Figure 1.1 |
Core prognostic frameworks in a PHM system. Source: based on IEEE (2017). |
Figure 1.2 |
Framework diagram for a PHM system. Source: based on CAVE3 (2015). |
Figure 1.3 |
Graph of the exponential CDF with λ = 3. |
Figure 1.4 |
Graph of the Weibull CDF with β = 1.2 and η = 5. |
Figure 1.5 |
Graph of the exponential PDF with λ = 3. |
Figure 1.6 |
Graphs of gamma PDFs. |
Figure 1.7 |
Graphs of Weibull PDFs. |
Figure 1.8 |
Failure rates of Weibull variables. |
Figure 1.9 |
Failure rates of gamma variables. |
Figure 1.10 |
Failure rate of the standard normal variable. |
Figure 1.11 |
Failure rate of the lognormal variable with μ = 0 and σ = 1. |
Figure 1.12 |
Logistic failure rate with μ = 0 and σ = 1. |
Figure 1.13 |
Gumbel failure rate. |
Figure 1.14 |
Log‐logistic failure rate with μ = 0. |
Figure 1.15 |
Integration domain. |
Figure 1.16 |
High‐level block diagram of a PHM system. |
Figure 1.17 |
A framework for CBM for PHM. Source: after CAVE3 (2015). |
Figure 1.18 |
Taxonomy of prognostic approaches. |
Figure 1.19 |
Example of an FFP signature – a curvilinear (convex), noisy characteristic curve. |
Figure 1.20 |
Ideal DPS transfer curve superimposed on an FFP signature. |
Figure 1.21 |
Ideal DPS, degradation threshold, and functional failure. |
Figure 1.22 |
Normalized and transformed FFP and DPS transformed into FFS. |
Figure 1.23 |
Ideal FFS – transfer curve for CBD. |
Figure 1.24 |
Variability in DPS transfer curves. |
Figure 1.25 |
FFS transforms of the DPS plots shown in Figure 1.23. |
Figure 1.26 |
FFS and prognostic information. |
Figure 1.27 |
FFS transfer curve exhibiting distortion, noise, and change in degradation rate. |
Figure 1.28 |
Example plots of an ideal RUL and ideal PH. |
Figure 1.29 |
Example plots of an ideal SoH transfer curve and PH accuracy. |
Figure 1.30 |
Example of RUL with an initial‐estimate error of 100 days. |
Figure 1.31 |
Random‐walk with Kalman‐like filtering solution for a high‐value initial‐estimate error. |
Figure 1.32 |
Random‐walk with Kalman‐like filtering solution for a low‐value initial‐estimate error. |
Figure 1.33 |
Example of FFS data exhibiting an offset error, distortion, and noise. |
Figure 1.34 |
Utility function of price. |
Figure 1.35 |
Utility function of expected lifetime. |
Figure 1.36 |
Shape of the objective function. |
Figure 2.1 |
Block diagram showing three approaches to PHM. Source: based on Pecht (2008). |
Figure 2.2 |
Precision and complexity: relative comparison of classical PHM approaches. |
Figure 2.3 |
Model‐based approach to development and use. |
Figure 2.4 |
Model‐use diagram. Source: based on Medjaher and Zerhouni (2013). |
Figure 2.5 |
A framework for CBM for PHM (CAVE3 2015). |
Figure 2.6 |
Transition diagram. |
Figure 2.7 |
Example of a fault tree showing an RTD fault and an RTD‐usage model. |
Figure 2.8 |
Example plots of Weibull distributions. |
Figure 2.9 |
A special system structure. |
Figure 2.10 |
HALT result – 30 of 32 FPGA devices failed (Hofmeister et al. 2006). |
Figure 2.11 |
Family of failure curves, failure distribution, and TTF. |
Figure 2.12 |
Diagram of data‐driven approaches. |
Figure 2.13 |
A special neural network. |
Figure 2.14 |
Comparison of model‐based (PoF) and data‐driven prognostic approaches. |
Figure 2.15 |
Relative comparison of PHM approaches – PoF, data‐driven, and hybrid. |
Figure 2.16 |
Example diagram of a heuristic‐based CBM system using CBD‐based modeling. |
