Characterization method of IGBT comprehensive health index based on online status data
Introduction
IGBT and VDMOS, as core devices in the field of power electronics, are widely used in many fields such as modern energy, manufacturing, and communications. With high voltage, high current and other harsh working conditions, these devices are subject to a lot of stress factors such as heat stress, mechanical stress and electrical stress, and prone to failure. Their failure may cause system loss, sudden shutdown, long maintenance time, and high cost. Therefore, the stable operation of the devices must be ensured. VDMOS causes extensive investigation on its reliability due to its excellent performance, such as HTRB (high temperature reverse bias), HTGB (high temperature gate bias), NBTI (negative bias temperature instability) and/or irradiation, so on [[1], [2], [3]]. Due to its mature application, the attention of IGBT device is paid to the research of online status evaluation and life prediction. These power electronic devices are very important and have a variety of degradation model, so how to characterize the health of online devices is an important issue [4]. The research results of IGBT can also be used for VDMOS.
The health characterization and monitoring methods of IGBT can be divided into three categories: (1) Failure physical method [5], which needs to explore the complex failure mechanism of the system to establish a mathematical model. (2) Data-driven method [6], which avoids the shortcomings of failure physical method, and has the advantage of real-time and good adaptability [7,8]. However, this method requires the development of a predictive model based on sensor data (3) Fusion method. Combining the advantages of the failure physical and the data-driven method, and overcoming the limitations of them, this method can provide more accurate and convenient results.
There are many degradation mechanisms in IGBT, which contains package-level degradation such as bond wire shedding and solder thermal fatigue, and chip-level degradation such as increased body resistance and gate oxide aging. Moinul et al. [9] propose an auxiliary particle filter (APF) method using Vce-on data to estimate remaining useful life (RUL). Mominul Ahsan et al. [10] propose an method based on Neural Network (NN) and Adaptive Neuro Fuzzy Inference System (ANFIS) models using collector-emitter voltage (Vce) information to predict RUL. Richard Mandeya et al. [11] propose an method using gate–emitter pre-threshold voltage Vge(pre-th) to monitor the failure of IGBT.
These methods have a problem that they don't consider the diversity of device degradation mechanisms, and the constructed health index (HI) is only suitable for evaluating a certain type of degradation. In other words, a single parameter is not enough to characterize the overall health level of the device. For example, using the parameter Vce-on cannot characterize the health level of the device in the gate oxide aging. In the application of the IGBT, the characterization of the health level should be comprehensive and various degradation mechanisms should be considered. Device reliability degradation mechanisms are diverse, occur simultaneously and influence each other, and they will cause changes in multiple parameters. Using changes in multiple characteristic parameters to reflect the operating status of the device will get a more accurate characterization result. However, how to use the changes of multiple parameters to obtain a comprehensive characterization index to quantify and intuitively reflect the health level of the device is a difficult problem. Prasanna Tamilselvan [12] proposes a multi-attribute fusion method based on weighted majority voting, but this method requires a lot of prior data and the model is complex. There is no doubt that a fast and convenient health index characterization method that comprehensively considers multiple degradation mechanisms of the device is very valuable.
This paper proposes an IGBT comprehensive health index characterization method based on online state data of aging life. By introducing the failure physical model of parameter degradation and combining with the data-driven method, the new method can characterize the comprehensive health level of the device under multiple degradation mechanisms. By eliminating the overlapping and invalid information in multiple characteristic parameters, amplifying and fusing the effective information, the comprehensive health index (CHI) is finally obtained. The CHI method solves the problem that a single characteristic parameter cannot characterize the comprehensive health level of the device, and that multiple characteristic parameters cannot be quantitatively represented and intuitively reflect the comprehensive health level of the device. Both the failure physical and the data-driven method are used, so that the method can not only use the results of the existing physical analysis, but also has the advantages of good adaptability. The algorithm is verified by the IGBT online state data of aging life provided by NASA.
