Prognostic Health Management Techniques for Power Electronics
Presented By: Mark Scott
Date: 10/15/2018
Where are Power Electronics Found?
Consumer Electronics
Tesla Model S
Aircraft and High Speed Rail
http://www.caranddriver.com/tesla/model-s
http://www.bombardier.com/en/about-us/history.html
Google Server Room
ABB SVC: Norway
SpaceX Dragon V2
https://planetsurprises.wordpress.com/201 2/10/18/google-server-room/
http://www.abb.com/cawp/seitp202/c36f4e6 2da52ab46c1257670003690d3.aspx
http://www.electronicsweekly.com/u ncategorised/space-spacex-dragonv2-ready-crew-2014-05/
Mark Scott, Ph.D.
scottmj3@miamioh.edu
2
More Applications
250 MW PV Plant (Portugal)
Fowler Ridge Wind Farm Steve Grundy: https://flic.kr/p/aDy58Y
http://www.sfgate.com/business/article/PG-E-plansbig-investment-in-solar-power-3199510.php
Portable Oxygen Concentrator ITER Tokamak
https://www.spinlife.com/
https://www.iter.org/sci/makingitwork
Mark Scott, Ph.D.
scottmj3@miamioh.edu
3
Introduction I.
Motivation for Prognostic Health Management Systems in Power Electronics
II. Failure Mechanisms in Power Electronics Capacitors Semiconductor Switching Devices III. Approaches to PHM in Power Electronics IV. Electromagnetic Spectral based PHM Approach (E-PHM)
Theory Results
V. Conclusion
Mark Scott, Ph.D.
scottmj3@miamioh.edu
4
What is PHM? Prognostic Health Management (PHM) Systems: • •
Monitor the health of individual components in a system to ensure they remain in the safe operating area. Estimates the remaining useful life of a component or system.
Goal: • •
Minimize unexpected failures from occurring. Extract the maximum amount of useful life from a system.
Mark Scott, Ph.D.
scottmj3@miamioh.edu
5
What is PHM? Prognostic Health Management (PHM) Systems Goals: • •
Minimize unexpected failures from occurring. Extract the maximum amount of useful life from a system.
Mark Scott, Ph.D.
scottmj3@miamioh.edu
6
PHM: Motivation
1. 2. 3. 4.
•
Maintenance accounts for 10 % to 20 % of the total operating costs for an aircraft in the commercial airline industry ($3.6 M per aircraft per year) [1].
•
In aviation, up to 40 % of equipment is falsely diagnosed and removed as the result of ambiguous and labor intensive test procedures [1], [2].
•
It is recommended to change the DC Link capacitors in uninterruptable power supplies (UPS) after 50,000 hours even though their lifetimes is estimated at 150,000 [3].
•
Semiconductor switching devices and the capacitors are the most likely elements to fail in power electronics [4].
•
These failures primarily occur due to extreme stress and aging. Their damage indicators are often the same [4]. International Air Transport Association, Airline Maintenance Cost Executive Commentary. Montreal CA: Maintenance Cost Task Force, 2015 U. PeriyarSelvam et al., “Analysis on Costs for Aircraft Maintenance,” Advances Aerospace Science Appl. vol. 3, no. 3, pp. 177-182, 2013. “Capacitors age and capacitors have an end of life - white paper,” in Emerson Network Power, 2015. Y. Avenas et al., “Condition Monitoring: A Decade of Proposed Techniques,” IEEE Ind. Electron. Mag., vol. 9, No. 4, pp. 22-36, Dec. 2015.
Mark Scott, Ph.D.
scottmj3@miamioh.edu
7
Capacitors Parameters and Aging Capacitor Types: • Aluminum Electrolytic • Metalized Poly Propylene Film • Multilayer Ceramic Capacitors Parameters: • Capacitance (CDC) • Equivalent series resistance (RESR) • Equivalent series inductances (LESL)
Equivalent circuit of a non-ideal capacitor.
Capacitance and equivalent series resistance variation with degradation [1]. 1.
H. Solimann et al., “A Review of Condition Monitoring of Capacitors in Power Electronics, ” in IEEE Trans. Ind. Appl., vol. 52, no. 6, pp. 4976-4987, Nov./Dec. 2016.
