by 6
6000 0.058

Wood-tie track, frozen ballast andd subgrade 9

9000 0.044

The relation
r betw
ween the raail deflections at the wheel-rail
w coontact pointt and

reelative deflection (Yrel) is shown inn Figure 3-6. This non-liinear relation is based on
o the

W
Winkler mod
del and supeerposition. For
F relativelyy small deflections (0~00.2 inch), Yrel
Y is

abbout 60% off the total deflection (defflection at thhe wheel-Raiil contact pooint).
 33

Figure 3-6: Relation between the Total Deflection and Yrel (Winkler model)

3.3 Effects of Track Geometry

The measurement of relative deflection (Yrel) uses the wheel/rail contact line as a

reference as shown in Figure 3-1. The measurement assumes the unloaded rail is

perfectly straight. However, if the rail has a significant pre-existing geometry variation

over a length comparable to the four feet between the measurement point and wheel/rail

contact point, the system’s measurement will be affected. Large vertical “dips” that

occur over a short length of track affect the measurement result.

The relationship between modulus and geometry is complex. In real track, areas

of geometry variations often correlate with areas of modulus variations and vice versa.
 34

A case study was chosen to investigate this relationship. Figure 3-7 shows a section of

track where there is a significant geometry variation and a significant modulus variation.

Measurements at the site indicated that the unloaded rail drops by 0.5” over a length of

about 200”. A geometry variation of this shape is significant and easily visible. The

light colored ballast seen at this site also suggests tie “pumping” and low track stiffness.

Figure 3-7: An Example Site with Both Significant Unloaded Geometry and Low
Track Stiffness

Relative rail deflection (Yrel) from the measurement system at this site is 1.1”.

Simulations, based on the Winkler model, have been conducted to quantify the effects of

track geometry on the measurement of relative deflection.

 35

Figure 3-8 shows an example simulation result. In this simulation, a section of

track has both geometry and modulus variations. The unloaded track geometry is

described in the top subplot in Figure 3-8. It has a maximum “dip” of 0.5” in depth and

it occurs over 200” (between 100” and 300”) of track.

In the simulation it was assumed that the modulus over this section of track varies

as a cubic curve with a minimum at the center of the geometry variation, (the middle

subplot in Figure 3-8). The bottom subplot in Figure 3-8 shows the Yrel measurement

for this site. Here, the “total” measurement replicates the value of 1.1” as it did in the

real measurement when the measurement system passed over the location shown in

Figure 3-7. To create this value it was found that the modulus for this location had to

drop from 3000 psi (assumed as a reasonable value for “normal” track) to 800 psi in

addition to the unloaded geometry profile. This measurement is then broken into two

“elements” – a modulus element and a geometry element. The geometry element is the

measurement that would be made if the same unloaded geometry (top subplot) existed on

a perfectly rigid track. The modulus element in the remaining portion is the total

measurement minus the geometry element.

 36

Figure 3-8: Simulation on the Effects of Track Geometry

It can be seen that in this case the contribution of geometry (the geometry element)

is about equal to the contribution of modulus (the modulus element). However, both are

required to make the measurement large.

Now, the simulation can be used to study the relative contribution of geometry

and modulus as the length of the geometry variation (L) and the depth of the geometry (d)

vary. The simulation result is shown in Figure 3-9.

 37

Figure 3-9: Effects of Unloaded Geometry of Various Length (L) and Depth (d)

It can be seen in Figure 3-9 that there is a complex relationship between modulus

and geometry and that the effects vary depending on the length (L) and depth (d) of the

geometry variation. The three-dimensional plot on the left shows the relative size of the

geometry element and the modulus element. It can be seen that there is a curve where

the elements are equal in magnitude.

The two graphs on the right show two cross sections of these surfaces. The top

right graph shows the effects of variations in the length of the geometry defect (L) at a

constant depth (d=0.5”). The bottom right graph shows the effects of variations in the

depth of the geometry defect (d) at a constant length (L=200”).

Again, the conclusions that can be drawn from these simulations are that (1) only

large vertical geometry defects occurring over a short distance significantly contribute to

the Yrel measurement, and (2) both geometry and modulus problems are generally

present to measure very large Yrel values.

 38

3.4 Eliminating the Effects of Track Geometry Variation

As seen in the simulation and analysis in the previous section, track geometry can

greatly affect the output of the system in terms of measuring rail deflection. In order to

eliminate the effects of track geometry variation and get the real rail deflection results,

rail profile data from track geometry measurement vehicles was introduced into the

system.

A track geometry vehicle is a railed vehicle used for non-destructive diagnosis of

railroad tracks. It measures various parameters including position, curvature, and

alignment of the track, smoothness and the cross-level of the two rails, etc. The space

curve channel of the geometry car uses multiple high-precision accelerometers onboard

to produce the rail profile.

3.4.1 10-ft ECO (End-Chord Offset) Calculation from Rail Profile

Figure 3-10: 10-ft ECO Calculation

As shown in Figure 3-10, P(x) is the rail profile from the space curve channel of

the track geometry data. The longitudinal position of the track is defined as x (units of
 39

foot). ECO(x) is the 10-ft end-chord offset when the leading wheel’s longitudinal

position is x. (ECO is positive if the string is above the rail.) Here, the 6 ft and 4 ft were

chosen because they are the distance between the two wheel axles and the distance from

the sensor head to the inboard wheel axle respectively.

From the geometry relation in Figure 3-10:

P( x − 6) − P( x) 6
= Equation 3-11
P( x) − [ P( x + 4) + ECO( x)] 4

Therefore,

2
ECO ( x ) = ⋅ [ P ( x ) − P ( x − 6)] + P ( x ) − P ( x + 4) Equation 3-12
3

3.4.2 Subtracting -ECO from Yrel

In Figure 3-11, P(x) is the vertical position of the inboard wheel-rail contact point

when the inboard wheel’s longitudinal position is x. P(x-6) is the vertical position of the

inboard wheel-rail contact point when the inboard wheel’s longitudinal position is x-6.

If it is assumed the two wheels always have the same space curve, then P(x-6) is the

vertical position of the trailing wheel-rail contact point when the inboard wheel’s

longitudinal position is x. P4(x) is the rail’s vertical position four feet ahead of the

inboard wheel when the inboard wheel’s longitudinal position is x.

 40

Moving

De

Figure 3-11: Deflection Calculation

Let A(x)=P4(x)-Yrel(x), then from geometry relation in Figure 3-11, we have:

P ( x − 6) − P ( x) 6
=
P ( x) − A( x) 4

So, 1
A( x ) = ⋅ [5 P ( x ) − 2 P ( x − 6)] Equation 3-13
3

Therefore, 1
P 4( x ) = A( x ) + Yrel ( x ) = ⋅ [5 P ( x ) − 2 P ( x − 6)] + Yrel ( x ) Equation 3-14
3

The vertical position of the rail at location x+4 may be determined when the

inboard axle is at location x. The vertical rail position at the same location may be

determined again when the inboard axle is actually at location x+4. Then, the difference

between these two measurements (P4(x)-P(x+4)) may be calculated by:

 41

“Deflection” = P4(x)-P(x+4)

1
= ⋅ [5 P ( x ) − 2 P ( x − 6)] + Yrel ( x ) − P ( x + 4)
3

2
= ⋅ [ P ( x ) − P ( x − 6)] + P ( x ) − P ( x + 4) + Yrel ( x )
3 Equation 3-15

= ECO ( x ) + Yrel ( x )

It should be noted that the so-called “Deflection” calculated above is not the

maximum deflection of the rail attributed to the loads. Rather, it is a relative deflection

of the rail from partially loaded (when the axle is four feet away) to fully loaded (when

the axle is right at the point).

Therefore, Yrel can be accounted for by two parts. One part is ECO, attributed to

the track geometry variations, and the other part is the deflection related to track modulus

variations.

3.5 Stress and Strain on the Rail

One potential output from the measurement system is that rail stress and strain

can be estimated from system’s measurement. Here, some primary studies are

demonstrated trying to relate Yrel data to rail stress and strain.

The bending moment on the rail is calculated as:

M ( x) = − EIw" ( x) Equation 3-16

The axial stress in the rail is given by

 42

M ( x) z
σ x ( x, z ) = Equation 3-17
I

wheree z is the verttical distance from the neutral

n axis.

Thereefore, the straain of the raiil is

σ x ( x, z ) M ( x) z
ε x ( x, z ) = = Equation 3-18
E E
EI

For tw
wo axles, suuperposition was applied to obtain the rail respponse from each

looad. The raiil bending moment

m cann be related to Yrel byy using the same param
meters

=30,000,0000psi; I=87.9iinch4 for RE

shhown in the model desccribed in secction 3.2, (E= E 132

raail; 32500 lb
b load on eaach axle) annd by combiining Equatiion 3-10, Eqquation 3-166 and

E
Equation 3-18. This analyysis is basedd on the Winnkler model with the asssumption thaat the

trrack has no geometry

g varriation.

Fig
gure 3-12: Relation be
etween Rail Bending Moment
M and
d Yrel
 43

The relations betw

ween Yrel and
a rail straain and betw
ween Yrel annd rail stresss are

plotted in Fig
gure 3-13 andd Figure 3-14 respectiveely.

Allowable
A Beending Stresss (AREA Manual
M (19966))

Figure 3-1
13: Relation
n between Rail
R Bendin
ng Stress (R
RE132, botttom) and Yrel
Y
 44

Figurre 3-14: Rellation betwe

een Rail Sttrain (RE13
32, bottom) and Yrel

Accorrding to thee AREA Manual

M (19966, Chapter 16, 10.2.2,55), the allow
wable

bending stress for a rail iss determinedd as follows:

σ σN
67y 8 67 8
70,000 − 20,000
σ= = 25,000lb / in 2
1{
.2 × 1{
.25 × 1{.15
1 × 1{
.15
La
ateral _ Rail _ Bendiing Track _ Condiition Rail _ Wear & Corrosion Unbaalanced _ elevation

wheree σy is the yieeld stress.

The denominator
d consists of four
f reductioon coefficiennts. Note thaat the inclusion of

thhe lateral raiil bending coefficient

c avvoids the neeed for a moore involvedd bending-toorsion

raail analysis with an unncertain forcce input. Thhe use of the

t rail weaar and corroosion

cooefficient sim
mplifies the analysis byy allowing thhe use of thee moment off inertia of a new

raail. The subttraction of thhe axial therrmal stress from

fr σy leadss to another simplification of

thhe analysis.
 45

Therefore, when Yrel is larger than 0.6”, the stress at the bottom of the rail will

exceed the allowable bending stress. However, it should be noted that Figure 3-13 is

based on the Winkler model and simple superposition which are not very suitable for

analyzing large deflection situations. In addition, Yrel measurements are usually

combined with geometry effects, and the complexity of the track (joints, switches,

bridges, etc.) will also contribute to large values of Yrel. All these complex factors

were not considered in the analysis in Figure 3-13, so it will be difficult to evaluate

stresses on the rail just based on Yrel measurements, especially when large deflections

occur.

3.6 Different Loads

From Equation 2-5, it can be observed that rail deflections are linearly

proportional to the applied loads given a constant track modulus. Based on the Winkler

model and superposition, the relations between Yrel and modulus under various loads are

illustrated in Figure 3-5. The values of load shown here are the loads on each axle.

