Diagnosis, Fault-tolerant and Robust Control for Quake Laboratory Mandatory Assignment in DTU Course 31320 Part A

Mogens Blanke Original version: March 5, 2014 version 9: March 19, 2014

Table 1:

Document revision sheet - list of changes

change kf uj are unknown vars, input is ij x 2 x 4 and x 2 x 6 C1 4 CUSUM detectors change change(s) Hyf (s)i(s) Hyf (s)f (s) u2 = k2 (1 fu2 )i2 u2 = kf (1 fu2 )i2 u3 = k2 (1 fu2 )i2 u3 = kf (1 fu3 )i3 u1 d, u2 i2 and u3 i3 ... u1 is used as a disturbance, unknown to you Cusum is to be applied on dataset 04, not 03 this version is void removed March 27 before 12:59 Correct dataset04, 05, 06 uploaded comment added remark added corrected removed clarication typo corrected typo corrected typo corrected clarication clarication typo corrected

version 2 2 2 2 3 4 5 5 6 6 7 8 8 9

where Table 1 after 1.9 Eq. 1.17 Question 2 Question 5 Eq. 1.19 Eq. 1.17 Question 2 State space model Question 6 Question 9 Deadline Question 6

changed

Chapter 1

Assignment part A - 2014

Introduction
This exercise presents a much simplied concept of damping of building response due wind load and a earthquake. The system considered is the laboratory shown in Figure 1.1. The variables are explained in Table 1.1. We are not able to shake the entire structure as an earthquake would do, we have three electro magnetic actuators that can provide a force on each of the oors. One actuator is used to act as an unknown disturbance from wind or a quake, the other two are used to damp the motion amplitudes. Acceleration is measured by an accelerometer on each oor. Floors are numbered {1, 2, 3} starting from below. Table 1.1: variable x1,3,5 x2,4,6 d1 u2,3 y1,2,3 unit [m] [m/s] [N ] [N ] m/s2

List of variables

description position of oors 1, 2, 3 [m] velocity of oors 1, 2, 3 [m/s] force due to wind or quake (on oor 1) actuators force on oors 2, 3 measured acceleration of oors 1, 2, 3

The dynamics of the quake laboratory setup is, when output is a vector of measured accelerations, the disturbance d1 acts on oor 1, actuation force u2

Figure 1.1: Structure of the building model.

and u3 act on oors 2 and 3, respectively, x 1 = x2 (1.1) 1 x 2 = ((k1 + k2 )x1 (c1 + c2 )x2 + k2 x3 + c2 x4 + d1 ) (1.2) m1 x 3 = x4 (1.3) 1 (k2 x1 + c2 x2 + (k2 + k3 )x3 (c2 + c3 )x4 + k3 x5 + c3 x6 + u2 ) x 4 = m2 (1.4) x 5 = x6 1 x 6 = (k3 x3 + c3 x4 k3 x5 (c3 )x6 + u3 ) m3 y1 =x 2 y2 =x 4 y3 =x 6. (1.5) (1.6) (1.7) (1.8) (1.9)

The electromagnetic actuators provide a force in proportion to the command current so u2 , and u3 provide a force through their respective force constants, uj = kf ij , 3 (1.10)

where kf being the force constant of the actuator and ij the current to the j th actuator. Note that this means uj are unknown variables, the known inputs are ij . The actuators have fast dynamics and a saturation both of which are neglected here. The parameters in Eq. (1.1 to 1.9) are listed in Table 1.2. Table 1.2: parameter mi ki ci kf

List of parameters
unit [kg ] [N/m] [N s/m] [N/A] description mass of oor i spring constant friction coecient actuator constant

value 2.78 1.78 103 0.25 4.2

1.1

State Space Model

= Ax + Bu x y = Cx + Du (1.11) (1.12)

The dynamic model in Equations 1.1 to 1.9, has the form:

Where u denotes the force from the three actuators, of which u1 is used as a disturbance, unknown to you, and the matrices are described by:
0 (k1 + k2 ) m1 0 A= k2 m2 0 0 1 (c1 + c2 ) m1 0 c2 m2 0 0 k2 m1 0 (k2 + k3 ) m2 0 k3 0 m3 0 0 1 0 m1 0 0 1 B= 0 m2 0 0 0 0 k2 m1 (k2 + k3 ) m2 k3 m3 0 c2 m1 1 (c2 + c3 ) m2 0 c3 m3 0 0 0 0 0 1 m3 c2 m1 (c2 + c3 ) m2 c3 m3 0 k3 m2 k3 m3 0 (1.15) 0 0 0 k3 m2 0 k3 m3 0

0 0 c3 m2 1 c3 m3

(1.13)