Mark Scott, Ph.D.
scottmj3@miamioh.edu
8
Capacitors Failure Modes Capacitor Type Aluminum Electrolytic
Failure Mechanism Electrolyte Vaporization, Electrochemical Reaction
Metalized Poly Propylene Film Multilayer Ceramic
Critical Stressor Temp, Voltage, Current
Failure Mode Open Circuit
Moisture corrosion, Temp, Voltage, Humidity dielectric loss
Open Circuit
Insulation degradation, Temp, Voltage, Vibration flex cracking
Short Circuit
Condition Monitoring: Implementation: 1. Offline 2. Online
Lifetime Indicators: 1. Capacitance (CDC) 2. RESR 3. Ripple Voltage (ΔVDC) 4. Volume 5. Temperature Mark Scott, Ph.D.
Methods: 1. Current sensors 2. Injecting signals 3. Adv. data algorithms
scottmj3@miamioh.edu
9
PHM: Failure Mechanisms in Power Electronics Power devices: Main wear-out mechanisms for semiconductor devices [1]: • Wire bond lift-off. • Baseplate solder joint cracking. • Chip solder joint cracking. Block diagram of a power module [1]. Failure Mode Wire Bond Lift-Off Solder Joint Degradation Gate Oxide Degradation Junction Temperature
Damage Indicator
Increased forward voltage (VCE or VDS) Reduced parasitic ringing Increase in thermal resistance (RTH) Increased forward voltage (VCE or VDS) Changes in threshold voltage (VTH) Changes in on-resistance (RDS_ON) and forward voltage (VCE) Changes in switching speed (dV/dt)
[1] H. Wang et al., “Transitioning to Physics-of-Failure as a Reliability Driver in Power Electronics,” IEEE J. Emerging & Selected Topics in Power Electron., vol. 2, no. 1, pp. 97-114 , Mar. 2014.
Mark Scott, Ph.D.
scottmj3@miamioh.edu
10
E-PHM: Overview Approach: The EM spectrum is analyzed to preform PHM of the systems. Implementation: An embedded system measures spectral content of current transducers (CT) and provides PHM feedback about power electronics. Adv.:
(1) Enables In-situ measurements. (2) Integrates into existing control hardware. (3) Minimal additional components required (Primarily software based).
Mark Scott, Ph.D.
scottmj3@miamioh.edu
11
E-PHM: Basic Theory Overview: Spectral content of power electronics hardware provides a unique signature of the operating conditions and system hardware. •
Operating Conditions: dV/dt of switching devices changes with temperature and load current. → Impacts EMI.
•
System Hardware: As passive and active components age, their associated parasitics change. → Impacts EMI.
Mark Scott, Ph.D.
scottmj3@miamioh.edu
12
E-PHM: Theoretical Analysis Theory: Using the equivalent differential mode (DM) circuit, the magnitude of the DM voltage and capacitor’s resonant frequency are impacted by aging.
DM Voltage vs. Frequency
Equivalent DM Circuit Theoretical Results • Voltage at the resonant frequency is 10 dB higher for aged capacitors. • Resonant frequency also shifts higher from 120.3 kHz to 126.8 kHz as the components age.
Mark Scott, Ph.D.
Capacitor Parameters New
Aged
CDC
50 µF
40 µF
RESR
6 mΩ
12 mΩ
LESL
35 nH
35 nH
scottmj3@miamioh.edu
13
E-PHM: Experimental Validation Approach: • Preliminary testing is performed on an EMI test bench. • Modular three phase inverter that enables DC Link capacitors to be exchanged. • The spectrum is measured for multiple classes of DC link capacitors. • EMI spectrum is processed offline using a support vector machine.
Modular Three Phase Inverter.
EMI Test Setup at Miami University Mark Scott, Ph.D.
scottmj3@miamioh.edu
14
E-PHM: Experimental Validation Cont’d Execution: • Capacitor interface board enables C DC and RESR to be altered. • Five classes of DC link capacitors are evaluated: – N6A0, N5A1, N3A3, N1A5, N0A6.