Two axles as in Figure 3-4 were taken into consideration in this model. This figure

shows that given a certain modulus, the Yrel measurements increase linearly as the loads

increase. Therefore, the weight of the measurement vehicle is a major factor to be

considered when designing the measurement system. Clearly, heavier weights will

generate larger deflections.

 46

Figure 3-15: Effect of Different Loads on Yrel-Modulus Relation

 47

4 MEASUREMENT SYSTEM

In this chapter, all of the components of the measurement system will be

described in detail. These components include sensors, GPS and encoder, power supply

and management system, remote supervision, data downloading system, and a database

for testing results. The methods and procedures of image processing and calibration

will be illustrated. The potential errors in the measurements will also be analyzed.

4.1 Instrumentation

The measurement system is installed on a refurbished hopper car. The hopper

car is filled with sand such that the total weight of the car and sand is about 260,000 lb.

As shown in Figure 4-1, the system includes two rigid beams attached to the side frames

(structural members that connect the axles of a truck) of the hopper car, two sensor heads

holding cameras and lasers at the ends of the rigid beams, a solar panel array, a GPS

(Global Positioning System) antenna on the top of the car and an enclosed box containing

two computers, a data acquisition (DAQ) boards and a GPS receiver for on-board image

processing and data computation.

 48

Figure 4-1: System Instrumentation

All of these components will be described in detail in the following sections.

4.1.1 Sensor System

The sensor head is attached to highly rigid steel beams that are fastened to the

side frames as shown in Figure 4-2. The beams are bolted to the side frames without

modification of the side frames.

 49

Side Frame

Rigid Beam

Figure 4-2: Hopper Car Truck and Rigid Beams Assembly

These beams suspend the sensor heads out in front of the wheel/rail contact point

and over the top of the rail. Investigations into rigidity have shown that the beams are

sufficiently rigid to be considered stationary relative to the side frames (Norman, 2004).

Therefore, the measurement will not be affected by the bending or movement of the

beams. An actual view of the assembly is shown in Figure 4-3.

 50

Rigid Beam

Sensor Head

Figure 4-3: The Rigid Beam on the Side Frame

A sensor head holds a digital vision system and two line lasers as shown in Figure

3-2. The camera and line lasers are rigidly attached to an enclosed sensor assembly

which is mounted at the end of the rigid beam on the side frame. These lasers are

projected at an acute angle (~40˚). They cross and create curves (because of the curved

profile of the rail head) across the surface of the rail.

The assembly of the sensor head is shown in Figure 4-4. The two laser holders are

adjustable to ensure that the lasers project at an appropriate angle. The height of the

camera is also adjustable. This ensures that well-proportioned images will be captured.
 51

Caamera Positiion
L
Laser

Lasser

Figure 4-4: Sensor He

ead Assem
mbly

Ambient light, sppecifically suunlight, ruinns the image of the laserr lines acrosss the

raail and makes it imposssible for thee image proccessing proggram to disttinguish the laser

liines. Thereefore, a shrooud assemblly is made to

t shade thee sunlight. The shroudd is a

liight aluminu
um frame bolted
b to thee bottom off the rigid beam. Rubbber landscaaping

m
material is attached
a alonng the bottoom edge to add furtherr shading. This material is

fllexible, mak
king it resistaant to rocks and debris, but is also continuously
c y solid, makiing it

reesistant to wind
w effects.. As shownn in Figure 4-5, even inn very sunnny conditionss, the

shhroud provid
des adequatee shading forr the lasers and
a camera image.
 52

Shrouds

Figure 4-5: Shrouds Attached to the Rig

gid Beams

A typical sensor image

i is shoown in Figurre 4-6. The video imagee shows the laser

beams interseecting the toop of the raill. Images arre captured and
a processeed in real tim
me to

ouutput the disstance betweeen the laser lines (d in Figure

F 3-2).

Figure 4-6:
4 Typical Test Image
e
 53

This measurement is geometrically related to the height of the sensor above the

rail (h in Figure 3-1). As the sensor moves closer or farther from the rail surface the

distance between the laser lines changes. Using a calibration technique which will be

described in detail in later sections, Yrel can be calculated from the distances between the

laser lines in the images. As the system’s output, Yrel is the relative displacement

between the rail surface under the camera and the wheel/rail contact line. Yrel can then

be mathematically related to track stiffness and modulus (Lu et al, 2007; McVey, 2005;

Norman, 2004).

The real-time image processing and data management is performed by computers

on board. These computers are installed in an enclosed box as shown in Figure 4-7. A

GPS receiver and two DAQ boards are also installed in this box. All of the cables and

wiring for the equipments are run through flexible conduits. This sealed box protects

the computers and other electrical equipments from harsh environments.

 54

Figure 4-7: Enclosed Box for Computers

4.1.2 GPS and Encoder

The measurement system includes a GPS receiver. The latitude and longitude

are reported in real time and recorded in the output data. The GPS data are used to get

milepost information by comparing it with Precision Measurement Vehicle (PMV) data

and provides accurate coordinates for each location of interest.

The GPS receiver used in the system is a NovAtel OEM4 model which is capable

of absolute single-point positioning accuracies of 1.8 meters circular error probable (CEP)

(GDOP (geometric dilution of precision) < 2; no multipath). However, this level of

accuracy is really only an estimation, and may vary widely depending on numerous GPS

system biases, environmental conditions, as well as the GPS receiver design and

engineering quality.
 55

Numerous factors will influence the single-point position accuracies of any GPS

receiving system. As the following list will show, a receiver’s performance can vary

widely when under the influences of these combined system and environmental biases:

1. Ionospheric Delays. The Earth’s ionospheric layers cause varying degrees of

GNSS (Global Navigation Satellite System) signal propagation delay.

2. Tropospheric Delays. The Earth’s tropospheric layer causes GNSS signal

propagation delays.

3. Ephemeris Errors. Some degree of error always exists between the broadcast

ephemeris’ predicted satellite position and the actual orbit position of the

satellites.

4. Satellite Clock Errors. Some degree of error also exists between the actual

satellite clock time and the clock time predicted by the broadcast data.

5. Multipath. Multipath signal reception can potentially cause large pseudo

range and carrier phase measurement biases.

In general, all these factors combined may cause an error of up to 60 feet. GPS

errors may cause problems when data from multiple tests are compared in trending

analysis, because data from different tests may be out of alignment. As a result, data

pre-processing will be needed to align the data before comparison.

In order to obtain more accurate and reliable GPS readings, a more sophisticated

differential GPS system was proposed to upgrade the current GPS system. Some

differential GPS systems can achieve an accuracy to within centimeters.

 56

Due to the GPS error, there may be some stretches in the data in terms of

mileposts. An encoder has been introduced into the system to eliminate this stretching

problem caused by the GPS error.

4.1.3 Power Supply and Management System

Two rugged computers are used to process images in real time and save the data.

In order to reduce power consumption when the testing vehicle is not moving, the whole

measurement system enters a “sleeping” mode in which the lasers and cameras are turned

off and the PCs remain in standby mode. The total power consumption is about 50

watts when the system is in full-on testing mode and 10 watts in sleeping mode.

An on-board power supply and management system was developed to make

automated testing possible. As shown in Figure 4-8, four solar panels installed on the

top of the testing vehicle provide the power source for the measurement system. The

solar panels are rated at 400 watts maximum. A battery pool consisting of eight

deep-cycle marine batteries is used as energy storage and as a buffer to provide stable and

consistent power to the system during both day and night. The eight batteries have a

total capacity of 400 Amp-hours which can supply 4 days of continuous testing or 16

days of “sleep” mode without inputs from the solar panels. The batteries are enclosed in

the black box as shown in Figure 4-8 along with a solar panel voltage regulator which

manages the battery recharge process and prevents the batteries from being overcharged.

Inside of the battery box, two watt meters were also installed to provide information

about the measurement system’s energy usage and energy input from the solar panels.
 57

Figure 4-8: Power Supply System

The power supply system has proved to be sufficient and reliable during

automated tests in March and April of 2008. In these tests, the batteries were returned

fully charged after the six-day, 1300-mile-long journey.

4.1.4 Remote Supervision and Data Downloading

The voltage across the batteries, the current input from the solar panels and the

ambient temperature are monitored by the computer in the white box shown in Figure

4-8. This information can be recorded and sent back to a server on the internet through

wireless communication which enables the power supply system to be remotely

monitored.

The data which were logged and uploaded in real time from the computer on

board to a remote server through the internet during the test in April of 2008 are
 58

displayed in Figure
F 4-9. As shownn by the currrent and volttage traces in
i this figuree, the

batteries weree charged duuring the dayytime. Thee output of thhe solar paneels was meassured

ass high as 15
50 watts. T highest output
The o from the solar paanels usuallyy occurred in
i the

m
morning to co
ompensate for
f the powerr consumption during thhe nights. While
W in sleeeping

m
mode, the battteries can be
b fully chargged before noon
n and thee voltage of the batteriess will

reemain constaant for the reest of the dayy.

Figure 4-9:: Power Supply System

m Monitorin
ng Information in the April
A 2008 Test
T

4.1.5 Testin
ng Results Database
D

The teesting vehiccle has been tested on thhousands off miles of trrack. As a result,

huuge amounts of data haave been prooduced from

m these tests. Thereforee, a databasee was

seet up to orgaanize and manipulate

m thhese data. A website was
w created to
t search thrrough

thhe database and to deveelop plots. A screenshot from thiss website is shown in Figure
F

4-10. The website

w can also produce lists of exxception locaations from the databasee and

plot them on maps as shoown in Figure 4-11.

 59

Figure 4-10: Database Website Screenshot

Figure 4-11: Exception Locations List from the Website

 60

4.2 Real-time Image Processing

4.2.1 Basic Algorithm

In the typical image captured during the test shown in Figure 4-6, the laser lines

are easily identifiable on the top of the rail. The image processing program scans

through all the pixels on each horizontal line of the image and finds the peaks of the pixel

intensities which represent the locations of the laser lines. Subsequently, the laser lines

can be reproduced as shown in Figure 4-12. After scanning through the image and

obtaining the valid points on each laser line, some points may be considered as valid but

out of the curve due to reflections on the rail surface or other noise captured in the images

as shown in Figure 4-13. These outliers are filtered out and cubic curves are applied to

fit the remaining valid points.

Figure 4-12: Reproduced Laser Curves

 61

Figure 4-13: An Imperfect Image Example

4.2.2 Verification Using Neural Networks

The current image processing method for the measurement system has some

limitations when the system works under certain extreme conditions and the images

captured are not ideal. The image in Figure 4-6 is a typical one captured by the system.

In this image, the laser curves only appeared on the top of the rail and they are clear and

easy to identify. However, images like the one in Figure 4-14 are not ideal to process.

In this image, the lasers lines projected on the bottom of the rail and tie were captured by

the camera and bright reflections appear on top of the rail. These effects may cause the

program to obtain incorrect distances between the laser curves.

 62

Figure 4-14: Variouss Images Captured

C byy the Camerra

In ord
der to avoidd false alertts caused by incorrect image proccessing, a neural
n

network is in
ntroduced innto the systeem to verifyy the imagee processingg. Based onn the

mage processsing algorithhm, a few elements

im e werre determineed to be impportant indiccators

of how well images were processed.

