PME 3132

PME 3132: Maintenance Engineering and Management

Tutorial Questions

January 20, 2020

Introduction

1. Discuss 3 possible consequences of unforeseen failure.

• Costly downtime - loss of revenue due to non productive time

• Expensive maintenance - due to propagation of failures to other parts
• Catastrophic failures - some unforeseen failures could lead to loss of life, or damage
to the environment.

2. What do you understand by the term Prognostics and Health Management

(PHM)?

- Keywords: Condition monitoring, diagnostics, prognostics and a decision module (main-

tenance scheduling or reliability control).

3. What is the objective of PHM?

To increase Availability, Reliability and Safety of technical systems.

4. PHM involves fault diagnostics and prognostics. Why is prognostics more

challenging than diagnostics?

+ Diagnosis involves identifying an event that has already taken place.

+ Prognosis involves predicting an event that is yet to place and there are many uncer-
tainties such as manufacturing/installation errors, variation in operating conditions,
etc that make it difficult to predict future events.

5. What are the key challenges facing implementation of PHM technologies?

• Examples of fault progression are difficult to find due to periodic maintenance and
component replacement
• Sensor noise makes it hard to distinguish small, gradual deviations in performance
• Limited sensor sets
– Sparse set of sensors
– Limited information from sensors - Discrete open/closed sensors

1
 PME 3132

6. Explain a scenario where condition monitoring can be used in quality control.

+ Condition monitoring can be used to map the condition of a production machinery to

the quality of a product. Example: Tool wear in machining. Tool wear affects the
condition of a work piece such as surface roughness and monitoring tool wear can
help in establishing at what wear level the surface roughness deteriorates.
+ Process parameters in electro-discharge machining can be used to monitor the quality
of a product.

7. What are the three levels at which enabling technologies of PHM interact?

Figure 1: Interaction of PHM technologies at different levels

8. Apart from the maintenance department, what other departments in an or-

ganization would benefit from PHM?

Figure 2: PHM Technology consumers

2
 PME 3132

9. Discuss the main sources of information in the design of PHM systems.

• FMEA / FMECA - Failure modes, Effects (and Criticality) – which failure modes
to go after
• Fault Tree Analysis - Propagation Models
• Designers / Reliability Engineers - System knowledge and insight, Expected / nom-
inal behavior of the system
• Seeded Failure Testing / Accelerated Life Testing - Data, Failure signatures, Effects
of environmental conditions
• Fielded Systems - Sensors measurements, Maintenance logs, Fleet Statistics, Perfor-
mance Validation

10. When setting the failure threshold of a health index, why is it important to
incorporate a safety margin?
To avoid late failure prediction, to account for uncertainties (See Figure 3).

Figure 3: Defining thresholds

11. Explain/discuss how maintenance strategies have evolved with time. What is
the current state of the art maintenance?

Figure 4: Evolution of maintenance strategies

3
 PME 3132

PHM Costing

12. Using a Remaining Useful Lifetime diagram, discuss the benefits and limita-
tions of different maintenance strategies.

Figure 5: Effects of maintenance strategies

Table 1: Benefits and limitations of maintenance strategies

Maintenance strategy Benefits Limitations
Corrective maintenance - Low initial cost of implementation - Leads to expensive downtime since
faults may propagate to other com-
ponents. Loss of revenue due to un-
planned downtime.
- Maximizes useful lifetime - May lead to catastrophic failure.
Preventive maintenance - Avoids catastrophic failures - Does not maximize life of compo-
nents
- Reduces maintenance costs and - Cumulative maintenance cost are
unscheduled downtime high
Condition-based mainte- - Avoids catastrophic failures and - Initial cost of implementation may
nance unscheduled downtime, hence in- be high.
creasing reliability, availability and
safety.
- Maximizes the lifetime of compo-
nents/units.

