IGBT剩余寿命预测(2013)

A Simple State-Based Prognostic Model for Predicting Remaining UsefulLife of IGBT Power ModuleAlireza Alghassi1, Suresh Perinpanayagam2, Ian K Jennions3IVHM CENTER, CRANFIELD UNIVERSITYIVHM center, Cranfield UniversityConway House, Medway Court, University Way, Cranfield Technology Park,MK43 0FQ eMilton Keynes, UKTel.: +44 (0)1234 75111 EXT 2870Fax: +44 (0)1234 758331E-Mail: a.alghassi@URL: /ivhm/Keywords«IGBT», «Reliability», «Prognosis», «SSBP»AbstractHealth management and reliability are fundamental aspects of the design and development cycle of power electronic products. This paper presents the prognostic evaluation of a power electronic IGBT module. To achieve this aim, a simple state-based prognostic (SSBP) method has been introduced and applied on the data which was extracted from an aged power electronic IGBT and its remaining useful life was determined.IntroductionThe trend toward More Electric Aircraft (MEA) and Electric Vehicle Systems are fostering increasing requirements for higher performance power electronic systems. More Electric Vehicles propose to use more electrical power to drive vehicle subsystems such as Traction Motors/Inverters, Auxiliary Motors/Inverters, Energy Storage, etc. Power electronic converters have recently generated great interest among researchers and industrialists working in this area[1].On the other hand, the inherent advantages of Insulated Gate Bipolar Transistor (IGBT) provide lower on-state resistance, higher breakdown voltage and thermal conductivity, and closer thermal expansion coefficients with better mechanical characteristics resulting in it as the main power electronic switch that is employed in power converters [2].Since IGBT power semiconductors is one of the most costly components in power electronic converters, it is beneficial to investigate the long-term reliability and the sizing of the semiconductors used in power converters. On the other hand, it is important to use a reliability assessment method that can take into account the actual operational and environmental conditions. So by employing a prognostic and health management system which assesses and predicts the reliability of a product in its actual application environment and in real time, it is possible to determine the reliability and the end-of-life period of power converters [3].Therefore, it can be stated that the aim of this work is to estimate the life consumption of a power converter which is employed in electric aircraft applications. Firstly, reliability prediction methods are investigated. Secondly, the simple state-based prognostic method is presented and then, the results of predicting remaining useful life of the power electronic IGBT is indicated.Reliability Prediction MethodsPrognostic and health management is a system which integrates the sensing and interpretation of relevant recorded data for assessing and predicting the reliability of a product in its actual application environment. It is needed to identify the failure modes and mechanisms that can take place in electrical components in the first step for employing a prognostic and health management (PHM) system [2, 4]. To identify the main failure mechanisms, the precursor parameters such as voltage, current, temperature, amongst others, have to be identified and monitored. The recorded data is then used in a PHM system to help predict the remaining life time. In this section, a review of the concept of prognostic and health management of systems is presented.Prognostic and Health ManagementPrognostics and diagnostics are the key players in service planning, maintenance and minimizing the down state of equipment. Diagnostics focuses on the detection, isolation and identification of failure when they occur whilst prognosis focuses on predicting failure before it occurs. This means that technical prognostics could be understood as an extending/complementary element of technical diagnosis. Prognosis can be referred to as the ability to predict how much time is left or remaining useful life (RUL) before a failure occurs given an observed machine condition variable and past operational profile. The observed condition can be attributed from physical characteristics or process performance of its failure. For instance, some condition parameters that can be used in prognostics are acoustic data, temperature, moisture, humidity, weather, voltage and current[5].Technical prognosis, which is being considered as a part of PHM, is a relatively new field of research and it is still considered as the weakest point in the condition-based maintenance processing chain. There are several applications of prognostics methods but the results and accuracy vary and are not always sufficient even if researchers claim so. Although several patents have been registered and many journal and conference papers have been published, the field of technical prognosis is still quite new and not well researched. In particular, robust real system applications are still missing[6]. Precursor Parameters of IGBTA few failure precursors have been reviewed for packing level-failures in [6] and it has been wildly reported that wire bond degradation can be monitored by the drifts in the voltages V CE(on) solder layer degradation causes an increment of thermal resistance R th [6, 8]. To date, measurements could only take place in a laboratory environment when the power module is disconnected from the power converter. Once the power module has become an integral part of the converter, measurements become unattainable. Hence, the challenges posed for in-situ monitoring was discussed