Figure 2.17 |
Diagram comparison of model‐based and CBD‐signature approaches to PHM. |
Figure 2.18 |
Simplified diagram of a switch‐mode power supply (SMPS) with an output filter. |
Figure 2.19 |
Unreliability plots for three models for capacitor failures (Alan et al. 2011). |
Figure 2.20 |
Example of the output of a SMPS. |
Figure 2.21 |
Example of a ringing response from an electrical circuit to an abrupt stimulus. |
Figure 2.22 |
Relationship of prognostic specifications (PD and PDα) to RUL and PH. |
Figure 2.23 |
Simulated change in resonant frequency as filter capacitance degrades. |
Figure 2.24 |
Experimental change in resonant frequency as filter capacitance degrades. |
Figure 3.1 |
Diagram of classical and CBD prognostic approaches for PHM systems. Source: based on Pecht (2008). |
Figure 3.2 |
Functional block diagram for CBD signature data and processing flow. |
Figure 3.3 |
Example of CBD containing feature data and noise (FD + Noise). |
Figure 3.4 |
Example of signals at an output node of an SMPS. |
Figure 3.5 |
Example of a damped‐ringing response (Judkins and Hofmeister 2007). |
Figure 3.6 |
Modeling a damped‐ringing response. Source: based on Judkins and Hofmeister (2007). |
Figure 3.7 |
Example of a CBD signature: FD is the resonant frequency of a damped‐ringing response. |
Figure 3.8 |
CBD signature and levels for SMPS lot 1 (top) and SMPS lot 2 (bottom). |
Figure 3.9 |
Functional block diagram for FFP signature and processing flow. |
Figure 3.10 |
FFP signatures using a fixed value and a calibrated value for nominal frequency. |
Figure 3.11 |
FFP signature, calibrated value for nominal frequency, failure threshold at 0.6 and 0.7. |
Figure 3.12 |
FFS signatures for FL = 0.6 (top) and FL = 0.7 (bottom). |
Figure 3.13 |
DPS and FFP signatures for data shown in Figure 3.11. |
Figure 3.14 |
FFP and DPS Showing FL = 0.4 and FL = 0.5. |
Figure 3.15 |
DPS‐based FFS (FL = 0.65) and FFP‐based FFS (FL = 0.7). |
Figure 3.16 |
Examples of signatures: decreasing (top) and increasing (bottom) slope angles. |
Figure 3.17 |
Other examples of signatures: decreasing (top) and increasing (bottom) slope angles. |
Figure 3.18 |
Simulated CBD‐based signature: FD = CBD − NM. |
Figure 3.19 |
Differences: experimental and simulated signatures. |
Figure 3.20 |
Simulated (top) and comparison of experiment (bottom) FFP signatures. |
Figure 3.21 |
Simulated DPS (top) and experimental DPS (bottom). |
Figure 3.22 |
Simulated FFS from simulated (top) and from experimental DPS (bottom): FL = 0.65. |
Figure 3.23 |
Illustration of point‐by‐point FNL comparison. |
Figure 3.24 |
Illustration of total FNLE comparison. |
Figure 3.25 |
Example plot of FFP‐based and DPS‐based FNLi. |
Figure 3.26 |
SMPS output and extracted damped‐ringing response. |
Figure 3.27 |
CBD signature and FFP signature. |
Figure 3.28 |
FFP‐based FFS and DPS. |
Figure 3.29 |
DPS‐based FFS. |
Figure 3.30 |
Procedural diagram for producing a DPS‐based FFS. |
Figure 3.31 |
Linear DPS (left side) and nonlinear DPS (right side) data plots. |
Figure 3.32 |
Simulated CBD plots using nonlinear and linear degradation rates. |
Figure 3.33 |
Comparison of ideal DPS to DPS from data for a nonconstant degradation rate. |
Figure 3.34 |
Node‐based framework for supporting failure progression signatures. |
Figure 4.1 |
Block diagram for offline modeling of CBD signatures. |
Figure 4.2 |
Flow diagram for developing signature models. |
Figure 4.3 |
