Remaining Useful Life Prediction of IGBT Modules Across Working Conditions Based on ProbSparse Self-Attention

doi:10.16183/j.cnki.jsjtu.2021.538

Abstract

Abstract:

In order to improve the accuracy of remaining useful life (RUL) prediction of insulated gate bipolar transistor(IGBT) modules across working conditions to enhance their reliability, an RUL prediction method based on the ProbSparse self-attention mechanism and transfer learning was proposed based on the transient thermal resistance features of IGBT modules under different working conditions. An accelerated aging test bench of the IGBT module was designed ang built to perform power cycling experiments in different temperature ranges, and state data of full life-time under different working conditions were collected. Transient thermal resistance change data during the IGBT module degradation were calculated, and the transient thermal features that can accurately reflect the aging state of the IGBT module were extracted and selected. These features were used to predict the RUL of IGBT modules across different working conditions based on the proposed method. The experimental result shows that the accuracy of the proposed RUL prediction method of IGBT modules across working conditions outperforms other compared methods. Particularly, the RUL prediction accuracy during the early degradation stage has been significantly improved.

Key words: insulated gate bipolar transistor(IGBT) module, transient thermal resistance, remaining useful life (RUL) prediction, ProbSparse self-attention, transfer learning

CLC Number:

ZHONG Zhiwei, WANG Yuxiang, HUANG Yixiang, XIAO Dengyu, XIA Pengcheng, LIU Chengliang. Remaining Useful Life Prediction of IGBT Modules Across Working Conditions Based on ProbSparse Self-Attention[J]. Journal of Shanghai Jiao Tong University, 2023, 57(8): 1005-1015.