and a hardware solution was presented in [7]. In fact, a system has been designed and set up to be capable performing robust experiments on IGBT to induce and analyze prognostic indicators. The overview of the electrical test system is shown in Fig. 1. The data collection was done on the thermal overstress ageing test and the temperature was controls within the range beyond the rated temperature (150° ) of the IGBTs IRG4BC30K which is measured individually and relays determine the sequence of the measurement. The parameters characterized are threshold voltage, breakdown voltage and leakage current.The V CE (on) was measured across the collector-emitter terminals of the transistor where the emitter is directly connected to the ground of the power supply and the collector with resistor is in series to the positive lead of the power supply. And also configuration consists at gate voltage of 15 V and Gate switching Frequency of 1 with duty cycle of %40 which the gate was driven by an independent power supply. From the results, we observe that the V CE (on) reduces with aging. There is an increased scatter in the voltage values as a result of variation in the time taken for each transistor to latch-up. The lowered voltage drop across the transistor with aging indicates reduced effectiveresistance of the transistor as this parameter is measured at a constant current. The reduction in the effective resistance of the transistors with aging is indicated by the reduction in V CE (on) results.Fig.1: overview of the electrical test system for measuring precursor parameters of IGBT [7] Remaining Useful Life (RUL)RUL and its attributes are the outcome of prognostics and are used in prognostic assessment by applying appropriate metrics and additional criteria. There is a wide range of methods dealing with RUL computation and calculation.A significant amount of research has been undertaken to develop prognostics models over recent years. By design, models are subject to specific assumptions and approximations, some of which are mathematical, while others relate to practical implementation issues such as the amount of data required to validate and verify a proposed model. Selection of an appropriate method is crucial for success in condition-based program deployment and is related to the previously-mentioned return on investment attribute. Adequate model selection necessarily requires mathematical understanding of each model type and its basic advantages and disadvantages.Each of the prognostics methods and approaches independently has its strengths and weaknesses and sometimes, a hybrid methodology is used, which profits from the advantages of all methods. Furthermore, it is quite common that the prognostics framework is part of the diagnostics framework and cannot be always isolated. Several prognostics frameworks have been developed and described and one of the best methods is data-driven approaches. A data-driven approach uses the ordinarily-observed operating data (power, vibration and acoustic signals, temperature, pressure, oil debris, currents, voltages, calorimetric data, frequency response) to track, approximate and forecast the system degradation behavior [8]. Measured input/output data is the major source for getting a better understanding of the system degradation behavior. The data-driven approaches rely on assumptions that the statistical data are relatively unchanged unless a failure occurs in the system. Data-driven prognosis is based on statistical and learning techniques from the theory of pattern recognition. These range from multivariate statistical methods (static and dynamic principle component, linear andquadratic discriminants, partial least squares and canonical variance analysis) to black-box methods based on artificial neural networks (probabilistic neural networks, multi-layer perceptron, radial basis functions), graphical models (Bayesian networks, hidden Markov model), self-organizing feature maps, signal analysis (filters, auto-regressive models, FFT, decisions trees) and fuzzy rule based systems [7].Most of the work in data-driven prognostics has been for structural prognostics. Many of those systems use vibration sensors to monitor the health of rotating machinery, such as helicopter gearboxes. Some systems monitor the exhaust gases or the oil stream from the engine for contamination that could indicate a fault[9].Dynamic Wavelet Neural Network (DWNN) utilization and RUL estimation of bearings are other examples of the current research in this area. Neural networks were trained by using vibrations signals from the damaged bearings with different levels and signs of wear. This approach appears to be accurate enough for diagnostic and prognostic purposes [6].The ability to transform and to reduce large amount of noisy data into a smaller valid and meaningful data set is the major advantage of data-driven approaches. The major disadvantage is the dependency on quality and quantity of operating data, which is a driving key element of prognostic accuracy and reliability. In summary, the data-driven approaches are preferred in the case when large amounts of run-to failure data sets are available in the required operational range and system models are not availableRemaining Useful Life Prediction by Simple State-Based Prognostic Model A simple state-based prognostic (SSBP) method is a statistical model for modeling systems that evolve through a finite number of discrete states [10]. The SSBP process basically has three steps. These are clustering, cluster evaluation, and RUL calculation. Procedures of calculating RUL by