13. How would you reduce the implementation costs of a condition monitoring
system?

• Check for already installed sensors

• Check if performance evaluation data can be used for monitoring
• Subcontracting maintenance services - Use of industrial IoT Softwares or PdM 4.0
Softwares

4
 PME 3132

14. Since avoidance of failure may not be practically possible, what is the main-
tenance tradeoff taken into account when costing?

• Perform maintenance so that the remaining useful life (RUL) and is minimized (use
up as much of the life as possible), while simultaneously avoiding failures (unsched-
uled maintenance) or
• Find the optimum mix of scheduled and unscheduled maintenance that minimizes
the life-cycle cost

15. With respect to maintenance costing, how is the optimum maintenance inter-
val/time obtained?

Figure 6: Maintenance costing

16. Explain a situation where corrective maintenance would be preferred com-

pared to condition based maintenance.

+ Systems that have constant failure rate - from the maintenance cost per unit time
curve, the minimum cost occurs at the end of life.

17. How would you establish if there is any benefit in installing a condition mon-
itoring system?
By evaluating the return on investment (ROI).

Return - Investment
ROI = .
Investment
• If ROI < 0 there is no direct cost benefit
• If ROI = 0 break even point (no impact on cost)
• If ROI > 0 there is a direct cost benefit - worth investing.

5
 PME 3132

General Condition Monitoring

18. What are the key elements of condition based maintenance and how are they
connected?

Figure 7: Elements of condition monitoring

19. What are the factors to consider when designing a condition monitoring sys-
tem?

• Critical components of the machine - those prone to failure and contribute to huge
downtimes
• Measurable quantities that may indicate condition - helps in sensor selection
• Presence of integrated sensors, e.g. for performance evaluation
• Kind of data, single value, etc. - helps in selection of sampling frequency and time
• Expected lifetime of component.
• Data logging intervals

20. Assuming you are installing a condition monitoring system onto a rotating
machinery (vibration measurements). Considering that the signal is analogue,
how would you reduce the storage/archiving requirements?

• Identify the appropriate data logging intervals - based on expected lifetime of the
equipment.
• Identify the appropriate sampling duration and frequency - Nyquist sampling rule.
• Sampling duration should be longer than the rotational frequency.

21. Explain why expert knowledge is still important in condition monitoring.

• Expert knowledge helps in defining failure thresholds

• Expert knowledge can be used in prioritizing critical components.
• Helps in signal processing and isolation of faulty components

6
 PME 3132

22. Not all components in a machinery require monitoring. How/what tools can
you use to identify critical components

• Failure mode and effect analysis (FMEA/FMECA)

• Fault tree analysis
• Risk assessment

23. Figure 16 shows a condition monitoring set-up of a ball bearing while (b) shows
the frequency spectrum of a faulty bearing resulting from accelerometer A1
measurements. The shaft rotates at a speed of 1350 rpm. The characteristic
frequencies are BPFI = 107 Hz, BPFO = 73 Hz, BSF = 57.5 and FTF = 9
Hz. Diagnose the most likely fault on the bearing and give reasons.

0.1

0.08
Amplitude (g)

Output 0.06

shaft 0.04
0
0.02

0
0 100 200 300 400 500
Frequency (Hz)

(a) Ball bearing section (b) Frequency spectrum from accelerometer A1 measure-
ments
Figure 8: Bearing condition monitoring

From the frequency spectrum, BP F I and its harmonics are present at high amplitudes
signifying a fault on the inner ring of the bearing.