(1.14)

(k + k ) 1 2 m1 k2 C= m2 0

(c1 + c2 ) m1 c2 m2 0

c3 m2 c3 m3

 1 m1 D= 0 0

0 1 m2 0

(1.16)

0 1 m3

Question 1
Make a structural analysis to determine: a) a complete matching on the unknown variables, b) the parity relations in symbolic form, c) investigate the properties you nd relevant for a structural analysis. You need to allow integration through dierential constraints for this system, this means the initial value of velocity is known to you when you wish to integrate acceleration, and likewise for position being integrated from velocity. These initial values will be provided by a state observer, that you should disregard in your analysis. Now introduce the following faults in the model of the system, y1 y2 y3 u2 u3 = x 2 + fy1 = x 4 + fy2 = x 6 + fy3 = k2 (1 fu2 )i2 = k3 (1 fu3 )i3 (1.17)

where fy1 to fy3 are additive, and fu2 and fu3 are multiplicative.

Question 2
Determine the matrices A, B, C, D, Ex , Ey , Fx , Fy in a model (notation from chapter 3 in DFTC) where the multiplicative faults have been changed to additive ones. Input is i2 (t), i3 (t), disturbance is d(t), measurements y(t), faults are in f (t). Note: in the Qlab model description, we use u1 d, u2 i2 and u3 i3 , hence we have two known inputs, and one unknown disturbance. Hint: show rst how you can rewrite the actuator equation to u2 (t) and similarly for u3 . = kf (1 fu2 (t))i2 (t) = kf i2 (t) + 2 (t) (1.18)

Question 3
Calculate the transfer function matrices Hyi (s), Hyd (s), Hyf in the expression y (s) = Hyi (s)i(s) + Hyd (s)d(s) + Hyf (s)f (s) 5 (1.19)

Investigate which stepwise faults will give rise to a change in mean value of the output, and which will give oscillatory responses that will die out slowly. Investigate also the response in output to a stepwise disturbance.

Question 4
Determine a set of residual generators using a left nullspace design in the frequency domain. The script Qlab NullspaceDesignScript.m has been uploaded. Demonstrate through a simulation in Simulink that the residual generators you nd are insensitive to input and to disturbances. The attached Matlab les are recordings from the quake laboratory. The recordings are, nofaultnoinput01.mat is without disturbances and without input withfaultnoinput02.mat is with a fault but without input nofaultwithinput03.mat has input on the actuators and a disturbance acting part of the time withfault04.mat has input on the actuators and a disturbance acting part of the time, and it has one fault withfault05.mat is a vector residual (time history) that has a fault withfaultwithinput06.mat is a vector residual (time history) that has another fault

Question 5
Determine the variance of sensor noise in the three measurements y1 , y2 , y3 from dataset 01. Use the ARL method to determine the smallest mean value you could detect in the signal with a time to detect of less than 2 seconds and mean time between false alarms of more than 1 month. Let CUSUM detectors work with hypothetical 1 from 0.1 to 1 times the standard deviation of the signals y1 , y2 , y3 and test these on the dataset 02. Use 4 values of 1 in these intervals. Determine the interval in which the change(s) are and determine the time you could detect the change(s).

Question 6
Using the residual generator you designed above, test it on dataset 03 to verify insensitivity to input and disturbance. Dataset 04, comprises a fault on one of the sensors. You may use you own residual generator to make residuals. Alternatively, you may your CUSUM detector on dataset Qlab04.r1 and Qlab04.r2 to 6

detect the fault. Discuss, using the Hrf (s) why isolation of this fault is possible or not possible.

Question 7
Design a GLR detector and determine which fault is in dataset 05. Explain how you have chosen the threshold h and the window length M and discuss how the choice of these two values impact the diagnostic performance of your GLR test.

Question 8
Dataset 06 has an actuator fault. Discuss how such a fault could be detected, isolated and perhaps estimated. Demonstrate your method on dataset 06 or on the simulation model if you should wish.

Practical notes
Group work of two persons is encouraged. Groups cannot exceed two persons. If anybody are solving the assignment alone, you need only to solve questions 1,2,3 and 5,6,7. The deadline for the report was Thursday March 27, 2014, before 12:59. Please note that max. 18 pages are allowed for part A, excluding front matter. Any pages in excess of 18 (excluding front matter) will be discarded. Layout must be reasonable: A4 paper, 20mm margins and font size 11pt. The report need be delivered in electronic form, as a .pdf le, via the Campusnet assignment system. Only pdf les are accepted. Please do write your name(s) and student number(s) at the front page and as running head on each page of your your report. In addition, do not forget page numbers.