First, the number of valid poinnts on each laser

l line is a key param
meter in the im
mage

prrocessing. The prograam scans thrrough all thhe pixels onn each horizzontal line inn the

im
mage and fiinds the peaaks of the intensity whhich represennts the locaation of the laser

cuurves on thee top of the rail. Howeever, if theree is no obvioous peak onn some horizzontal

liine, the prog

gram will coonclude thatt no laser cuurve was foound in this location. After

sccanning thro
ough all of the
t horizonttal lines, onnly the validd points com
mpose the cuurves.

T
Therefore, the number off valid pointts on each laaser curve inndicates how
w well the cuurves

w interpreted.
were
 63

Another factor consists of determining the number of outliers. There may be some

points valid but out of the curve. The program will filter out these outliers. Better

image processing is achieved with fewer outliers.

The standard deviation of the fitting curve is a third consideration in the image

processing. After filtering out all the outliers, the program will fit remaining valid

points to a cubic curve and calculate the standard deviation of the fitting. Smaller

standard deviation indicates better image processing.

The position of the center lines between the two lasers presents yet another issue

for image processing, because in general the two laser curves should be symmetric along

the vertical center line of the image.

Finally, the change in rail position from the last two images also plays an

important role in the image processing. When the system passes by turnouts, there will

be changes in rail and occasionally multiple parallel rails are captured in the image.

When the lasers are projected onto multiple rails, it is important to pick the correct rail.

Therefore, the position of the rail in the images should not drift dramatically.

The above five parameters were used as inputs to construct a neural network as

shown in Figure 4-15. Two units are used in the hidden layer. About 50 images were

processed and the five parameters resulting from those images were saved as training

samples. Each of these images were then reviewed and assigned to an output based on

whether the program processed the image correctly. The network was then trained

using these samples and the weighted values were determined.

 64

Figure 4-15: Neural Network Structure

The trained neural network was then implemented into the image processing

program. Each image is verified based on the output from the network. A threshold of

0.6 was used to decide if the image was processed correctly. Figure 4-16 shows the

results output from the program without using neural networks. A number of outliers

exist which are caused by incorrect image processing.

Figure 4-16: Processing Results without Using a Neural Network

 65

The results from the program with the neural network are shown in Figure 4-17.

After each image was processed, the neural network was used to verify if the processing

was successful. If the output from the neural network indicated the image was

processed incorrectly, the result from current processing is discarded and the reading

from the last image is saved. As shown in Figure 4-17, all of the outliers were

eliminated which shows that the neural network works well to identify the incorrect

image processing.

Figure 4-17: Processing Results Using Neural Network Verification

4.3 Calibration Approach and Procedure

As shown in Figure 3-1, Yrel is the relative displacement between the rail surface

under the sensor and the wheel/rail contact line. Yrel is the measurement system’s

output.

The system processes images in real-time and obtains the number of pixels

between the two laser lines in the images. This number of pixels is the system’s direct
 66

measurement. In order to convert this number of pixels into the value of Yrel in inches,

a calibration must be conducted.

The purpose of the calibration procedure is to obtain the relation between the

system’s direct measurement (n, the number of pixels between the laser lines in images)

and the expected output (Yrel).

Figure 4-18 shows how to convert the number of pixels into the actual distance in

inches. An image of a ruler on top of the rail was captured. The six inch ruler

corresponds to 208 pixels in the captured image. Therefore, one inch in the image

corresponds to 208/6=34.67 pixels (i.e. if the number of pixels between the two lines is n,

the actual distance d is n/34.67 inches).

Figure 4-18: Converting Number of Pixels into Distance in Inches

Now the problem consisted of finding the relation between d and Yrel.
 67

If at one moment, Yrel=Yrel* and d=d* are known, and the ratio of ΔYrel and Δd,

ΔYrel
(R= ) is also known, then the relationship between ΔYrel and Δd may be
Δd

determined as follows:

First, ΔYrel and Δd may be represented as:

ΔYrel=Yrel-Yrel*

Δd=d-d*

Then it follows that,

Yrel = Yrel * + ΔYrel = Yrel * + R ⋅ Δd = Yrel * + R ⋅ (d − d *) Equation 4-1

Therefore, the calibration problem consisted of finding Yrel*, d* and R which

involves the following steps.

4.3.1 Finding the Ratio of Δd and ΔYrel (R)

The ratio of Δd and ΔYrel (referred to as R) can be calculated based on the

geometry of the laser beams and the rail.

Figure 4-19 is the diagram of the sensor head and the rail.

The distance between the two laser heads is 28.112”;

The vertical distance from the laser heads to the top of the rail is 13”;

The distance between the two laser lines in the images (d) is 3.768” (131pixels);

Therefore, the ratio of ΔYrel and Δd is:

ΔYrel 13
R= = = 0.40778
Δd 28 .112 + 3.768
 68

Figure 4-19: Calibration (Geometry of Sensor Head and Rail)

1
Therefore, one pixel in the image represents ⋅ 0.40778 = 0.01176 inch of
34 .67

ΔYrel.

4.3.2 Verifying the Ratio of Δd and ΔYrel (R)

A special plate is made for this calibration procedure. The device consists of

seven steps as shown in Figure 4-20 and Figure 4-21. The lowest step is 1/4 inch high

while the other six steps are 1/8” high. When doing the calibration, this plate is placed

on top of the rail under the camera, and the two laser lines project onto the plate.

Therefore, an image similar to Figure 4-22 is captured by the camera.

 69

Figure 4-20:
4 Calibrration Plate on the Top
p of the Raiil

Figure 4-21: Calibration

n Plate on Top
T of the Rail
R (Side View)
V

Figure 4-22: Capturred Image of

o the Calib
bration Plate
e
 70

The image in Figure 4-22 is then processed. The distances between the laser

lines on each step of the plate were obtained and plotted in Figure 4-23 with respect to

the step’s height above the rail. As shown in Figure 4-23, a linear line fits the data

points very well (R2=0.9988). From the line fit, one pixel in the image represents

1
= 0.01178 inch of ΔYrel.
84.857

140

120

100

80
Pixes

60

40
y = -84.857x + 139.04
2
R = 0.9988
20

0
0 0.2 0.4 0.6 0.8 1 1.2
Distance above rail(inch)

Figure 4-23: Calibration Results

This result verifies the calibration results very well. Based on the geometric

relation, one pixel in the images was calculated to represent 0.01176 inch of Yrel. The

difference between the results from the two different methods is

(0.01178-0.01176)/0.01176 = 0.17% which is negligible.

This difference may be attributed to many factors including the limited resolution

of the captured images, the wide laser lines in the images, measurement error during the

calibrations, etc. Some of these factors will be discussed in detail in section 4.4.
 71

4.3.3 Finding Yrel* and d*

After calculating the ratio of Δd and ΔYrel, all that is needed is to find a reference

point where both the relative deflection (Yrel) and the distance between the laser lines (d)

may be determined.

Assuming that a section of track is perfectly uniform and the modulus over the

section is constant, the deflection of the rail under the axles will be identical over the

entire section. If a train moves over this section at a very slow speed, the rail deflection

at one fixed point as a function of time can be mapped from the time domain into the

space domain so that the static rail profile under the axles can be obtained. The

dynamic load factor can be ignored since the train speed is slow.

Figure 4-24 demonstrates the setup for capturing the rail deflection with a video

camera. A sticker is placed on the side of the rail head as a marker while a video

camera away from the track captures this sticker in its view. Another marker is placed

on the ballast four feet away from the sticker.

 72

Figure 4-24: Capturing the Rail Deflection with Video Camera

The measurement vehicle is pulled at a constant speed and passes by the video

camera. Meanwhile, the video camera captures and saves the video for post-processing.

An image from the captured video is demonstrated in Figure 4-25. As the wheels of the

measurement vehicle move closer to the position of the marker, the deflection of the rail

increases and the marker in the camera image goes down. When the wheels move away

from the marker, the marker goes up in the video.

 73

Figure 4-25: Captured Video Showing the Rail Deflection

An image processing program was developed to process the video and determine

the vertical position of the top edge of the marker as shown in Figure 4-25. Assuming

the deflection of the rail is zero when the locomotive and measurement vehicle are far

away from the marker, the vertical position of the marker in the video can then be

converted into the actual deflection of the rail which is plotted in Figure 4-26. When the

sensor head passes by the marker, the two ends of the shielding shroud will block the

marker from the camera view which causes the image processing program to fail during

that time. Therefore, Yrel*, the vertical height difference between the rail surface under

the camera and the wheel/rail contact plane, is shown in Figure 4-26. From these

calibration data, Yrel* is 0.045 inch.

 74

Figure 4-26: The Deflection Curve of the Rail from Calibration

During the calibration procedure, the measurement camera in the sensor head also

captures the video of the laser lines on the top of the rail. This video is post-processed

to capture the image of the marker placed on the ballast as shown in Figure 4-27. From

this image, the distance between the two laser lines (d*) is obtained.
 75

The Marker

Figure 4-27: Captured Image when Sensor Head Passes by the Marker

This calibration procedure assumes that the section of track (about 40 feet around

the sticker’s position) is uniform. However, track conditions may change from tie to tie.

Even though this procedure was done on a section of specially selected quality track,

there is no guarantee that modulus over the section is constant. Therefore, the accuracy

of the result is worthy of further examination.

4.3.4 Verifying Calibration Results

After obtaining the ratio of Δd and ΔYrel, Yrel* and d*, the distance between the

two laser lines in the images can be converted into the measurement results, Yrel, by

applying Equation 4-1. However, as stated above, it is difficult to accurately determine

Yrel* and d*. Therefore, another procedure was developed to verify Yrel*, d* and the

entire calibration results.

 76

A mechanical shop for railroad vehicles (shown in the satellite map in Figure 4-28)

is involved in this procedure. The track in this shop is assumed to be absolutely stiff such

that no rail deflection will be considered when trains move across this section of track.

This assumption is reasonable since the track in the shop is in good condition and the rail

is supported by a quality concrete foundation.

Figure 4-28: Yrel Data from the Mechanical Shop

The results of the Yrel data for this section of track are plotted in Figure 4-28.

As shown in the information section of this figure, the average of Yrel over this section

of track is 0.0018”. Since this value is very close to zero, it confirms the calibration

results very well. The variations of Yrel, which are relatively small (standard deviation

is less than 0.1”), are attributed to the pre-existing geometry variations of the track.
 77

4.4 Error Analysis

System’s measurement error will be discussed in this section. The potential

factors causing measurement errors include limited sampling rate, the wide laser lines in

images and the laser beam drifting.

4.4.1 Errors Caused by Limited Sampling Rate

The cameras in the current measurement system produce analog video signals in

NTSC standard which is 30 frames per second. As a result, individual data points are

spaced approximately every two feet when the measurement vehicle travels at 40 mph.

However, Yrel measurements may change considerably within two feet since certain

track modulus and geometry variations occur over extremely short distances. Therefore,

this limited sampling rate may result in measurement errors.

For example, 50 feet of data is presented in one-foot increments in Figure 4-29.

The highest peak within the 50 ft range is point B, which is 0.77”. If the test is

conducted at a speed of 40 mph, the system will record readings every two feet. As a

result, the peak at point B could be missed while only the data points at A and C would

be recorded. Hence, an error of 0.04” (the difference between point A and B) would

result. Since the errors based on limited sampling rate depend on how the actual data

varies over short distances they are difficult to quantify. Some locations such as joints

are much more sensitive to this sampling rate issue because the actual Yrel data can

change dramatically over a few ties. Based on examination of the data from past tests,

the error could be up to 0.2” at some locations.