Figures/Tables 17

Fig.1

Fig.2

Fig.3

Fig.4

Fig.5

Tab.1

Fig.6

Fig.7

Tab.2

Fig.8

Tab.3

Fig.9

Tab.4

Fig.10

Fig.11

Fig.12

Tab.5

References 28

[1]	CARAMEL C, AUSTIN P, SANCHEZ J L, et al. Integrated IGBT short-circuit protection structure: Design and optimization[J]. Microelectronics Journal, 2006, 37(3): 249-256. doi: 10.1016/j.mejo.2005.09.028 URL
[2]	刘嘉诚. 基于机器学习算法的IGBT寿命预测研究[D]. 合肥: 合肥工业大学, 2020.
	LIU Jiacheng. IGBT life prediction based on machine learning algorithm[D]. Hefei: Hefei University of Technology, 2020.
[3]	YANG S Y, BRYANT A, MAWBY P, et al. An industry-based survey of reliability in power electronic converters[J]. IEEE Transactions on Industry Applications, 2011, 47(3): 1441-1451. doi: 10.1109/TIA.2011.2124436 URL
[4]	GOPIREDDY L R, TOLBERT L M, OZPINECI B. Power cycle testing of power switches: A literature survey[J]. IEEE Transactions on Power Electronics, 2015, 30(5): 2465-2473.
[5]	唐圣学, 张继欣, 姚芳, 等. IGBT模块寿命预测方法研究综述[J]. 电源学报, 2021: 1-29.
	TANG Shengxue, ZHANG Jixin, YAO Fang, et al. An overview of lifetime prediction methods for IGBT power module[J]. Journal of Power Supply, 2021: 1-29.
[6]	HUANG Q, PENG C, WANG L C, et al. Effects of current and bonding wires damage on high-power IGBT module reliability by electro-thermo-mechanical simulation[C]//2020 21st International Conference on Electronic Packaging Technology. Guangzhou, China: IEEE, 2020: 1-4.
[7]	张亚玲, 李志刚. 基于加速老化试验的IGBT寿命预测模型研究[J]. 电气传动, 2016, 46(10): 72-75.
	ZHANG Yaling, LI Zhigang. Research on IGBT lifetime prediction models based on accelerated lifetime test[J]. Electric Drive, 2016, 46(10): 72-75.
[8]	田航, 陈民武, 张振宇, 等. 牵引负荷对补偿装置功率模块寿命预测的影响及分析[J]. 电气化铁道, 2019, 30(1): 24-28.
	TIAN Hang, CHEN Minwu, ZHANG Zhenyu, et al. Influence to prediction of service life of power modules of compensation device caused by traction load and its analysis[J]. Electric Railway, 2019, 30(1): 24-28.
[9]	刘子英, 朱琛磊. 基于Elman神经网络模型的IGBT寿命预测[J]. 半导体技术, 2019, 44(5): 395-400.
	LIU Ziying, ZHU Chenlei. IGBT life prediction based on Elman neural network model[J]. Semiconductor Technology, 2019, 44(5): 395-400.
[10]	葛建文, 黄亦翔, 陶智宇, 等. 基于Transformer模型的IGBT剩余寿命预测[J]. 半导体技术, 2021, 46(4): 316-323.
	GE Jianwen, HUANG Yixiang, TAO Zhiyu, et al. Residual useful life prediction of IGBTs based on transformer model[J]. Semiconductor Technology, 2021, 46(4): 316-323.
[11]	VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[EB/OL]. (2017-12-06)[2021-12-30]. https//arxiv.org/abs/1706.03762.
[12]	CAO Y D, DING Y F, JIA M P, et al. A novel temporal convolutional network with residual self-attention mechanism for remaining useful life prediction of rolling bearings[J]. Reliability Engineering & System Safety, 2021, 215: 107813. doi: 10.1016/j.ress.2021.107813 URL
[13]	MO Y, WU Q H, LI X, et al. Remaining useful life estimation via transformer encoder enhanced by a gated convolutional unit[J]. Journal of Intelligent Manufacturing, 2021, 32(7): 1997-2006. doi: 10.1007/s10845-021-01750-x
[14]	ZHOU H Y, ZHANG S H, PENG J Q, et al. Informer: Beyond efficient transformer for long sequence time-series forecasting[EB/OL]. (2021-03-28)[2021-12-30]. https//arxiv.org/abs/2012.07436.
[15]	GEHRING J, AULI M, GRANGIER D, et al. Convolutional sequence to sequence learning[C]//34th International Conference on Machine Learning. Sydney, Australia: PMLR, 2017: 1243-1252.
[16]	LONG M S, CAO Y, CAO Z J, et al. Transferable representation learning with deep adaptation networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019, 41(12): 3071-3085. doi: 10.1109/TPAMI.2018.2868685 pmid: 30188813
[17]	曾东, 孙林, 周雒维, 等. 基于加速老化试验IGBT性能退化特征参量的可靠性评估[J]. 电工电能新技术, 2019, 38(7): 20-28.
	ZENG Dong, SUN Lin, ZHOU Luowei, et al. Reliability evaluation of IGBT performance degradation characteristic parameters based on accelerated aging test[J]. Advanced Technology of Electrical Engineering and Energy, 2019, 38(7): 20-28.
[18]	刘宾礼, 刘德志, 唐勇, 等. 基于加速寿命试验的IGBT模块寿命预测和失效分析[J]. 江苏大学学报(自然科学版), 2013, 34(5): 556-563.
	LIU Binli, LIU Dezhi, TANG Yong, et al. Lifetime prediction and failure analysis of IGBT module based on accelerated lifetime test[J]. Journal of Jiangsu University (Natural Science Edition), 2013, 34(5): 556-563.
[19]	张全. 电子元器件加速寿命试验技术浅析[J]. 空间电子技术, 2016, 13(6): 86-88.
	ZHANG Quan. Analysis on the accelerated life test technique of electronic components[J]. Space Electronic Technology, 2016, 13(6): 86-88.
[20]	姬煜轲, 侯婷, 何智鹏, 等. 一种柔直换流阀用压接型IGBT功率子模块加速老化试验方法[J]. 南方电网技术, 2021, 15(5): 1-11.
	JI Yuke, HOU Ting, HE Zhipeng, et al. An accelerated aging test method of press-pack IGBTs based power submodules for VSC-HVDC converter valve[J]. Southern Power System Technology, 2021, 15(5): 1-11.
[21]	SMET V, FOREST F, HUSELSTEIN J J, et al. Ageing and failure modes of IGBT modules in high-temperature power cycling[J]. IEEE Transactions on Industrial Electronics, 2011, 58(10): 4931-4941. doi: 10.1109/TIE.2011.2114313 URL
[22]	陈一高, 陈民铀, 高兵, 等. 基于瞬态热阻的IGBT焊料层失效分析[J]. 中国电机工程学报, 2018, 38(10): 3059-3067.
	CHEN Yigao, CHEN Minyou, GAO Bing, et al. Evaluation of solder failure of an IGBT module based on transient thermal impedance[J]. Proceedings of the CSEE, 2018, 38(10): 3059-3067.
[23]	MOROZUMI A, YAMADA K, MIYASAKA T, et al. Reliability of power cycling for IGBT power semiconductor modules[J]. IEEE Transactions on Industry Applications, 2003, 39(3): 665-671. doi: 10.1109/TIA.2003.810661 URL
[24]	卓鹏程, 严瑾, 郑美妹, 等. 面向滚动轴承全生命周期故障诊断的GA-OIHF Elman神经网络算法[J]. 上海交通大学学报, 2021, 55(10): 1255-1262.
	ZHUO Pengcheng, YAN Jin, ZHENG Meimei, et al. GA-OIHF Elman neural network algorithm for fault diagnosis of full life cycle of rolling bearing[J]. Journal of Shanghai Jiao Tong University, 2021, 55(10): 1255-1262.
[25]	WANG Y, PENG Y Z, ZI Y Y, et al. A two-stage data-driven-based prognostic approach for bearing degradation problem[J]. IEEE Transactions on Industrial Informatics, 2016, 12(3): 924-932. doi: 10.1109/TII.2016.2535368 URL
[26]	LIU H, DAI Z, SO D R, et al. Pay attention to mlps[J]. Advances in Neural Information Processing Systems, 2021, 34: 9204-9215.
[27]	PAN S J, TSANG I W, KWOK J T, et al. Domain adaptation via transfer component analysis[J]. IEEE Transactions on Neural Networks, 2011, 22(2): 199-210. doi: 10.1109/TNN.2010.2091281 pmid: 21095864
[28]	SUN B C, SAENKO K. Deep coral: Correlation alignment for deep domain adaptation[C/OL].(2016-11-24)[2021-12-30]. https//link.springer.com/chapter/10.1007/978-3-319-49409-8_35.

