Prediction of IGBT junction temperature using improved cuckoo search-based extreme learning machine
Introduction
IGBT is now widely applied in power systems such as electric vehicles and high-speed traction, etc. [1]. However, the failure rate caused by power semiconductor devices and the photovoltaic inverter with IGBT accounts for about 34% and 37% of the total failures, respectively [2], [3]. Accordingly, the IGBT reliability should be accurately evaluated for IGBT-based device operation.
It is known that some crucial factors causing IGBT failures include collector-emitter saturation voltage (VCE(sat)), collector current (Ic), junction temperature (Tj), thermal resistance and case temperature (Tc), etc. [4]. Among them, the failure of power electronic systems related to the temperature rising accounts for 55% of the total failures [5], [6], [7]. Literature [8], [9] indicated that junction temperature is a key factor to seriously affect IGBT life and reliability.
At present, the direct junction temperature measurement based on optical or contact devices is used to measure the IGBT junction temperature. However, it may suffer from some disadvantages like high cost, inaccuracy or physical device damage [10]. Alternatively, there are three major indirect junction temperature measurement methods, e.g. physical model [11], [12], probability statistics [13], [14] and machine learning [15], [16]. The physical model is established on the basis of the internal physical structure, material properties and mechanical properties of devices [17]. The probability statistics method is applied to form a temperature distribution function from experimental data [18]. The machine learning models [19] are developed to find the relationship between the independent variables and junction temperature in the IGBT.
The content structure of this article is as follows: The second section reviews existing methods such as direct methods, indirect methods, and artificial intelligence methods for measuring the junction temperature. The third section obtains the experimental data of IGBT through IGBT experiment, processes the experimental data, and proposes a multi-parameter IGBT junction temperature prediction model. In the fourth section, junction temperature prediction model is established. Section 5 analyzes and discusses the junction temperature prediction results of each model. The sixth section is the conclusion of this article.
Section snippets
Literature review
The reliability of the IGBT can be evaluated via directly or indirectly measuring the junction temperature. Optical measurement method and physical contact method [20], [21] are classified as direct methods. On the other hand, physical models, probability statistics and machine learning are indirect methods, which have attracted more attention currently [22], [23], [24], [25]. For example, a physics-based transient electrothermal model proposed in literature [23] considered the
Set-up of the data collection experiment
In order to obtain the experimental data for VCE(sat), junction temperature, Ic and number of cycles of the IGBT, this study designs the accelerated aging and the single pulse tests. The steps of the data collection experiment are illustrated as follows. Firstly, perform the accelerated aging test. After every 1000 times test is completed, the IGBT module is put into the thermostat. Then, adjust the thermostat temperature by experimental setting at the range of [0 °C, 100 °C] until it reaches
The principle of ELM
ELM is an algorithm based on the single hidden layer feedforward neural network with fast learning speed and high predication accuracy, being widely used in various fields [44].
Firstly, N sets data is selected as the input sample X, where xj = [xj1, xj2, ⋯, xjn]⊤; Y is the output sample, yj = [yj1, yj2, ⋯, yjm]⊤; V is taken as the number of hidden layer nodes.
Sigmoidal function is chosen as the activation function of the hidden layer, which is shown in Eq. (1).
The relationship between the
Performance analysis
In the study, total 386 groups of data obtained from IGBT tests are used for Tj prediction. In the model, VCE(sat), Ic and the aging cycle number are taken as the input variables, while Tj is taken as the output variable. The first 270 groups are used for training process, and the last 116 groups are used for testing process. For comparison, ICS-ELM, CS-ELM, ELM and Support Vector Machine (SVM) are involved in the prediction of IGBT junction temperature. In addition, Mean Absolute Error (MAE),
Conclusions
It is well known that the junction temperature of IGBT is regarded as the most crucial factor to determine its working efficiency as well as failure situation. With increasing power electronics applications in industry, more attention has been drawn on the reliability, safety and stability in the device with IGBT. In the proposed ICS-ELM model, the IGBT junction temperature was firstly selected as the main failure precursor. Through accelerated aging and single pulse tests in IGBT, the
CRediT authorship contribution statement
Conceptualization, Boying Liu, Guolong Chen; Methodology, Boying Liu and Guolong Chen.; Software, Boying Liu., Guolong Chen and Hsiung-Cheng Lin; Formal analysis, Boying Liu and Hsiung-Cheng Lin.; Data curation, Boying Liu, Guolong Chen and Weipeng Zhang; Writing—original draft preparation, Boying Liu, Guolong Chen, Hsiung-Cheng Lin, Weipeng Zhang and JiaQi Liu.; Writing—review and editing, Hsiung-Cheng Lin, Weipeng Zhang and JiaQi Liu. All authors have read and agreed to the published version
Declaration of competing interest
The authors declare that we have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References (51)
- et al.
Junction temperature management of IGBT module in power electronic converters
Microelectron. Reliab.
(2014) - et al.
"A unified MMC reliability evaluation based on physics-of-failure and SM lifetime correlation," (in English)
Int. J. Electr. Power Energy Syst.