 78

Figure 4-29: Limited Sampling Rate Causing Measurement Errors

To address the problem caused by the limited sampling rate, high-speed cameras

are proposed to upgrade the system’s sampling rate to 120 frames per second.

Therefore, the data can be collected every six inches at a testing speed of 40 mph. This

will greatly improve the system’s measurement repeatability and accuracy.

4.4.2 Measurement Resolution

The measurement system is based on image processing. The most direct

measurement output is the number of pixels between the two laser lines in the images.

Therefore, the measurement resolution is directly determined by the resolution of the

captured images.
 79

The current frame grabbers are set to capture images in half frame mode which is

320x240 pixels. Based on the calibration results, for images of 320x240 pixels, one pixel

in each image represents 0.0118 inch of Yrel. Hence, the measurement system’s

resolution is 0.0118 inch.

However, the current cameras and frame grabbers can easily be upgraded to

capturing images of 640x480 pixels. In that case, the system’s resolution may be

improved to 0.0059 inch.

4.4.3 Laser Line Width

The image-processing program identifies locations of the laser lines by finding

peaks of pixel intensities. However, as shown in Figure 4-30, a laser line can be as wide

as three or four pixels in a captured image. Therefore, it is difficult for the image

processing program to precisely distinguish which pixel should represent the location of

the laser lines, especially when two adjacent pixels have the same intensity. Although the

curve fitting algorithm used in the image processing program greatly reduces the error

caused by this factor, it is reasonable to conclude that an error of ±1 pixel (±0.0118” in

Yrel) may result.

 80

Figure 4-30: Laser Line Width

4.4.4 Laser Beam Drifting

The laser line generator’s boresight accuracy is rated at a maximum of 2.5mm/m.

Boresight accuracy, known as pointing accuracy, is a measure of the angular difference

between the beam propagating axis (where the laser beam is pointing) and the mechanical

axis (where the laser housing is pointing). Due to the nature of the measurement

principle and the calibration procedure, the boresight accuracy has no effect on the

measurement result because the measurement is not related to the mechanical axis of the

laser housing at all. As a result, no measurement error is caused by the boresight

inaccuracy.

However, pointing stability, a measure of how much the laser beam alignment

drifts over a period of time, will cause errors. The laser line generator’s pointing

stability is rated at <50 μrad. The geometry relation between the lasers and the rail is

shown in Figure 4-31, where H is the vertical height of the laser generator above the rail,

α is the angle between the laser beam and the vertical direction, Δα is the drifted angle

and Δd is the drifted distance of the laser line on the rail.

 81

From the geometry relation, we know that

Δd = H ⋅ tan(α ) − H ⋅ tan(α − Δα ) Equation 4-2

From the calibration, we know H=13.6, α=50°, and the maximum Δα is 50μrad.

Therefore, Δd is calculated to be 0.00165 inch. Based on the geometric relation from

the calibration, 0.00165 inch in Δd will result in a ΔYrel of 0.00165x0.853=0.0014 inch.

Considering the worst case in which both laser beams are drifting towards the

camera at the same time or away from the camera at the same time, the maximum error

will be doubled. Therefore, the maximum error of Yrel that the laser beams drifting can

cause is 0.0028 inch, which is negligible.

Figure 4-31: Laser Beam Drifting

 82

5 FIELD TESTING AND DATA ANALYSIS

5.1 Revenue Service Testing

Since 2006, the measurement system has conducted revenue service tests in many

locations including the Burlington Northern Santa Fe (BNSF) Railway’s St. Joseph

subdivision in Kansas, and the Union Pacific Railroad (UPRR)’s South Morrill

subdivision, Sidney subdivision, Ravenna subdivision and Powder River subdivision in

Nebraska. Some short distance tests such as between Lincoln and Ashland, NE and

between Lincoln and Tecumseh, NE were also conducted.

Recent tests focused on the 160 miles of heavy axle load freight line of the Union

Pacific Railroad on the South Morrill subdivision of the North Platte division. The tests

were performed on Main 1 (primarily unloaded with approximately 50 MGT/year) and

Main 2 (primarily loaded with approximately 250 MGT/year). Repeated tests were

conducted on this subdivision in October 2006, December 2006, February 2007, April

2007, January 2008, April 2008, and June 2008. Tests were done at speeds up to 60

mph in a work train consist. The automated testing ability of the system made these

tests easy to conduct and cost-efficient. These repeated tests were performed in

different seasons, under different weather conditions, an d at different speeds, providing

tremendous amounts of data for analysis. Figure 5-1 shows the measurement vehicle in

the consist of a coal train during a test.

 83

Figure 5-1: System in Revenue Service Testing

An example of data recorded over a distance of one mile is shown in Figure 5-2.

The plot shows the relative deflection of the rail, Yrel (as defined in Figure 3-1), as a

function of GPS coordinates given in degrees of longitude and latitude. The data are

overlaid on a satellite image (Google Map). In this figure it is possible to qualitatively

trace changes in relative deflection to specific track events such as grade road crossings,

culverts, and bridges.

 84

Figure 5-2: Yrel Data Overlaid on a Satellite Map

Figure 5-3 shows data sampled over one mile of track. The plot shows the

relative rail deflection between the measurement point and the wheel/rail contact point

(Yrel as defined in Figure 3-1). Also shown are the mean and standard deviation of the

data. The mean represents the average value over the past 0.1 miles of track and the

standard deviation corresponds to the mean over that same distance. This distance was

chosen arbitrarily as a characteristic length. The mean and standard deviation are

therefore dependent on the direction of travel (right to left in this plot).

This section is an interesting example because it corresponds to a high-speed

crossover between MP 0.6 and 0.9. A non-insulated joint bar (near 0.9) is located at the

right end of the crossover. The relative deflection at this point became very large (over

1”) indicating a very low track modulus. The joint bar failed and caused a derailment
 85

only two weeks after the track modulus measurements were taken. This location

represented the second highest relative deflection measurement recorded over the nearly

350 miles of track tested for this portion of the test. This spike in relative displacement

obviously caused a jump in both the mean and standard deviation of the data. The other

end of the crossover (near MP 0.65) also displayed a rise in the standard deviation,

indicating a rough section of track.

Figure 5-3: A Rough High-Speed Crossover

In contrast to Figure 5-3, an example of one mile of quality track is shown in

Figure 5-4. Yrel and therefore modulus were both very consistent over this section of

track. The consistent mean and relatively low standard deviation emphasize the track’s

quality.
 86

The results of these tests suggest that information in these measurements may be

useful in indicating unsafe sections of track in need of repair. The contrast between the

two figures suggests a quantitative method to more rationally schedule and prioritize

track maintenance.

Figure 5-4: Track with Consistent Modulus

5.2 Measurement Repeatability

The multiple tests over the same section of track (South Morrill Subdivision of

UPRR) allow for comparison between tests. This comparison highlights the high

repeatability of the measurement. At most locations the measurements were very

similar for each test.

 87

Measurements over the same section of track from tests in December of 2006,

February of 2007 and April of 2007 are shown in Figure 5-5. The measurements from

the three different tests show almost no difference over this section of track, which

indicates the system’s measurement is highly repeatable. The high repeatability of data

as shown in Figure 5-5 is observed over most sections of the 320 miles of track in this

subdivision.

Figure 5-5: Measurements from Multiple Tests

5.3 System Measurement under Various Conditions

The results from multiple tests in South Morrill subdivision also provided

abundant data for studies on system’s measurement under various conditions such as at

different testing speed, under different weather condition, and for different size of rail.
 88

5.3.1 Results From Different Train Speeds

The results of South Morrill subdivision suggest that the measurements are not

strongly related to the train speed. The average train speed for the 0.1 miles shown in

Figure 5-6 are 48, 22, and 35 mph, respectively.

Figure 5-6: Measurements from Different Testing Speeds

Repeated tests over the same section of the track at different speeds were

performed during a special validation test on the Yoder subdivision. Four different

speeds (20, 30, 40 and 50 mph) were tested. For most of the sections, the measurements

from the different speeds are highly repeatable, which indicates that the speed is not a

significant factor in the measurements at those locations. However, at some locations

differences do exist between the different speeds. The section shown in Figure 5-7 is an

example of these locations. The values of the peaks at MP224.4685 corresponding to

the four different speeds within this short section (21, 30, 40 and 48 mph) are 0.607”,

0.687”, 0.692” and 0.77” respectively. The Yrel measurements increased 0.163” when
 89

the train speed increased from 21 mph to 48 mph. This can be explained by the

increased dynamic load when the train moves at higher speeds.

Figure 5-7: Dynamic Loads Effecting Measurements

5.3.2 Results From Different Seasons

Although most of comparisons between tests showed extreme repeatability as

shown in Figure 5-5 and Figure 5-6, there were some variations in certain sections of

track. The most noticeable variation is the comparison between the December, 2006

test and the other tests. At multiple locations of the track, the relative deflection data

(Yrel) from the test in December, 2006 has an offset compared with the results from other

tests as shown in Figure 5-8. The test in December, 2006 was special because it was

conducted in extreme weather conditions with falling snow and an ambient temperature

around a high of 20° F. Figure 5-8 shows similarity in the shape of the measurements,
 90

but a shift in the Test 2 measurement. The December 2006 test has a lower mean value

indicating the track is stiffer (smaller Yrel). This could indicate a change in track

modulus associated with frozen subgrade that may contain moisture. These results

suggest the system could be useful to quantify seasonal variations in track modulus

caused by factors such as variations in subgrade moisture.

Figure 5-8: Variations of the Measurements

Interestingly, the offset of the measurements shown in Figure 5-8 only occurred at

some track sections. Figure 5-9 shows a section of track where only the measurements

on the left half of the figure has this offset while on the right half of the figure the results

from the three tests are very similar. This could indicate variations of subgrade moisture

can lead to greater (or lesser) seasonal variations in track modulus.

 91

Figure 5-9: Variations in Some Sections of Track

5.3.3 Results from Different Rail Sizes

One method used to inspect the change in Yrel data over long distances up to an

entire subdivision is to plot the average Yrel measurement over every 500 feet. The

data from three tests on the South Morrill subdivision are presented using this procedure

as shown in Figure 5-10. The sizes of the rail over the entire subdivision are also

displayed. A general correlation exists between the 500 feet Yrel averages and the rail

sizes. For example, the size of the rail between MP82 and MP104 is 141 lb per yard.

The data from all three tests display a noticeable decrease in Yrel over these 22 miles.

However, it is difficult to single out one factor’s effect since the track performance is

affected by many interconnected sources. Further theoretical studies and data analysis

are needed to examine how Yrel data varies as the rail size changes.
 92

Figure 5-10: Average Yrel from Three Tests on Different Size Rail

5.4 Measurement Validation

A special validation test was conducted in October, 2008 on the UPRR’s Yoder

Subdivision between Cheyenne, WY, and Yoder, WY. The purpose of this validation

test was to confirm the measurements collected by the system. Three methods were

proposed and performed.

5.4.1 String Measurements

As described in section 3.1, Yrel is the distance from the rail surface under the

camera to the wheel/rail contact plane. Therefore, the method depicted in Figure 5-11 to

measure Yrel is very straightforward. Here, a string is pulled to pass the bottoms of the

two wheels. Then the distance from the top surface of the rail under the camera to the

string is the Yrel reading at this location.