24. Figure 17(a) shows the vibration signal of a boiler circulating pump operating
at a shaft speed of 1800 rpm, while (b) shows the FFT frequency spectrum.
The pump impeller has 6 vanes. Identify the possible cause(s) of the pump
vibration.
Ans: Cracked/Broken Vane - vane characteristic frequency is present, Misaligned shaft -
high amplitude at shaft frequency

5
0.2
X: 180.1
X: 30.38
Amplitude (g)

Amplitude (g)

0.15 Y: 0.2225
Y: 0.1137
0
0.1

0.05

-5 0
0 0.2 0.4 0.6 0.8 1 0 100 200 300 400 500
(a) Time (s) (b) Frequency (Hz)

Figure 9: Pump condition monitoring data

7
 PME 3132

25. Figure 10 shows the frequency spectrum of a faulty bearing resulting from
accelerometer measurements. The shaft rotates at a speed of 1500 rpm. The
bearing parameters are given in Table 2. Diagnose the most likely fault on
the bearing.
Inner ring fault - BPFI and its harmonics present

0.25

0.2 X: 235.7
Amplitude (g)

X: 118 Y: 0.1418
0.15 X: 24.67 Y: 0.206
Y: 0.09056
0.1

0.05

0
50 100 150 200 250 300 350 400
Frequency (Hz)
Figure 10: Frequency spectrum of a faulty bearing

Table 2: Table Q4(c)

Contact angle (φ) Pitch diameter Ball diameter Number of balls

0o 1.245” 0.235” 8

PHM Methodologies

26. Discuss the various categories of condition monitoring data.

• Single value type, where the only one sample is recorded at each prediction interval.
Typical examples include quasi-static condition monitoring data such as tempera-
ture, pressure, load, oil particle count.
• Continuous signal type, where at each prediction interval, the signal is sampled for
a predetermined duration at a specified sampling frequency, e.g. vibration, force,
electric voltage, electric current.
• Multi-dimensional type, e.g. thermal images, X-Ray images.

27. Describe the possible condition monitoring data for a gearbox

• Vibration measurement
• Temperature
• Oil particle/debris count
• X-Ray images

8
 PME 3132

28. What hardware do you need to collect condition monitoring data from machines or pro-
duction lines?

• Data acquisition card (used to log production signals)

• Sensors
• Ethernet cable
• Display device (PC or tablet)

29. What do you understand by the term feature extraction? In what domains
can features be extracted from raw data?
Dimensional reduction and deriving values (features) intended to have useful information
and non-redundant. Time domain, Frequency domain, Time-frequency domain.

30. Why is feature selection important when using data-driven condition moni-
toring? How is selection done?

• Not all features contain useful information on condition of a system.

• Monotonicity for regression
• Separation distance for fault/health state classification.

31. What is the importance of time synchronous averaging (TSA). What kind of
systems would TSA be most suited?

• The TSA is well suited for gearbox analysis.

• It allows the vibration signature of the gear under analysis to be separated from
other gears and noise sources in the gearbox that are not synchronous with that
gear
• Variations in shaft speed can be corrected, such that the spreading of spectral energy
into an adjacent gear mesh bin is minimized

32. Fault diagnosis involves 3 key elements. What are they?

+ Diagnostics involve: Anomaly detection → Isolation of faulty component → Identifi-

cation of fault.

33. Explain how reliability based prognostics can be implemented. What are its
advantages and disadvantages?
Utilizes distributions of historical failure times of similar systems. Usually for low risk
systems with no sensor network, e.g. electronic components.

34. Using a block diagram, discuss the workflow of data driven diagnostics/prognostics
based on machine learning. What are the main challenges involved in imple-
menting a data-driven based prognostics approach?

9
 PME 3132

Figure 11: Reliability based PHM

Table 3: Advantages/disadvantages of reliability based phm

Advantages Disadvantages
No condition monitoring data is required Low accuracy
Easy to implement No information on incipient faults
Difficult to implement at component level

Figure 12: ML based data driven methods

35. Discuss the components of model-based diagnostics/prognostics.

36. Explain how a simple 2-state discrete event fault model works.

37. How can you apply unsupervised machine learning (clustering) in condition
monitoring?
Anomaly detection. If the data clusters in more than one group.

38. Explain how classification algorithms can be used for prognostics.

10
 PME 3132

Figure 13: Model-based PHM

Figure 14: Simple 2 state FMS fault model

Through health state estimation.