Power function #1: increasing curves, decreasing slope angles. |
Figure 4.4 |
Power function #1: increasing curves, increasing slope angles. |
Figure 4.5 |
Power function #2: decreasing curves, decreasing slope angles. |
Figure 4.6 |
Power function #2: decreasing curves, increasing slope angles. |
Figure 4.7 |
Power function #3: increasing curves, vertically asymptotic, dPi < P0. |
Figure 4.8 |
Power function #4: decreasing curves, vertically asymptotic, dPi < P0. |
Figure 4.9 |
Power function #5: increasing curves, horizontally asymptotic. |
Figure 4.10 |
Power function #6: decreasing curves, horizontally asymptotic. |
Figure 4.11 |
Power function #7: increasing curves, slightly curvilinear, decreasing slope angles. |
Figure 4.12 |
Power function #7: increasing curves, slightly curvilinear, increasing slope angles. |
Figure 4.13 |
Power function #8: decreasing curves, slightly curvilinear, decreasing slope angles. |
Figure 4.14 |
Power function #8: decreasing curves, slightly curvilinear, increasing slope angles. |
Figure 4.15 |
Power function #9: increasing curves, vertically asymptotic, dPi < P0. |
Figure 4.16 |
Power function #9: increasing curves, horizontally asymptotic, dPi < P0. |
Figure 4.17 |
Power function #10: decreasing curves, vertically asymptotic, dPi < P0. |
Figure 4.18 |
Power function #10: decreasing curves, horizontally asymptotic, dPi < P0. |
Figure 4.19 |
Exponential function #11: increasing curves, vertically asymptotic. |
Figure 4.20 |
Exponential function #12: decreasing curves, vertically asymptotic. |
Figure 4.21 |
Exponential function #13: increasing curves, horizontally asymptotic. |
Figure 4.22 |
Exponential function #14: decreasing curves, horizontally asymptotic. |
Figure 4.23 |
Simulated FFP signatures: power function #1. |
Figure 4.24 |
Simulated DPS from FFP signatures: power function #1. |
Figure 4.25 |
Simulated FFP signatures: power function #3. |
Figure 4.26 |
Simulated DPS from FFP signatures: power function #3. |
Figure 4.27 |
Simulated FFP signatures: power function #5. |
Figure 4.28 |
Simulated DPS from FFP signatures: power function #5. |
Figure 4.29 |
Simulated FFP signatures: power function #7. |
Figure 4.30 |
Simulated DPS from FFP signatures: power function #7. |
Figure 4.31 |
Simulated FFP signatures: power function #9. |
Figure 4.32 |
Simulated DPS from FFP signatures: power function #9. |
Figure 4.33 |
Simulated FFP signatures: exponential function #11. |
Figure 4.34 |
Simulated DPS from FFP signatures: exponential function #11. |
Figure 4.35 |
Simulated FFP signatures: exponential function #13. |
Figure 4.36 |
Simulated DPS from FFP signatures: exponential function #13. |
Figure 4.37 |
Simulated FFP and DPS signatures: power function #1 for n = 2.0. |
Figure 4.38 |
Simulated FFP and DPS signatures: power function #3 for n = 2.0. |
Figure 4.39 |
Simulated FFP and DPS signatures: power function #5 for n = 1.5. |
Figure 4.40 |
Simulated FFP and DPS signatures: power function #7 for n = 0.75. |
Figure 4.41 |
Simulated FFP and DPS signatures: power function #9 for n = 0.25. |
Figure 4.42 |
Simulated FFP and DPS signatures: exponential function #11 for P0 = 100`. |
Figure 4.43 |
Simulated FFP and DPS signatures: exponential function #13 for P0 = 150. |
Figure 4.44 |
Example plots: FFS (top) and FFP and DPS (bottom): power function #1. |
Figure 4.45 |
Example plots: FFS (top) and FFP and DPS (bottom): power function #5. |
Figure 5.1 |