 93

Figure 5-11: String

S Meassurement Diagram

As sh
hown in Figuure 5-12, a specially
s maade magneticc ruler is atttached to the rail

duuring the fieeld measureement to maake the readding process easier. Thee string is tiightly

held from thee two ends as

a it barely toouches the bottom
b of the two wheells. The distance

frrom the top surface of thhe rail underr the cameraa to the stringg is read to compare
c witth the

Y reading from the meeasurement system.

Yrel s

Figure 5-12: Field String Measurement

The string
s measuurement wass performed at three loccations durinng the validdation

teest. The measurement

m results and the Yrel meeasurements from the teesting vehiclle are

liisted in Tab
ble 5-1. Thhe field strinng measurem
ments and the
t Yrel meeasurements each

differ by less than 0.1 incch.

 94

Table 5-1: String Measurement and Yrel Measurement

Yrel String Measurement Difference

Locations
(inch) (inch) (inch)

#1 1.000 1.004 0.004

#2 0.703 0.610 0.093

#3 0.703 0.669 0.004

The string measurements from these three locations closely matched the Yrel

measurements from the testing vehicle, which suggests that the system’s Yrel

measurements are correct. However, this type of string measurement is not very

accurate. A significant practical limitation is ensuring that the string perfectly passes

the bottom of each of the two wheels – a rather difficult task. This limitation can easily

contribute a relatively large error in the measurement results. To provide more accurate

results, further methods are proposed and described in the following sections.

5.4.2 Survey Measurements

This method uses a surveyor’s total station as shown in Figure 5-13 to

independently measure the vertical deflection of the rail. Such instruments are

commonly available with an indicated accuracy of less than one millimeter. In this

scenario, rulers used as targets are attached to the side of the railhead by strong magnets

as shown in Figure 5-13.

 95

Figure 5-13: Instruments Used in Survey Measurements

Figure 5-14 demonstrates the measurement scenario. Thirteen rulers are placed

every three feet on the side of the rail so that the total measurement range is 36 feet.

This distance is usually enough to cover the entire deflection basin. First, the unloaded

profile of the rail is measured by using the total station and a glass Porro prism

commonly used in surveying. Next, the total station is used to measure the height of

each ruler on the side of the rail while the rail is unloaded. After the measurements are

recorded, the locomotive parks the measurement vehicle on this section of the track such

that the inboard axle is directly on top of the center ruler as shown in Figure 5-14. The

total station is then used to take the height measurement of each ruler a second time. The

difference between the two height measurements for the same ruler is the rail deflection

at that ruler’s location. Adding the deflection measurement to the unloaded rail profile,

the loaded rail profile can then be determined. Analyzing the resulting measurements

would allow for the confirmation of the measured Yrel reading.

 96

Figure 5-14: Measurement of Vertical Rail Position by Surveying

Survey measurements were conducted at two locations during the validation test.

Measurement results from one of these locations are presented in Figure 5-15. The

curve demonstrates the loaded profile of the rail with one wheel at position 0 and the

other at a position of 70 inches since the distance between the two axles is 70 inches.

The profile points at these two locations are connected with a line which is then extended

by -48 inches in the horizontal direction because the horizontal distance between sensor

head and inboard axle is 48 inches. As a result, the distance from the end of the line to the

rail at position -48 inches is the Yrel measurement at this location. As shown in the

figure, Yrel measures 0.732 inch at this location.

 97

Figure 5-15: Survey Measurement Results

The testing vehicle’s measurements are compared with the survey measurements

from each site as shown in Table 5-2. The two methods produced almost identical

results at both sites. This suggests that the testing vehicle’s measurements are accurate

and reliable.

Table 5-2: Comparison between System Measurement and Survey Measurement

Yrel Measurement from Yrel Measurement

Difference (inch)
testing vehicle (inch) from surveying (inch)
Site A 0.738 0.732 0.006
Site B 0.150 0.150 0.000
 98

5.4.3 Absollute Deflectiion Measurrements by Cameras

C

Similaar to the calibration

c p
procedure d
described inn Section 4.3.3, a com
mmon

caamcorder is used to obbserve the veertical displacement of the rail as shown in Figure
F

5-16. A wh
hite magnetiic strip is affixed
a to thhe rail to usse as a targget. The caamera

reecords the viideo of the white

w strip ass the movingg train passees the locatioon. The viddeo is

thhen post pro

ocessed to determine the
t vertical location off the strip over
o time. This

prrocess is mu
uch simpler than using LVDT’s
L as described
d in Section 2.3 and will proovide

coomplete infformation onn the passiing train. A section of sampledd deflection data

prrocessed fro
om a video of a coal train is presennted in Figuure 5-17. In
I this sectioon of

data, a seriess of coal hooppers causees the rail too deflect abbout one hallf inch and three

heavier locom
motives at thhe end of the train to defllect the rail over
o 0.9 inchhes.

Figure 5-16:
5 Wayside Camera
a Measurem
ments Setup
p
 99

Coal hopppers

Three locom
motives

Figure 5-17:
5 Samp
ple Data of Absolute
A Deflection fro
om Waysid
de Camerass

Durin
ng the validdation test, this
t type off absolute deflection
d m
measurement
t was

coonducted at three locatioons where suurvey measuurements werre also perfoormed. Thee raw

data of deflecction over tiime from onne of the loccations is plootted in Figuure 5-18. Here,
H

thhe deflection
n caused by the locomootive, the tesst vehicle (U
UNLX002) and
a the geom
metry

caar are identiified respecttively. Thee maximum deflection of

o the rail caaused by thee test

vehicle load occurred att point A, which

w is the moment when
w the axxles pass byy the

he magnetic strip.
position of th The maximum absolute
a defl
flection was determined to be

1.1 inches forr this locatioon. Similarr plots were made

m for thee other two locations,
l annd the

abbsolute defleections weree evaluated.

 100

Figure 5-18: Deflectio

on Data from
m Camera Measurement

Table 5-3 containns all of the deflection

d daata from the surveying measurement
m ts and

thhe camera video

v at the three locatiions. The three
t deflection measurrements from
m the

caamera are very

v close too the corressponding onnes from thee surveying technique. The

differences arre less than 0.1” at eachh location, soo the two typpes of measuurements suupport

onne another.

Table 5--3: Comparrison betwe

een Absolutte Deflectio
on Measurements from
m
Survveying and Camera

Deflecttion from surrveying Deeflection froom camera Differencce

(inch) (inchh) (inch)

Location #1
# 0.35 0.344 0.01
Location #2
# 1.14 1.1 0.04
Location #3
# 0.7 0.8 0.1
 101

5.5 Trending Analysis

One focus of recent testing on the UP’s South Morrill subdivision is on using

trends in measurements made over time to better predict track performance and better

schedule maintenance. It is hoped that this will ultimately lead to a reduction in

derailments.

5.5.1 Data Pre-processing

Due to GPS misalignment, data from different tests may not exactly coincide with

each other in terms of milepost. Figure 5-19 shows the two sets of data from the same

section of track. An offset exists between the two curves.

Measurements from two tests over approximately 0.05 miles of track are shown in

Figure 5-19. The figure clearly shows that the measurements from both tests have

similar shape. However, the two sets of data have an obvious horizontal offset. The

offset represents errors in the milepost location associated largely with GPS error in

localizing the data. Despite errors in GPS measurement, the shape of the curve is

maintained because the relative GPS measurement (one data point with respect to the

next) is much more accurate than two independent absolute GPS measurements.
 102

Offset

Figure 5-19: The Original Data from Two Tests

For the purpose of trending, it is desirable to remove the offsets so that relative

comparisons can be made over short sections of track. The relative comparisons would

evaluate one measurement relative to a previous measurement made at the same location

at an earlier time.

To remove the offset in milepost, the cross correlation function is introduced to

mathematically quantify the offset. Cross correlation is a standard method of estimating

the degree of correlation between two sets of measurements. Consider two series x(i)

and y(i,), both of length N, where i=0,1,2...N-1. The cross correlation, R̂xy , at delay m

is defined as :

N −1
1
Rˆ xy (m) =
N
∑ x ( n) y ( n + m)
n =0
Equation 5-1

where m=-(N-1),…,-2 -1,0,1,2,…,N-1

 103

For various values of m, R̂xy is in the range − 1 ≤ Rˆ xy ≤ 1 . The bounds indicate

maximum correlation while 0 indicates no correlation. A high negative correlation

indicates a high correlation but where one series is the inverse of the other series.

The results of applying the cross correlation function to the two series in Figure

5-19 at numerous values of m (between -100 and 100) is shown in Figure 5-20. This

figure shows the cross correlation as a function of offsets. Clearly, the cross correlation

reached a maximum when the offset was 8 feet. The value of the cross correlation at this

offset is 0.8, indicating a high correlation between the data (this also suggests the

measurements are highly repeatable).

Figure 5-20: Cross Correlation

Based on the cross correlation plot, the second series in Figure 5-19 was shifted

by 8 feet and the new plot is shown as Figure 5-21. Now the two series of data line up

well, which is helpful and convenient for further data comparison and analysis.
 104

Figure 5-21: The Shifted Data from Two Tests

5.5.2 A Trending Example

Some trending results from the South Morrill subdivision are now presented.

Several tests were conducted over the same approximately 160 miles of this sub-division

at three-month intervals. Figure 5-22 shows the relative deflection measurement (Yrel)

over 0.1 miles of track from MP A.70 to A.80 (exact mileposts are removed). Here the

third axis (into the page) illustrates the time interval between the tests in months. Figure

5-22 clearly demonstrated the changes in this section of track as a function of time.

Specifically, two locations are singled out over time. It can be seen that MP A.76 is not

changing quickly over time while MP A.74 corresponds to a peak in the measurement

(soft spot) that is increasing over time (becoming softer).

 105

Figure 5-22: Data from 3 Tests at MP A.74

From the measurements shown in Figure 5-22, two sections at MP A.74 and A.76

are extracted as an example and shown as a function of time in Figure 5-23. Since three

tests were performed, three data points are shown in each of the plots for these milepost

locations. Given these data, a prediction can be made based on the trends. In this case

a line is fitted to the data and used as the prediction. A correlation can be produced to

indicate how well the line fits the three data points. Having only three points (three

tests) of course may not accurately predict the trend, and clearly the prediction has

uncertainty. More testing will improve the prediction. With more available tests it

may be desirable to use other curves (rather than linear extrapolation) to improve the

prediction.
 106

Figure 5-23: Trending at MP A.74 and A.76

To further illustrate the possible usefulness of this technique MP A.74 is shown

again in Figure 5-24. Here an assumption is made that a given threshold of relative

displacement, Yrel, would be undesirable. Based on previous measurements, a value of

1.2 inches is chosen and indicated by the red dashed line. Now, the linear prediction

can be used to estimate the time required to reach this threshold. In this case, five and a

half months from the last test is the window for maintenance. The accuracy of this

prediction is difficult to quantify, however, this is a tool that can be used to prioritize

maintenance based on actual track data. It is also possible to apply this technique to

other track measurements such as gauge, gauge restraint, cross-level, and other standard

measurements.
 107

Figure 5-24: Test Data at MP A.74 as a Function of Time

The above approach can now be applied over the entire section of track between

A.70 and A.80. This range is broken into 12 ft bin lengths and a curve fit is created for

each bin. The Yrel measurements are shown in Figure 5-25 for the three tests. A

subplot is created that represents the slope of the trending line (linear curve fit in Figure
∆
5-24). Here the slope, given by in units of inches per month, is approximated as
∆

the difference between the two tests. When both differences are equal it represents a

linear change over time. The figure shows that MP A.74 is changing approximately

linearly over time at a relatively fast rate while the other part of the track in this section

shows little change. The location of MP A.74 corresponds to a muddy road-crossing,

and the needed maintenance schedule can now be estimated (given some amount of

uncertainty).
 108

Figure 5-25: Trending from MP A.70 to A.74

5.5.3 Trending Results for Bridge Approaches

Data used in this section come from tests performed on the BNSF’s Creston, St.