39. Describe possible ways in which the target in machine learning can be defined
for prognostics. What is/are the limitation(s) of this/these approaches?

• RUL - Not suitable for highly variable data

• HI - Need a method to derive the RUL
• Virtual HI - Need a method to derive RUL

40. What are the sources of uncertainty in prognostics? How can the uncertainties
be incorporated in prognostic methods?

• Modeling uncertainties –Epistemic: Numerical errors, Unmodeled phenomenon, Sys-

tem model & Fault propagation model
• Input data uncertainties –Aleatoric: Initial state (damage) estimate, Variability in
the material, Manufacturing variability
• Measurement uncertainties –Prejudicial: Sensor noise, Sensor coverage, Loss of in-
formation during preprocessing, Approximations and simplifications

11
 PME 3132

• Operating environment uncertainties –Combination: Unforeseen future loads, Un-

foreseen future environments, Variability in the usage history data

These uncertainties can be represented as a probability distribution on the initial state

or within the features for machine learning.

Structural Health Monitoring

41. Describe two scenarios (example of structures) that demonstrate the impor-
tance of structural health monitoring.

• Failure of wind turbine mast due to storms - weak points such as cracks in welds
initiate failure
• Damage to aircraft fuselage due to cracks at fastened joints
• Damage of river bridges due to floods
• Failure of bridge due to fatigue loads

42. Describe two methods commonly used in SHM.

(a) Passive monitoring - utilizing excitations from normal usage of structure - acoustic
emission sensors
(b) Active monitoring - involves periodic excitation of structure and measurement of the
corresponding response - piezoelectric sensors

43. How is SHM different from NDT?

• SHM utilizes integrated sensors in the structure and can be used for continuous
monitoring. It minimizes human involvement.
• NDT involves transporting the equipment to the equipment/structure site where the
testing is done.

44. As the chief Engineer of a DB, you have identified that there is a section of a
bogie that fails frequently at a point where it is riveted. Describe a monitoring
method that you would propose for this system.

Performance evaluation

45. What is the importance of performance evaluation of condition monitoring

techniques?

• For end-of-life predictions of critical systems, it becomes imperative to establish a

fair amount of faith in the prognostic systems before incorporating their predictions
into the decision-making process.

12
 PME 3132

Figure 15: Elements of SHM

• A maintainer needs to know how good the prognostic estimates are before he/she
can optimize the maintenance schedule.
• Algorithms should be tested rigorously and evaluated on a variety of performance
measures before they can be certified.

46. What would you say is the current trend in condition monitoring?

• Ensemble methods
• Big data analytics, IoT
• Deep learning

13
 PME 3132

Figure 16 shows a condition monitoring set-up of a ball bearing while (b) shows the
frequency spectrum of a faulty bearing resulting from accelerometer A1 measure-
ments. The shaft rotates at a speed of 1350 rpm. The characteristic frequencies
are BPFI = 107 Hz, BPFO = 73 Hz, BSF = 57.5 and FTF = 9 Hz. Diagnose the
most likely fault on the bearing and give reasons.

0.1

0.08
Amplitude (g)
Output 0.06

shaft 0.04
0
0.02

0
0 100 200 300 400 500
Frequency (Hz)

(a) Ball bearing section (b) Frequency spectrum from accelerometer A1 measure-
ments
Figure 16: Bearing condition monitoring

Figure 17(a) shows the vibration signal of a boiler circulating pump operating at a
shaft speed of 1800 rpm, while (b) shows the FFT frequency spectrum. The pump
impeller has 6 vanes. Identify the possible cause(s) of the pump vibration.

5
0.2
X: 180.1
X: 30.38
Amplitude (g)

Amplitude (g)

0.15 Y: 0.2225
Y: 0.1137
0
0.1

0.05

-5 0
0 0.2 0.4 0.6 0.8 1 0 100 200 300 400 500
(a) Time (s) (b) Frequency (Hz)

Figure 17: Pump condition monitoring data

14