Example of a non‐ideal FFP signature and an ideal representation of that signature. |
Figure 5.2 |
Plots of a family of FFP signatures and DPS transfer curves. |
Figure 5.3 |
Plots of a curvilinear FFP, the transform to a linear DPS (top), and the transform to an FFS (bottom). |
Figure 5.4 |
Offline phase to develop a prognostic‐enabling solution of a PHM system. |
Figure 5.5 |
Diagram of an online phase to exploit a prognostic‐enabling solution. |
Figure 5.6 |
Plot of a non‐ideal CBD signature data: noisy ripple voltage, output of a switched‐mode regulator. |
Figure 5.7 |
Example plots: non‐ideal FFP signature data (top) and transformed DPS data (bottom). |
Figure 5.8 |
Example plots: non‐ideal and ideal FFS data (top) and FNL (bottom). |
Figure 5.9 |
Example plots: non‐ideal and ideal FFS data (top) and FNL (bottom) after using a NM of 3.0 mV. |
Figure 5.10 |
Example plot: non‐ideal ADC transfer curve. |
Figure 5.11 |
Example plots: ideal input and non‐ideal output from an ADC. |
Figure 5.12 |
Temperature (a), voltage (b), and current (c) plots. |
Figure 5.13 |
Temperature‐dependent (a) and temperature‐independent (b) plots of calculated resistance. |
Figure 5.14 |
Example of a damped‐ringing response. |
Figure 5.15 |
Sampling diagram for Example 5.8. |
Figure 5.16 |
ADC example: saw‐tooth input, sampling, digital output value. |
Figure 5.17 |
Simulated data before (top) and after (bottom) filtering of white (random) noise. |
Figure 5.18 |
Degradation signature exhibiting a change in shape. |
Figure 5.19 |
Experimental data: temperature measurements for a jet engine. |
Figure 5.20 |
Differential signature from temperature measurements for each of two engines. |
Figure 5.21 |
Temperature data and differential signatures: four engines on an aircraft. |
Figure 5.22 |
Composite differential‐distance signature. |
Figure 5.23 |
Example of a noisy FFP signature. |
Figure 5.24 |
Example of a noisy FFP signature: calculated nominal FD value, reduced NM value. |
Figure 5.25 |
Example of a smoothed (3‐point moving average) FFP signature. |
Figure 5.26 |
Example of a DPS from a smoothed FFP signature. |
Figure 5.27 |
Example of an FFS from a smoothed FFP signature. |
Figure 5.28 |
Example of a {FNLi} plot from a smoothed FFS. |
Figure 5.29 |
Example random‐walk paths and FFS input. |
Figure 5.30 |
Example plots of input FFS data and adjusted FFS data. |
Figure 5.31 |
Example of a {FNLi} plot from an adjusted FFS. |
Figure 5.32 |
Ripple voltage: plots of an unsmoothed (top) and smoothed (bottom) FFP signature. |
Figure 5.33 |
Ripple voltage: plots of an unsmoothed (top) and smoothed (bottom) FFS data. |
Figure 5.34 |
Smoothed FFS: for NM = 3.0 mV (top) and for NM = 2.0 mV (bottom). |
Figure 6.1 |
Core prognostic frameworks in a PHM system. Source: after IEEE (2017). |
Figure 6.2 |
A framework for CBM‐based PHM. Source: after CAVE3 (2015). |
Figure 6.3 |
Random walk with Kalman‐like filtering solution for a high‐value initial‐estimate error. |
Figure 6.4 |
Random walk with Kalman‐like filtering solution for a low‐value initial‐estimate error. |
Figure 6.5 |
Block diagram showing three approaches to PHM. |
Figure 6.6 |
Model development and use. |
Figure 6.7 |
Diagram: model‐based and CBD‐signature approaches to PHM. |
Figure 6.8 |
Example diagram: heuristic‐based CBM system using CBD‐based modeling. |
Figure 6.9 |
Procedural diagram for producing a DPS‐based FFS. |
Figure 6.10 |
Examples of CBD, FFP, DPS, and DPS‐based FFS. |