Joseph, and Ravenna subdivisions. Tests were performed over two bridge approaches

as shown in Figure 5-26 and Figure 5-27. The two soft approaches to the bridge

corresponding to high Yrel can be identified in the figures. Both bridges are concrete

ballast deck bridges. Trending from two tests on the bridge at MP B.6 is shown in

Figure 5-26. An examination of the two measurements of Yrel indicates little change in

the measurement. This is confirmed in the trending analysis where the rate of change

was never more than 0.02 inch per month.

 109

Figure 5-26: A Bridge at MP B.6

Trending for a second bridge near MP C.64 is shown in Figure 5-27. Here, three

tests were performed, and again an examination of each of the Yrel measurements

indicates little change in the measurement near the bridge. This is confirmed in the

trending analysis. However, a few significant observations may be made. First, the

difference between the June and February tests is consistently more substantial than the

difference between the August and February tests. It could be speculated that more

moisture was present in the track structure in June as compared to both February and

August. Second, a significant change can be seen near MP C.60. Here the June test

indicates both stiff and soft locations with large variations in Yrel as compared to the

other tests. This is also clearly indicated in the trending results. It is suggested that such

a “blip” in the trending might warrant further investigation by a track inspector.

 110

Figure 5-27: A Bridge at MP C.63

5.6 Implementation of Track Geometry Data

Based on the analysis in section 3.4.2, combining Yrel data and space curve data

from the track geometry vehicle can provide information regarding vertical track

displacement. In other words, using ECO data calculated from rail profile data, the

element in Yrel contributed by the track geometry variation can be eliminated. ECO

data is calculated by using the equations in section 3.4.1. Since Yrel data and ECO data

come from two different measurement systems which have two separate GPS systems, an

offset between the position readings from each GPS system is expected. As a result, an

alignment algorithm is required for implementing geometry data into Yrel measurements.

The Yrel and ECO data were aligned by the same method used in the trending

analysis, which included calculating the cross correlation between the two sets of data
 111

and shifting the distance where maximum cross correlation occurs. The procedure is

done by a program written in C++ and described in detail as follows.

The geometry data are measured in increments of one foot, however Yrel data is

not. As a result, an interpolation is needed to transform Yrel data into one foot

increments so that the cross correlation function can be applied to the two sets of data.

This interpolation was done by the function “interp1” (one-dimensional data interpolation)

in MATLAB. The method “pchip” (piecewise cubic Hermite interpolation) was used.

After the Yrel data is interpolated into one foot increments, it is then divided into

sections of 300 ft. The length of 300 ft was chosen because it appeared to be able to

create higher cross correlation for most of the sections.

Then, the GPS latitude and longitude data for each 300-ft section are used to

extract the corresponding 300 ft of ECO data, as well as the 300 ft of ECO data before

and after the corresponding 300 ft. Altogether, 900 ft of ECO data will be available for

the next step of the procedure. For example, over the 300 feet (MP19.536 to MP19.478)

shown in Figure 5-28, the GPS latitude and longitude for the starting data point (A) is

given as 41.185069 -101.425016. The program searches through the ECO data and

finds the point where the GPS is the closest to point A which is 41.185073 -101.4250164

(point B). Then, the 600 ft after point B and 300 ft before point B are extracted for the

cross correlation calculation and data shifting.

 112

Figure 5-28: Data alignment for ECO and Yrel data

The next step is to calculate the cross correlation between the two series which

include the 300 ft after point A in the Yrel data and the 300 ft after point B in the original

ECO data. The program calls a subroutine to finish the calculation and return the cross

correlation values as a function of m (-150 ≤ m ≤ 150).

Now the value of m corresponding to the maximum cross correlation value is

determined. For the example data shown in Figure 5-28, the offset value was 43 ft and

the maximum cross correlation was 0.9351 which indicates a very high correlation

between the two data series. After the offset value is determined, the ECO data is

simply shifted by this offset distance. For the section shown in Figure 5-28, the 300

data points from the original ECO data (from the 43rd point after point B to the 342nd

point after B) are cut off to match the 300 ft of Yrel data. This process is then repeated

for every 300 ft long section until all 158 miles of data are analyzed.
 113

Finally, relative deflection values are simply calculated by subtracting “-ECO”

from Yrel.

5.7 Exception Criteria

The Yrel data presents a multitude of information about the track condition but

can be difficult to interpret. Therefore, a list of exception criteria is being propagated to

automate and simplify the interpretation of such rail deflection and track modulus

information. In this case, the exception criteria identify points of interest in the data.

5.7.1 Mathematical Formulation of Exception Criteria

The preliminary exception criteria are based on identifying distinct changes in the

condition of the track. Changes are identified relative to a mean and standard deviation

of the surrounding track. The mean may be calculated as:

n
di
μ=∑ Equation 5-2
i =1 n

Where μ is the mean, di is a single data point, and n is the number of data points

over a characteristic length of track. Currently, 0.1 miles is used as the characteristic

length.

Given the mean as calculated in Equation 5-2, the standard deviation can be

determined by:
 114

∑d i −μ
σ= i =1 Equation 5-3
(n − 1)

With these two definitions, several exception criteria may be generated. One

standard criterion is to create a deviation ratio by computing the ratio (σratio) of the

current deviation with the standard deviation such as:

di − μ
σ ratio = Equation 5-4
σ

This is often used in industry as a quality control metric. For example, a given

parameter should never vary beyond “six sigma”. The difficulty with this metric in the

evaluation of the relative deflection is that it can lead to false exceptions. For example,

a smooth section of track such as the one in Figure 5-4, has a consistent mean and the

standard deviation is very small (around 0.05”). As a result, any medium sized change

(e.g. greater than 0.30”) will create an exception when in reality this medium sized

change in data does not represent a problem.

A more basic criterion is to compare the difference, given by Δ, between an

individual data point and the mean as in:

Δ = di − μ Equation 5-5

This criterion is straightforward and easy to apply.

 115

5.7.2 Prioritized Exception Results

The above exception criteria may only capture a small percentage of the

information contained in the relative displacement data. However, past tests suggest

that they are useful in identifying track sections in need of maintenance.

A test was conducted on August 9, 2006 from Bill, WY to North Platte, NE.

The test was performed on the Union Pacific Railroad’s Powder River and South Morrill

subdivisions. The line consisted of double track with approximately 250 MGT/year on

main track No. 2 and 50 MGT/year on main track No. 1. The test was conducted in a

special work train consist. Testing was done at speeds of up to 60 mph and the total test

length was approximately 270 miles.

Table 5-4 shows the results of this test. Values for both the difference criteria

exceptions, Δ, and the deviation ratio exceptions, σratio, are shown. The sites are ranked

in descending order of the Δ criterion over the entire 270 miles of the test (i.e. #1 having

the largest value of Δ=1.424”, #2 having the second highest value of Δ=0.989”, and so

on). Under this criterion, mile post A.47 was the “worst” section over the 270 miles of

the test. Note that the exact mile post numbers were changed to letters so as not to

identify specific sites of track.

 116

Table 5-4: Prioritized Exceptions over 270 miles of Track

Track
MP Δ σ µ σ ratio
Feature
1 A.47 Signal 1.424 0.062 0.112 22.85
2 B.89 Turnout 0.989 0.065 0.142 15.12
3 C.97 Signal 0.973 0.100 0.121 9.72
4 D.65 Unknown 0.970 0.112 0.078 8.64
5 E.51 Turnout 0.919 0.122 0.099 7.56
6 F.95 Turnout 0.916 0.133 0.11 6.91
7 G.97 Unknown 0.828 0.127 0.129 6.53
8 H.30 9' CBC 0.815 0.115 0.097 7.10
9 I.17 Unknown 0.800 0.085 0.119 9.42
10 J.58 Turnout 0.796 0.083 0.118 9.56
11 K.43 Crossing 0.773 0.054 0.119 14.23
12 L.44 Crossing 0.753 0.120 0.098 6.30

Several observations may be made from this exception list. First, there is not an

exact correlation between the difference criterion, Δ, and the deviation ratio criterion,

σratio. For example, site MP F.95 is the sixth highest when ranked by the difference

criterion, but has a relatively small deviation ratio. This is a result of a large standard

deviation surrounding the turnout (a rough turnout). Therefore, even though there is a

large data reading at this site, the deviation ratio is relatively low (as compared to site

K.43 for example).

5.7.3 Consequences of the Exception Criteria

Post processing of the data indicated a correlation with two of the top ten

deflection locations and the location of two derailments. These were within 30 days of

the track modulus measurement date.

 117

The exception list was generated based on the Δ criterion and was independent of

the knowledge of the derailments. The site with the second highest difference

exception, MP B.89, coincided with the location of a defective field weld which caused

an eight car derailment 14 days after the test. The site with the tenth highest exception

at MP J.58 coincided with the location of a broken joint bar derailment 30 days after the

test.

The site with the second highest Δ was mile post B.89. The raw data from that

location are shown in Figure 5-29 for one mile of track. The exception at MP B.89 is

clearly visible with a large peak in relative displacement. The standard deviation is also

larger around this turnout. This large relative displacement suggested a problem with

the joint 14 days before it caused a derailment.

Figure 5-29: Site of Broken Field Weld 14 Days after Test

 118

The site with the tenth highest Δ was at mile post J.58. The raw data from that

location are shown in Figure 5-30 for one mile of track. The exception at MP J.58 is

also visible with a large peak in Yrel measurements. The standard deviation is also

larger around this turnout with both ends of the turnout clearly visible. The track failed 30

days after the data was collected. The assumption is that the low modulus continued to

degrade, resulting in increasingly larger defections until failure of the non-insulated joint.

Figure 5-30: Failed Non-Insulated Joint 30 days post-test

5.7.4 Using “Yrel+ECO” as an Exception Criterion

Following the procedure described in section 5.6, the data from the vertical track

deflection measurement system and track geometry car can be combined and the relative

deflections (“Yrel+ECO”) can be calculated. These “Yrel+ECO” results eliminate the

 119

effects of track geometry variations and are more directly related to the track supporting

foundation. Therefore, it is reasonable to consider “Yrel+ECO” as an exception

criterion. “Yrel+ECO” is used as a major criterion to select locations for field

investigations, and the subsequent visits suggest that this criterion is fairly useful in terms

of identifying track support problems.

5.8 Comparison between Different Measurement Systems

5.8.1 Comparison of Vertical Track Deflection (VTD) with Vehicle-Track

Interaction (VTI)

A track quality measurement system known as Vehicle Track Interaction (VTI) is

currently in use by the Union Pacific Railroad. This system has been successfully

integrated into revenue service. The VTI system uses accelerometers mounted on

various locations of a railroad vehicle to measure vertical accelerations. The system is

used on both locomotives and hopper cars and has been fully integrated into revenue

service with sophisticated communication and data processing infrastructure.