The structure of the paper are as follows: Section 2, the main principles and steps of the method are described; Section 3, IGBT online state data are used to verify the algorithm; Section 4, the advantages and disadvantages are discussed. Finally, the paper is summarized.
Section snippets
Principles and methods
The initial application of the health index was human health evaluation, which was to predict people's life span. Comprehensive evaluation was carried out through comprehensive analysis of evaluation items related to life span, so it had been widely used in the field of life and medicine. The operating state of the device has some similarities with the human life activity state, but it is simpler than the human health state assessment. The health index was a quantitative index calculated based
Results and analysis
The online state data used in this paper were provided by Prognostics CoE of NASA Ames [22]. The data set contained online state aging data from 6 devices, one with a DC gate bias and the remaining with a square signal gate bias. The step of the experiment was to apply a square wave signal to the gate, and in order to ensure that the device operates in a stable high temperature environment, kept the case temperature at (268 °C, 270 °C) temperature range until the IGBT had a latch-off effect.
Discussion
In this research, we proved that by extracting and fusing effective information of features, the CHI characterization method can characterize the comprehensive health level of IGBT. Moreover, the ideas and steps of this method have very important reference value. Firstly, the physical model was used for feature selection based on previous work, which ensured the effectiveness of features. Secondly, deep learning algorithms such as KPCA was used to process the data, extract valid information and
Conclusion
Obtaining more information is of great help to make more accurate judgments. There are many degradation mechanisms and degradation parameters for IGBT, but they contain some overlapping and invalid information. The core of the health index characterization method is that different parameters contain different degradation information, and by eliminating overlapping and invalid information, and extracting and fusing the effective information, the more accurate and stable results can be obtained.
CRediT authorship contribution statement
Zhang Jinli: Conceptualization; Data curation; Formal analysis; Investigation; Methodology; Software; Validation; Visualization; Roles/Writing – original draft; Writing – review & editing.
Hu Jinbao: Conceptualization; Data curation; Investigation; Methodology; Software; Validation; Visualization.
You Hailong: Funding acquisition; Data curation; Formal analysis; Supervision; Resources; Project administration; Writing – review & editing.
Jia Renxu: Funding acquisition; Formal analysis; Resources.
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgment
This work is supported by the Opening Project of Science and Technology on Reliability Physics and Application Technology of Electronic Component Laboratory (6142806190205).
References (22)
- et al.
NBTI and irradiation related degradation mechanisms in power VDMOS transistors[J]
Microelectron. Reliab.
(2018) - et al.
The threshold voltage degradation model of N Channel VDMOSFETs under PBT stress[J]
Microelectron. Reliab.
(2018) - et al.
A multi-attribute classification fusion system for insulated gate bipolar transistor diagnostics[J]
Microelectron. Reliab.
(2013) - et al.
A comparative analytical approach for gate leakage current optimization in silicon MOSFET: a step to more reliable electronic device[C]//2016 3rd International Conference on Electrical Engineering and Information Communication Technology (ICEEICT)
IEEE
(2016) - et al.
Low magnetic field impact on NBTI degradation[J]
Microelectron. Reliab.
(2015) - et al.
Analysis of the results of accelerated aging tests in insulated gate bipolar transistors[J]
IEEE Trans. Power Electron.
(2016) - et al.
Thermal and structural simulation techniques for estimating fatigue life of an IGBT module[J]
Proceedings of the 20th International Symposium on Power Semiconductor Devices & IC's
(2008) - et al.
Stochastic RUL calculation enhanced with TDNN-based IGBT failure modeling[J]
IEEE Transactions on Reliability R
(2016) - et al.
Physics-of-failure, condition monitoring, prognostics of insulated gate bipolar transistor modules: a review[J]
IEEE Trans. Power Electron.
(2015) - et al.
A review of data-driven prognostics in power electronics[J]
IEEE 35th International Spring Seminar on Electronics Technology (ISSE)
(2012)