Figure 6.11 |
DPS‐based FFS and FFP‐based FFS. |
Figure 6.12 |
FNL plots for the FFS shown in Figure 6.11. |
Figure 6.13 |
Plots of a family of FFP signatures and DPS transfer curves. |
Figure 6.14 |
Offline phase to develop a prognostic‐enabling solution for a PHM system. |
Figure 6.15 |
Online phase to exploit a prognostic‐enabling solution. |
Figure 6.16 |
Multiple temperature signals: before and after differential‐distance conditioning. |
Figure 6.17 |
Extracted FD from fusing differential‐distance conditioned CBD. |
Figure 6.18 |
Block diagram of an example EMA subsystem. |
Figure 6.19 |
Offline modeling and development diagram. |
Figure 6.20 |
Design and analysis diagram. |
Figure 6.21 |
Continual (top) and periodic (bottom) sampling. |
Figure 6.22 |
Period‐burst sampling: sampling period (TS) and burst period (TB). |
Figure 6.23 |
Example of power supplies and EMA subsystems. |
Figure 6.24 |
SMPS output showing ripple period (TR) and sampling period (TS). |
Figure 6.25 |
Damped‐ringing response caused by an abrupt load change. |
Figure 6.26 |
Example of burst of burst sampling. |
Figure 6.27 |
Example of alerts issued using an unrealistic PHM system clock and/or polling method. |
Figure 6.28 |
Steps in test data due to low‐resolution fault injection. |
Figure 6.29 |
Diagram of a test bed to inject faults into a power supply. |
Figure 6.30 |
Diagram of a test bed to inject faults into an EMA. |
Figure 6.31 |
Sampled‐ and windowed‐phase currents: no load (top) and extra load (bottom). |
Figure 6.32 |
FFP signature of fault‐injected SMPS. |
Figure 6.33 |
Sampled‐phase currents: no load (top) and extra load (bottom). |
Figure 6.34 |
Current magnitude: no degradation (high), degraded transistor (reduced amplitude). |
Figure 6.35 |
Shifted levels due to a degraded power‐switching transistor. |
Figure 6.36 |
Illustration of using peak positive and negative threshold values. |
Figure 6.37 |
Special rms: threshold and truncation. |
Figure 6.38 |
Special rms applied to three phase currents. |
Figure 6.39 |
FFP signatures due to loading. |
Figure 6.40 |
Smoothed FFP signatures due to loading. |
Figure 6.41 |
FFS: EMA loading. |
Figure 6.42 |
Smoothed FFP signatures due to winding faults. |
Figure 6.43 |
FFS: EMA winding. |
Figure 6.44 |
Phase A currents: transistor fault in the positive Phase A portion. |
Figure 6.45 |
Phase B currents: transistor fault in the positive Phase A portion. |
Figure 6.46 |
Phase C currents: transistor fault in the positive Phase A portion. |
Figure 6.47 |
Peak‐RMS currents: transistor fault in the positive Phase A portion. |
Figure 6.48 |
Close‐up of the EMA motor winding. |
Figure 6.49 |
FFP: H bridge fault, sum of both halves of the Phase A current. |
Figure 6.50 |
Smoothed FFP: H bridge fault. |
Figure 6.51 |
Smoothed FFS: H bridge fault. |
Figure 6.52 |
Block diagram for an example of a robust PHM system. |
Figure 6.53 |
Architectural block diagram for a node definition. |
Figure 6.54 |
Example of a system node definition. |
Figure 6.55 |
Block diagram of an example node definition. |
Figure 6.56 |
NDEF: node status. |
Figure 6.57 |
NDEF: sampling specifications. |
Figure 6.58 |
NDEF: alert specifications. |
Figure 6.59 |
NDEF: Special Files specifications. |
Figure 6.60 |
NDEF: feature‐vector framework – (a) primary, (b) smoothing, (c) FFP‐DPS transform. |
Figure 6.61 |
NDEF: Prediction Framework. |
Figure 6.62 |
NDEF: Performance Services/Graphics. |