Results presented here are from a test conducted on a locomotive that passed over

the same section of track (the Powder River and South Morrill subdivisions) two days

after the vertical rail deflection measurements presented in section 5.7.2 and 5.7.3. The

relevant VTI results come from an accelerometer mounted to the axle of the locomotive

(below the suspension system). The acceleration data are used (with knowledge of the

locomotive’s mass and suspension characteristics) to estimate loads between the wheel

and rail.
 120

The VTI measurements and the vertical deflection measurements are

fundamentally different. For example, a smooth section of track with a low modulus

will show no significant difference in VTI measurements as compared to a smooth

section of high modulus track. However, this difference will be captured by the vertical

track deflection system.

5.8.2 Comparison of Revenue Service Results from VTD and VTI

A test of the same section of track (270 miles of the Powder River and South

Morrill subdivisions) was conducted using the VTI system on August 11, 2006 – two

days after the VTD measurements were made. Some results of this test are shown in

Table 5-5 as a comparison between the two measurements. Selections of the ranked

results from VTI measurements are shown in the left four columns. The right two

columns show how the same measurements appear in the Δ criterion exceptions from

Table 5-4. Only the locations common to both tables are shown in Table 5-5.
 121

Table 5-5: Prioritized Exceptions of VTI Data

VTI Acceleration Estimated Axle Δ

MP Rank of Δ
Rank (g) Force (lbf) (inch)
1 X.5 28.15 104,500 NA NA
: : : : : :
5 G.97 24.97 84,800 0.828 7
: : : : : :
27 J.58 27.27 69,650 0.083 10
: : : : : :
62 B.89 27.78 64,600 0.89 2
: : : : : :
123 A.47 22.09 59,000 1.424 1
: : : : : :
249 C.97 17.1 52,300 0.973 3

The most important conclusion to reach from Table 5-5 is that the two

measurements are fundamentally different. The highest force estimated from the VTI

data (at MP X.5) does not appear in the top twelve Δ criterion exceptions of Table 5-4.

The top VTD exception (at MP A.47) appears at #123 in the VTI measurements. The

two derailment locations described above appear on both lists (highlighted in both

tables). The derailment that occurred 14 days after the vertical modulus measurements

(MP B.89) appears as #2 on the VTD list and #62 on the VTI list. Similarly, the

derailment that occurred 30 days after the test (MP J.58) was #10 on the VTD list and

#27 on the VTI list. However, an enormous impact load at MP X.5 is not indicated in

the top 12 of the VTD measurements. Clearly, these are different measurements.
 122

5.8.3 Comparison of Revenue Service Results of Yrel and ECO

The relation between Yrel and ECO has been explained in section 3.4.2.

Although high correlations exist between Yrel and ECO, they are fundamentally different

measurements. Results from revenue service tests further confirm the fundamental

differences.

A test was conducted in South Morrill subdivision in April, 2007. The VTD

system and UPRR’s track geometry car (EC5) collected data over the same 160 miles of

track on the same day. Exception lists were produced based on Yrel measurements and

calculated ECOs from the geometry car’s space curve data.

Table 5-6 lists the top 20 locations in the VTD list along with their rankings in the

ECO list. Among the three highlighted sites, #4 and #17 in the list are the locations

where a derailment occurred one week before the testing day. The track was under

construction at these locations on the day of the test. The #1 item in the list is the

location where another derailment occurred two weeks before the test. Again, the track

was under construction on the day of the test.

It can be observed that nine of the top 20 sites in the VTD list did not show up on

the ECO list (ECO list includes 200 sites). Besides the derailment locations, only four

locations (#5, #9, #18 and #20 in VTD list) were identified by both top 20 lists. In

addition, these locations rank differently on each list.

 123

Table 5-6: Comparison between VTD and ECO (Ranked by VTD)

VTD Δ ECO ECO

MP DESCRIPTION
RANK (inches) RANK (inches)

1 1.18 1 1.1954 58.813 Derailment

2 1.13 N/A N/A 11.802 Unknown

3 1.06 N/A N/A 93.489 Signal

4 1.02 4 1.02 105.106 Derailment

5 0.94 7 1.0196 56.286 Road crossing

6 0.92 N/A N/A 14.37 Signal

7 0.9 N/A N/A 9.628 Unknown

8 0.88 93 0.625 55.307 Road crossing

9 0.87 28 0.8514 115.447 RR XING

10 0.86 12 1.0038 23.102 Signal

11 0.85 N/A N/A 38.938 Signal

12 0.84 N/A N/A 116.774 Unknown

13 0.83 N/A N/A 147.589 Unknown

14 0.83 NA N/A 31.294 Signal

15 0.8 43 0.7383 39.228 Road crossing

16 0.78 N/A N/A 100.149 Switch

17 0.76 25 0.8672 105.152 Derailment

18 0.76 9 1.0157 19.508 Road crossing

19 0.75 33 0.793 24.594 Road crossing

20 0.75 3 1.172 53.56 Road crossing

 124

A list ranked by ECO data was also produced as shown in Table 5-7. In this

table, 18 of the top 20 locations in the ECO list were also in the VTD list (VTD list

includes 90 sites). However, most of these locations have lower rankings in the VTD

list. Except for the derailment locations and five unknowns, almost all of the top 20

locations in the ECO list are road crossings.

 125

Table 5-7: Comparison between VTD and ECO (Ranked by ECO)

ECO ECO VTD

Δ(inches) DESCRIPTION
RANK (inches) RANK

1 1.1954 1 1.18 Derailment

2 1.1406 33 0.61 Road crossing

3 1.1172 20 0.75 Road crossing

4 1.0976 4 1.02 Derailment

5 1.0351 48 0.58 Crossover Switch

6 1.0312 72 0.52 Unknown

7 1.0196 5 0.94 Road crossing

8 1.0196 29 0.63 Culvert

9 1.0157 18 0.76 Road crossing

10 1.0156 22 0.7 Road crossing

11 1.004 82 0.51 Road crossing

12 1.0039 10 0.86 Road crossing

13 1.0001 81 0.51 Unknown

14 0.9961 46 0.58 Unknown

15 0.9687 56 0.55 Road crossing

16 0.9649 52 0.57 Road crossing

17 0.9414 59 0.55 Unknown

18 0.9336 N/A N/A Road crossing

19 0.9218 30 0.63 Unknown

20 0.8828 N/A N/A Road crossing

 126

5.9 Field Investigations

The VTD measurement system conducted a test in a coal train consist at the South

Morrill subdivision in June, 2008. About a month earlier, UPRR’s EC5 geometry car

collected track geometry data on the same subdivision. The track geometry data

provided by UPRR was then integrated into the VTD’s Yrel measurements so that Yrel

and ECO measurements could be compared and relative deflection measurements could

be calculated.

Based on both the VTD’s Δ criterion and the calculated relative deflection

measurements, 15 sites were selected and field investigations were completed on July 1st,

2008. The track visits identified the reasons for large Yrel and deflection measurements

at all sites with varying levels of maintenance urgency. Among them, one site was

taken out of service immediately and two additional sites were called for maintenance to

be repaired. A variety of track problems were identified including poor joints, broken

ties, muddy ballast, and crushed rail head, etc. Some of the sites will be discussed in

detail in the following sections. The actual mileposts for these sites are concealed in the

discussions.

5.9.1 Site 1: A Crushed Rail Head

The data from the first site is plotted in Figure 5-31. The peak in the Yrel data on

the north rail measures 0.89” while the ECO measurement is 0.34”. As a result, the

relative deflection is calculated as 0.55”.

 127

Figure 5-31: Data at the Crushed Rail Head Site

The pictures of this location are displayed in Figure 5-32. The white-colored

ballast indicates that strong pumping has occurred. The track inspector estimated the

pumping to be up to three quarters of an inch. The rail profile dropped down half an

inch at the crushed spot. This site was investigated two weeks after the test. The track

inspector suggested that the weather was damp and a flash flood warning was issued

around the testing time, so that the deflection was expected to be large.
 128

C
Crushed Rail Head

Figure 5-3
32: A Crush
hed Rail He
ead

5.9.2 Site 2: A Muddy Crossing

mber of roadd crossings are listed inn the VTD exceptions

A num e liist. Most of them

exxhibited fou
uled ballast and broken ties. Dataa from one particular site are show
wn in

Figure 5-33. On the soouth and norrth rails, tw

wo peaks of Yrel data may
m be obseerved.

T
They represen
nt the two appproaches too the road croossing. Thhe approach at
a the east siide of

b much woorse than thee west side based on thhe Yrel data. The
thhe crossing appears to be

deflections att both rails on

o the east appproach are almost
a 0.6”.
 129

Figure 5-33: Data at the Muddy Crossing Site

At least four ties were center cracked and some of them appeared to have

horizontal cracks at the bottom. The muddy ballast and the tie cavities indicated large

movements of the ties. Based on observation, the ties were estimated to have moved by

over 0.5”. This muddy area was located at an approach to a road crossing. The other

side of the road crossing was also muddy. This explains the two peaks in the Yrel data.

Again, the track was much wetter on the day of the test.
 130

Figure 5-34: The Muddy Crossing

5.9.3 Site 3: A Failing Joint

Approximately half of the sites in the VTD exception list are joints. The

measurement system demonstrated a notable ability to identify bad joints. Figure 5-35

shows the data at one particular joint. In this case, Yrel measured 1.21” at the joint with

an ECO reading of only 0.3”. Therefore, the deflection is over 0.8”.

 131

Figure 5-35: Data at the Failing Joint Site

Figure 5-36 shows the condition of the insulated joint. This supported joint is

rare in North America where suspended joints are dominant. The supporting tie in the

center was in a very poor condition. The tie was split and a large portion (left part in the

picture) was nearly separated from the rest of the tie. Applying pressure with a foot was

enough to deflect the tie. The bolts on the joint bars were loose and one of them would

actually rotate as a train passed. The track inspector commented that this joint was

probably among the worst 10% of all joints. A wayside camera measurement was taken

as a coal train passed the site. The video showed that the deflection at the joint was over

1.2 inches under the loads of the locomotives.

 132

Figure 5-36: The Failing Insulated Joint

5.9.4 Site 4: A Series of Broken Ties

The peak in the Yrel data from this site reads 0.95” while the ECO is 0.41”. The

site is located on a portion of tangent track with no joints or crossings. Field

measurements of the MCO (mid-chord offset) were consistent with the geometry car

measurements. As shown in Figure 5-37 the curve of Yrel data from this site has a

special shape when compared to the data from other sites. Unlike the sharp peaks in

Figure 5-35, the peak here is broader and high deflections occurred over a relatively long

distance.
 133

Figure 5-37: Data at the Broken Ties Site

A picture of this site is displayed in Figure 5-38 in which a series of broken ties

are clearly visible. The ties barely constrained the south rail. The clips were either

missing or unattached to the rail due to the cracks in the ties. The broken parts of the

ties could easily be lifted by hand. Recent geometry car tests and VTI tests did not

identify an exception at this location. The track inspector explained that defects like

these were difficult for track inspectors to notice. The track director immediately

stopped service on the track as shown in Figure 5-39.

 134

Figure 5-38: Six Broken Ties in a Row

Figure 5-39: Track Taken Out of Service

 135

5.10 Modulus Estimation

An important goal of the project is to obtain the track modulus information using

the system’s measurements. The relation between Yrel and track modulus based on the

Winkler model was discussed in Section 3.2. The track modulus determined by this

relation is an overall effective modulus. The method does not distinguish between

voids, poor ballast, soft subgrade, or broken rail components.