模块名	ΔT_j/ ℃	T_jm/ ℃	T_{j_min}/ ℃	T_{j_max}/ ℃	N/次
IGBT1	80	80	40	120	148 239
IGBT2	80	90	50	130	104 298
IGBT3	80	130	90	170	68 115

特征	编号	特征	编号
最小值	X₁	绝对中位差	X₁₂
最大值	X₂	峭度	X₁₃
极差	X₃	均方值	X₁₄
均值	X₄	均方根	X₁₅
中位数	X₅	绝对均值	X₁₆
第一四分位数	X₆	方根幅值	X₁₇
第三四分位数	X₇	脉冲因子	X₁₈
方差	X₈	波形因子	X₁₉
标准差	X₉	裕度因子	X₂₀
偏度	X₁₀	峰值因子	X₂₁
标准误差	X₁₁	双对数比率	X₂₂

网络模块	具体实现	参数	输出尺寸
输入层	-	-	50×12
嵌入层	一维卷积	卷积核为12×3×256	50×256
概率稀疏自注意力	线性变换	权重为50×256×32	50×32
模块	注意力计算	-	50×32
(n=8, d_k=32)	合并	-	50×256
后处理模块	层正则化	-	50×256
	一维卷积	卷积核为256×1×128	50×128
	一维卷积	卷积核为128×1×256	50×256
	层正则化	-	50×256
	一维卷积	卷积核为256×5×64	46×64
	全局池化层	-	1×64
全连接模块	全连接层	权重为64×32	32
	全连接层	权重为32×1	1

方法	IGBT1→IGBT2		IGBT1→IGBT3		IGBT2→IGBT1		IGBT2→IGBT3		IGBT3→IGBT1		IGBT3→IGBT2
方法	MSE	MAE	MSE	MAE	MSE	MAE	MSE	MAE	MSE	MAE	MSE	MAE
本文模型	0.001 6	0.031 9	0.001 0	0.024 3	0.002 6	0.041 9	0.001 6	0.033 0	0.005 8	0.058 5	0.001 6	0.032 5
gMLP	0.006 1	0.071 3	0.088 7	0.225 8	0.006 4	0.067 5	0.095 6	0.253 8	0.008 1	0.065 0	0.026 8	0.128 6
LSTM	0.009 3	0.074 3	0.027 3	0.129 4	0.009 2	0.081 7	0.033 1	0.155 7	0.013 5	0.084 5	0.027 6	0.129 4
TCA+gMLP	0.004 2	0.056 6	0.012 7	0.088 9	0.005 5	0.059 1	0.007 5	0.074 0	0.010 0	0.079 1	0.015 5	0.103 2
CORAL+LSTM	0.007 2	0.058 1	0.025 2	0.123 3	0.004 6	0.055 6	0.042 2	0.165 3	0.032 5	0.137 8	0.043 1	0.159 8

方法	训练时间/s	推理时间/s
传统自注意力机制	0.0167	0.0050
概率稀疏自注意力机制	0.0110	0.0031
相对提升比例	34.13%	38.00%