•
50 measurements were taken for each class: – 25 were used to train the SVM. – 25 were used to evaluate the SVM.
Individual Capacitor Parameters
RESR
CDC
New
6.0 μΩ
50 μF
Aged
23.6 μΩ
45 μF
Modular Three Phase Inverter
Capacitor Interface Board Mark Scott, Ph.D.
scottmj3@miamioh.edu
15
E-PHM Experimental Results Total Noise Measurements in 10 kHz to 200 kHz Range: • Difficult to notice differences between the spectrums.
Test Conditions New VDC
540 V
POUT
5 kW
fS
20 kHz
fc
60 Hz
Mark Scott, Ph.D.
scottmj3@miamioh.edu
16
E-PHM Experimental Results Total Noise Measurements in 10 kHz to 200 kHz Range: • A close examination at one of the switching frequency’s harmonics shows its magnitude increases with capacitor age.
Test Conditions New VDC
540 V
POUT
5 kW
fS
20 kHz
fc
60 Hz
Mark Scott, Ph.D.
scottmj3@miamioh.edu
17
E-PHM Experimental Results Machine Learning Algorithm Results: • 250 Tests conducted with 5 capacitor classes in the DC Link: – 6 New (N6A0), 5 New &1 Aged (N5A1), 3 New & 3 Aged (N3A3), 1 New and 5 Aged (N1A5), 6 Aged (N0A6).
•
SVM succeeds at identifying the brand new and aged DC link capacitance.
Test Conditions
Test Results – 10 kHz to 200 kHz Total Noise
SVM Decision
VDC
540 V
POUT
5 kW
fS
20 kHz
fc
60 Hz
Actual
New
N6A0 N5A1 N3A3 N1A5 N0A6
N6A0 25 0 0 0 0
N5A1 0 19 0 0 0
Mark Scott, Ph.D.
N3A3 0 5 20 5 0
N1A5 0 1 5 20 0
N0A6 0 0 0 0 25
scottmj3@miamioh.edu
18
Target Distributions Population Construction: • Sample size of 50 trials. • Mean and standard deviation from the mean were recorded. • Samples were modeled as a normally distributed population. • Populations overlap without a clear boundary.
Target Distributions at 6th Harmonic New N5A1 N3A3 N1A5 Aged
→ Shows that application is suited for machine learning.
Mark Scott, Ph.D.
scottmj3@miamioh.edu
19
E-PHM: Experimental Results Additional Results: 1. Significant differences in DM noise, but CM noise is more or less the same. 2. Magnitude of noise is higher for aged (‘old’) capacitors. 3. Machine learning performs well at low frequencies; struggles at higher ones. Validates theoretical analysis.
Experimental Results Common Mode Noise Mark Scott, Ph.D.
Experimental Results Differential Mode Noise scottmj3@miamioh.edu
20
Future Work Power Devices: •
Changes in spectrum with: – Device age. – Temperature.
Equivalent CM Circuit dV/dt for Turn On vs. Current
• •
14
•
Improving noise immunity. Replace LISN with current sensors. Implement SVM on an embedded system.
Switching Speed (ns)
Measurements:
16
12 10
8 6 4
Rohm (25 C) USCi (100 C) IXYS (25 C)
2
Rohm (100 C) Wolfspeed (25 C) IXYS (100 C)
USCi (25 C) WolfSpeed (100)
0
0
5
10
15
20
25
Drain Current (A) Mark Scott, Ph.D.
scottmj3@miamioh.edu
21
Conclusions Concept of Prognostic Health Management (PHM) Systems were introduced. Capacitor failure mechanisms were reviewed. Wear out conditions indicated by decrease in CDC and increase in RESR. Conditional monitor approaches for power devices were presented. Electromagnetic Spectrum based PHM (E-PHM) method was discussed. The approach successfully identifies changes in the health of the capacitors. Future work for E-PWM includes developing models for power devices.
Mark Scott, Ph.D.
scottmj3@miamioh.edu
22
Thank You for your Attention. Any Questions?
Mark Scott, Ph.D.
scottmj3@miamioh.edu
23