Using Equation 3-10 and Figure 3-5, the Yrel measurement can be converted into

track modulus. As mentioned earlier, Equation 3-10 is based on the analysis which

neglects track geometry variations. The discussions in Section 3.3 explained how using

“Yrel+ECO” would eliminate the effect of track geometry variations. Therefore, by

using the calculated relative deflection (“Yrel+ECO”) as Yrel in Equation 3-10, the

modulus measurement can be determined without the effects of track geometry.

The previous results may be implemented in further analysis of site four. The

Yrel and ECO data from the south rail of this site as shown in Figure 5-37 were

converted into track modulus. The calculated modulus over the same section of track

shown in Figure 5-37 is plotted in Figure 5-40. An extremely soft supporting

foundation is indicated since the modulus value drops from around 3,000 psi to a mere

200 psi.
 136

Figure 5-40: Track Modulus Calculated from Relative Deflection Data

 137

6 CONCLUSION

Track modulus is important because it significantly affects track performance and

maintenance requirements. Currently there is no vehicle available to measure track

modulus and track deflection at revenue speeds in real-time.

A system has been developed to make real-time vertical track deflection

measurements from a moving railcar. The deflection measurement can be used to

estimate track modulus based on mathematical models describing the relation between

loads and track deformation. The system consists of a loaded hopper car outfitted with a

camera/laser sensor system to detect the vertical deflection of the rail relative to the

wheel/rail contact point. In order to eliminate the effect of track geometry variations,

track geometry car data is introduced into the system.

All of the components of the measurement system have been described in detail.

The methods and procedures of image processing and calibration were illustrated. The

potential errors in the measurements were also analyzed, showing that the system can

provide fairly accurate and reliable measurements. Further improvements included

upgrading the sampling rate of the measurement system.

The measurement system has conducted revenue service tests over three thousand

miles of track. A special validation test was also performed. Based on the results from

these tests, a variety of analyses were conducted. The system’s measurements

demonstrated high repeatability. The influences of various testing conditions (testing

speed, testing seasons, rail size) on the testing results were also evaluated. Three

validation approaches have been developed, and results from the validation test confirm
 138

that the system measures accurate outputs. Trends in the data from different tests were

examined to better monitor the changes in the track quality. Exception criteria were

proposed and used to identify and prioritize track locations in need of maintenance.

These criteria proved to be valid, although further improvements are still possible. The

data from different systems (VTD and VTI, VTD and ECO) were then compared, and the

results showed that the VTD system provides unique and valuable information that is not

available from the other systems. Furthermore, the VTD system has notable ability to

indicate track support problems. Last but not least, the process of using the system’s

measurements to estimate track modulus was demonstrated.

 139

7 APPENDIX Program Flowchart

 140

8 REFERENCES

AREMA Manual, 1996, Chapter 16, 10.2.2,5

Arnold, R., Lu, S., Hogan, C., Farritor, S., Fateh, M., El-Sibaie, M., "Measurement of
Vertical Track Modulus From a Moving Railcar," Proceedings of the AREMA 2006
Annual Conference, Louisville, KY, September, 2006.

Beckwith, T. G., Marangoni, R. D., and Lienhard V, J. H., 1993, Mechanical

Measurements 5th Edition, Addison-Wesley Publishing Company, Inc. Reading, MA: pp.
464-469.
th
Boresi, Arthur P. and Schmidt, Richard J. 2003. Advanced Mechanics of Materials, 6
Edition. John Wiley & Sons, New York, NY: Chap. 5, 10.

Cai, Z., Raymond, G. P., and Bathurst, R. J., 1994, “Estimate of Static Track Modulus
Using Elastic Foundation Models,” Transportation Research Record 1470, pp. 65-72.

Carr, Gary A, 1999, “Dynamic Response of Railroad Track Induced by High Speed
Trains and Vertical Stiffness Transitions With Proposed Method of Measurement”,
Master Thesis, Tufts University. September 1999.

Carr, Gary A. and Greif, Robert, 2000. “Vertical Dynamic Response of Railroad Track
Induced by High Speed Trains,” Proc. of the ASME/IEEE Rail Joint Conference, pp.
135-151.

Chang, C. S., Adegoke, C. W., and Selig, E. T., 1980, “GEOTRACK Model for Railroad
Track Performance,” Journal Of the Geotechnical Engineering Division, Proc. Of the
American Society of Civil Engineers, November 1980, Vol. 106, No. GT11, pp
1201-1218.

Hogan, C., Dick, M., Lu, S., Farritor, S., Arnold, R., GeMeiner, W., Clark, D., "Track
Stiffness Measurement with Implementation to Rail/Vehicle Dynamic Simulation,"
Proceedings of the AREMA 2008 Annual Conference, Salt Lake City, UT, September,
2008.

Hogan, C., Design of a System to Measure Vertical Railroad Track Deflection From a
Moving Railcar and Implementation to Dynamic Railcar Simulation, Master Thesis,
University of Nebraska-Lincoln, August, 2007.
 141

Craig, J. J., 1989, Introduction to Robotics: Mechanics and Control, 2nd Edition,
Addison-Wesley Publishing Company, Inc., Reading, MA: pp 19-60 .

Davis, D. D., Otter, D., Li, D., and Singh, S., December 2003, “Bridge Approach
Performance in Revenue Service,” Railway Track & Structures, pp. 18-20.

Ebersohn, W., and Selig, E. T., 1994, “Track Modulus on a Heavy Haul Line,”
Transportation Research Record 1470, pp. 73-83.

Ebersohn, W., Trevizo, M. C., and Selig, E. T., 1993, “Effect of Low Track Modulus on
Track Performance,” International Heavy Haul Association, Proc. Of Fifth International
Heavy Haul Conference, pp. 379-388.

Federal Railroad Administration, March 2001, “T-16: FRA’s High Speed Research Car,”
Research Results, RR01-01.

Heelis, M. E., Collop, A. C., Chapman, D. N. and Krylov, V., 1999, “Predicting and
measuring vertical track displacements on soft subgrades,” Railway Engineering. 117

Kerr, Arnold D. “A Method for Determining The Track Modulus Using A Locomotive
Or Car On Multi-Axle Trucks,” Proceedings American Railway Engineering Association,
1983, Vol. 84, pp 269-286.

Kerr, Arnold D. “On the Stress Analysis of Rails and Ties,” Proceedings of American
Railway Engineering Association, 1976, Vol. 78, pp 19-43.

Kerr, A. D. and Shenton, H. W., “On The Reduced Area Method For Calculating The
Vertical Track Modulus,” American Railway Engineering Association Bulletin,
December 1985, Vol. 86, No. 703, pp 416-429.

Kierzenka J. and Shampine L. F., 2001, “A BVP Solver based on Residual Control and
the MATLAB PSE,” ACM TOMS, Vol. 27, No. 3, pp. 299-316.

Li, Dingqing, Hass, Kevin, and Meddah, Abe, June 2002, “Moving closer to
performance-based track geometry inspection,” Railway Track and Structures, pp. 15-17.

Li, Dingqing, and Selig, Ernest T., 1994, “Resilient Modulus for Fine-Grained Subgrade
Soils,” Journal of Geotechnical Engineering, Vol. 120, No. 6, pp 939-957.

Li, Dingqing, and Selig, Ernest T., 1998, “Method for Railroad Track Foundation Design
I: Development,” Journal of Geotechnical And Geoenvironmental Engineering, Vol. 68,
No. 7-8, pp 457-470.
 142

Li, Dingqing, Thompson, Randy, and Yoshino, Dio, 2003, “Evaluation of

track-gauge-strength degradation in revenue service,” Railway Track and Structures, pp.
19-21.

Lu, S., Arnold, R., Farritor, S., "On The Relationship Between Load and Deflection in
Railroad Track Structure," Proceedings of the AREMA 2008 Annual Conference, Salt
Lake City, UT, September, 2008.

Lu, S., Farritor, S., Arnold, R., GeMeiner, W., Clark, D., Al-Nazer, L., Carr, G.,
El-Sibaie, M., "Measurement of Vertical Track Modulus - Field Testing, Repeatability,
and Effects of Track Geometry," Proceedings of the 2008 Joint Rail Conference,
Wilmington, DE, April 22-24, 2008.

Lu, S., Hogan, C., Minert, B., Arnold, R., Farritor, S., Seeger, W., Oliver, M., Lees, H.,
Thornton, D., Fateh, M., Carr, G., "Vertical Track Modulus Testing On BNSFs Heavy
Haul Coal Line," Proceedings of the AREMA 2007 Annual Conference, Chicago, IL,
September, 2007.

Lu, S., Hogan, C., Minert, B., Arnold, R., Farritor, S., GeMeiner, W., Clark, D.,
"Exception Criteria in Vertical Track Deflection and Modulus," Proceeding of 2007
ASME/IEEE Joint Rail Conference & Internal Combustion Engine Spring Technical
Conference, Pueblo, CO, March 13-16, 2007.

McVey, B., Norman, C., Wood, N., Farritor, S., Arnold, R., Fateh, M., El-Sibaie, M.,
"Track Modulus Measurement From A Moving Railcar," Proceedings of the AREMA
2005 Annual Conference, Chicago, IL, September, 2005

McVey, B, A Nonlinear Approach to Measurement of Vertical Track Deflection From a

Moving Railcar, Master Thesis, University of Nebraska – Lincoln, May 2006.

Meyer, Marcus B. 2002, “Measurement of Railroad Track Modulus on a Fast Moving

Railcar,” Master Thesis, University of Nebraska – Lincoln, May 2002.

Norman, C., Measurement of Track Modulus From A Moving Railcar, Master Thesis,
University of Nebraska – Lincoln, May 2004.

Raymond, G. P., 1985, “Analysis of Track Support and Determination of Track

Modulus,” Transportation Research Record 1022, pp. 80-90.
 143

Read, D., Chrismer, S., Ebersohn, W., and Selig, E., 1994, “Track Modulus
Measurements at the Pueblo Soft Subgrade Site,” Transportation Research Record 1470,
pp. 55-64.

Selig, E. T., and Li, D., 1994, “Track Modulus: Its meaning and Factors Influencing It,”
Transportation Research Record 1470, pp. 47-54.

Stewart, Henry E., 1985, “Measurement and Prediction of Vertical Track Modulus,”
Transportation Research Record 1022, pp. 65-71.

Google map, http://maps.google.com.

Sussmann, T. R., Ebersohn, W., and Selig, E. T., 2001, “Fundamental Nonlinear Track
Load-Deflection Behavior for Condition Evaluation,” Transportation Research Record
1742, pp. 61-67.

Thompson, R., and Li, D., 2002, “Automated Vertical Track Strength Testing Using
TTCI’s Track Loading Vehicle,” Technology Digest, February.

Wanek, Mischa, 2004, “Quality instruments for rail-flaw detection,” Railway Track and
Structures, February, pp. 19-22.

Zarembski, Allan M. and Choros, John. “On the measurement and calculation of vertical
track modulus,” Proceedings of American Railway Engineering Association, 1980, Vol.
81, pp. 156-173.

Zarembski, A. M., and Palese, J., August 2003, “Transitions eliminate impact at
crossings,” Railway Track & Structures, pp. 28-30.