Figure 6.63 |
NDEF: Input & Output Files. |
Figure 6.64 |
NDEF: Checkpoint Library & File Name. |
Figure 6.65 |
NDEF: Device Driver Program ID and Units of Measure. |
Figure 6.66 |
NDEF: Other Program IDs. |
Figure 6.67 |
NDEF: End of Definition. |
Figure 6.68 |
NDEF updates to support node second SMPS (node 60). |
Figure 6.69 |
NDEF updates to support EMA 1 (node 51). |
Figure 6.70 |
Illustration and relationship of PH to BD, sample time, RUL, and EOL. |
Figure 6.71 |
Relationship of degradation times and an FFS. |
Figure 6.72 |
Illustration of the uncertainty of determining exactly when functional failure occurs. |
Figure 6.73 |
Uncertainty: prognostic distance. |
Figure 6.74 |
CBD at 1‐hour sampling (top) and at 24‐hour sampling (bottom). |
Figure 6.75 |
Family of failure curves, failure distribution, and TTF. |
Figure 6.76 |
Diagram of an initial estimate for PD. |
Figure 6.77 |
Comparison: FFS and plots of RUL and PH estimates. |
Figure 6.78 |
Prognostic bus (log file). |
Figure 6.79 |
SoH plot from DXARULE. |
Figure 6.80 |
Output file for SMPS using ARULEAV. |
Figure 6.81 |
Plots of the RUL, PH, and SoH estimates produced by ARULEAV. |
Figure 6.82 |
Output file for SMPS using DPS‐based FFS and ARULEAV. |
Figure 6.83 |
Plots of the RUL, PH, and SoH estimates using DPS‐based FFS and ARULEAV. |
Figure 6.84 |
CBD, FFP, and FFS for EMA node 51 (friction/load). |
Figure 6.85 |
EMA (load) plots: RUL, PH, and SoH estimates using DPS‐based FFS and ARULEAV. |
Figure 6.86 |
CBD, FFP, and FFS for EMA node 61 (winding). |
Figure 6.87 |
EMA (winding) plots: RUL, PH, and SoH estimates; DPS‐based FFS and ARULEAV. |
Figure 6.88 |
CBD, FFP, and FFS for EMA node 62 (power transistor). |
Figure 6.89 |
EMA (transistor) plots: RUL, PH, and SoH estimates; DPS‐based FFS and ARULEAV. |
Figure 6.90 |
PHM: high‐level control and data flow. |
Figure 6.91 |
Initialization: system nodes. |
Figure 6.92 |
System alerts, part 1. |
Figure 6.93 |
System alerts, part 2. |
Figure 6.94 |
Example of a GUI for a PHM system. Source: Ridgetop (2018). |
Figure 7.1 |
Example of a broad view of an ecosystem. |
Figure 7.2 |
Example of a five‐level model for health solutions. |
Figure 7.3 |
Example of alerts for SoH at or below 25%. |
Figure 7.4 |
Example of alerts for a damage‐detection approach. |
Figure 7.5 |
MTTF, TTF, and PITTFF0: CBD signature and failure distribution. |
Figure 7.6 |
Same MTTF for different failure distributions and signatures. |
Figure 7.7 |
Failure plots with average values for TTF and PD = PITTFF0. |
Figure 7.8 |
Estimated PD and actual PD: power supply (top) and EMA load (bottom). |
Figure 7.9 |
Estimated PD and actual PD: EMA winding (top) and EMA transistor (bottom). |
Figure 7.10 |
RUL and PH plot for PITTFF0 = 4800 and for PITTFF0 = 2290. |
Figure 7.11 |
RUL and PH plots for SMPS: before and after PITTFADJ = 2.0. |
Figure 7.12 |
RUL and PH plots for EMA 51: before and after PITTFADJ = 2.0. |
Figure 7.13 |
RUL and PH plots for EMA 61: before and after PITTFADJ = 2.0. |
Figure 7.14 |
RUL and PH plots for EMA 62: before and after PITTFADJ = 2.0. |
Figure 7.15 |
Plots: test results for six power supplies (top) and 12 EMAs (bottom). |
Figure 7.16 |
Bathtub curve showing three regions, MTBF, and a prognostic trigger point. |
Figure 7.17 |
Possible relationship of bathtub curve to failure distribution and MTTF. |
Figure 7.18 |
Multiple instances of CBD signatures and trigger points. |