(Mar 2019) - et al.
Extreme learning machine optimized by whale optimization algorithm using insulated gate bipolar transistor module aging degree evaluation
Expert Syst. Appl.
(2019) - et al.
A temperature spectrum density distribution based condition evaluation method and application in IGBT
Appl. Therm. Eng.
(2016) - et al.
Extreme learning machine: theory and applications
Neurocomputing
(Dec 2006) - et al.
An adaptive dynamic short-term wind speed forecasting model using secondary decomposition and an improved regularized extreme learning machine
Energy
(2018) - et al.
Variable complexity online sequential extreme learning machine, with applications to streamflow prediction
J. Hydrol.
(Dec 2017) - et al.
Regularization incremental extreme learning machine with random reduced kernel for regression
Neurocomputing
(2018) - et al.
Sales forecasting using extreme learning machine with applications in fashion retailing
Decis. Support. Syst.
(Dec 2008) - et al.
Improved cuckoo search for reliability optimization problems
Comput. Ind. Eng.
(Jan 2013)
Optimal heat exchanger network synthesis based on improved cuckoo search via Lévy flights
Chem. Eng. Res. Des.
Short-term photovoltaic power prediction on modal reconstruction: a novel hybrid model approach
Sustain. Energy Technol. Assessments
Nonlinear compact thermal model of the IGBT dedicated to SPICE
IEEE Trans. Power Electron.
In-situ health monitoring of IGBT modules of an on-line medium-voltage inverter system using industrial internet of things
CSEE J. Power Energy Syst.
Tunnel magnetoresistance-based short-circuit and over-current protection for IGBT module
IEEE Trans. Power Electron.
Remaining useful life estimation for thermally aged power insulated gate bipolar transistors based on a modified maximum likelihood estimator
Int. Trans. Electr. Energy Syst.
An industry-based survey of reliability in power electronic converters
IEEE Trans. Ind. Appl.
Investigation into IGBT dV/dt during turn-off and its temperature dependence
IEEE Trans. Power Electron.
Failure mechanism of die-attach solder joints in IGBT modules under pulse high-current power cycling
IEEE J. Emerg. Sel. Top. Power Electron.
Validation of lifetime prediction of IGBT modules based on linear damage accumulation by means of superimposed power cycling tests
IEEE Trans. Ind. Electron.
"In situ diagnostics and prognostics of solder fatigue in IGBT modules for electric vehicle drives," (in English)
IEEE Trans. Power Electron.
"Reliability analysis of the optimized Y-source inverter with clamping circuit," (in English)
Microelectron. Reliab.
"Electro-thermal-mechanical multiphysics coupling failure analysis based on improved IGBT dynamic model," (in English)
IEEE Access
"Reliability analysis and design of MMC based on Mission profile for the components degradation," (in English)
IEEE Access
Lifetime estimation of discrete IGBT devices based on Gaussian process
IEEE Trans. Ind. Appl.
Cited by (19)
Hybrid domain-based fast transient thermal evaluation method for power semiconductor modules in EV motor drive
2022, Microelectronics ReliabilityCitation Excerpt :Offline thermal evaluation at the early design stage is essential to minimize the unnecessary/wasted design margin of power devices and assess the reliability of the converter [1,2].
An improved extreme learning machine with self-recurrent hidden layer
2022, Advanced Engineering InformaticsCitation Excerpt :In [19], Lu et al. developed a model of transformer fault diagnosis via improved empirical wavelet transform (IEWT) and salp swarm algorithm (SSA) optimize kernel extreme learning machine (KELM), which has higher accuracy and robustness. According to the research by Liu et al. [20], a cuckoo search-based extreme learning machine (ICS-ELM) has been designed, which can predict junction temperature without measured-location sensitivity, the results show that the method reaches a high accuracy and fitting degree. In [21], Wang et al. proposed an ELM combined with multiple-point-AdaBoost to predict the wind speed, which enhanced the wind speed prediction at the target location and provides a new idea for the field of wind speed prediction.
Prediction of the void formation in no-flow underfill process using machine learning-based algorithm
2022, Microelectronics ReliabilityCitation Excerpt :The design parameters in the conventional capillary underfill process have been optimized analytically [9], but to date, there is no relevant design optimization work being reported for the no-flow underfill process. Several optimizations have been documented such as do comparison between decision trees, adaptive neuro fuzzy inference system (ANFIS) and artificial neural networks to optimize the stencil printing process for the pin-in-paste technology [10], using big data analytics in machine learning framework to predict the reliability of solder joints [11], prediction of component self-alignment in soldering using machine learning-based prediction method [12] and prediction of transistor junction temperature using cuckoo search-based extreme machine learning [13]. Even for classification techniques machine learning has been utilized by Huang et al. [14] to classify the defects in through silicon (TSV) and normal TSV.
Correlation Study on Voiding in Underfill of Large Quantity Ball Grid Array Chip Using Machine Learning
2025, Journal of Electronic Packaging, Transactions of the ASMETarget adaptive extreme learning machine for transfer learning
2024, International Journal of Machine Learning and Cybernetics