SSBP method is shown in Fig. 2. In this method, data from the different health states of multiple systems are using any clustering method in clustering stage of method. Then the RUL is estimated using the transition probabilities between health states.Fig. 2: Prognostic StepsThe IGBT dataset has seven run-to-failure samples. The dataset was separated into training and testing divisions. Five of them were used for the training and the rest were used for testing the model. Collector-emitter voltage (V CE) data was selected as a precursor parameter among the other sensory data collected, such as gate-emitter voltage or collector emitter-current, since the V CE degradation follows a monotonic trajectory and it was utilized in k-means clustering. Various numbers of health states representing the degradation starting from two to ten were tested. In the cluster evaluation part, a MATLAB(R) function (i.e. silhouette) was used in order to determine the best number of health states. Seven different health states was the best representative of the degradation process which was determined by using the silhouette function. In this scenario, IGBTs are considered to be failed when they are beyond the seven discrete health states. The first health state represents the brand new IGBT whereas the seventh one is observed to be close to failure. The testing collector-emitter voltage measurements are depicted in Fig. 2. K-means clustering basically defines discrete states whereas it is proposed to give discrete health states. Once the health states have been obtained, transition probabilities in between health states are calculated. Details of the transition probability calculation are given in [10].The testing process again starts with estimating the current health state of the IGBT. It is calculated using the cluster centroids obtained from the k-means results of the training dataset. Once the current health state of the IGBT has been obtained, the expected RUL is calculated simulating the transition probabilities obtained from the training dataset. Basically, a transition probability is the probability of changing the current state to another state. Transition probability information can be stated as a model for the calculation of the expected RUL of testing IGBTs. RUL estimation results for the testing IGBTs are shown in Fig. 4. The x-axis represents the life of each IGBT and the y-axis values represent the corresponding RUL values estimated using the SSBP model.Fig. 3: The IGBT collector-emitter voltage measurementsFig. 4: RUL estimation for the testing IGBTS .ConclusionFailure occur several states before making the system unusable. It is critical to identify and forecast these health states as the failure progresses for effective usage of the system. IGBT power electronic modules are one of the most important components of power converters. Although several studies on failure identi fication of IGBT power module are present in the literature, health state estimation and forecasting have not been reported. One of the most important dif ficulties in failure progression analysis is the inability to observe the natural progression of failures due to time constraint. Failures occur slowly and obtaining statistically enough failure progression data may take years. This paper has presented a simple state-based prognostic method (SSBP) for estimating RUL of IGBT power modules. SSBP method has been applied to data obtained from aged IGBT power electronic modules which is used in H-bridge and the RUL of an IGBT power module has been estimated.References[1]A. Kabir, C. Bailey, L. Hua, and S. Stoyanov, "A review of data-driven prognostics in power electronics," in Electronics Technology (ISSE), 2012 35th International Spring Seminar on , 2012, pp. 189-192. [2] C. Bailey, H. Lu, C. Yin, and T. Tilford, "Integrated reliability and prognosticsprediction methodology for power electronic modules," in Aircraft HealthManagement for New Operational and Enterprise Solutions, 2008 IET Seminar on , 2008, pp. 1-39.[3] B. Saha, J. R. Celaya, P. F. Wysocki, and K. F. Goebel, "Towards prognostics forelectronics components," in Aerospace conference, 2009 IEEE , 2009, pp. 1-7.[4]Y. Xiong, X. Cheng, Z. J. Shen, C. Mi, H. Wu, and V. Garg, "A Prognostic andWarning System for Power Electronic Modules in Electric, Hybrid, and Fuel CellVehicles," in Industry Applications Conference, 2006. 41st IAS Annual Meeting.Conference Record of the 2006 IEEE , 2006, pp. 1578-1584. Time (hour)L i f e T i m e (h o u r )[5] M. Musallam, C. M. Johnson, Y. Chunyan, L. Hua, and C. Bailey, "In-service lifeconsumption estimation in power modules," in Power Electronics and Motion Control Conference, 2008. EPE-PEMC 2008. 13th, 2008, pp. 76-83.[6] N. Patil, J. Celaya, D. Das, K. Goebel, and M. Pecht, "Precursor ParameterIdentification for Insulated Gate Bipolar Transistor (IGBT) Prognostics," Reliability, IEEE Transactions on, vol. 58, pp. 271-276, 2009.[7] G. Sonnenfeld, K. Goebel, and J. R. Celaya, "An agile accelerated aging,characterization and scenario simulation system for gate controlled power transistors,"in AUTOTESTCON, 2008 IEEE, 2008, pp. 208-215.[8] C. Y. Yin, H. Lu, M. Musallam, C. Bailey, and C. M. Johnson, "A prognosticassessment method for power electronics modules," in Electronics System-Integration Technology Conference, 2008. ESTC 2008. 2nd, 2008, pp. 1353-1358.[9] G. Niu, S. Singh, S. W. Holland, and M. Pecht, "Health monitoring of electronicproducts based on Mahalanobis distance and Weibull decision metrics,"Microelectronics Reliability, vol. 51, pp. 279-284, 2011.[10] O. F. Eker, F. Camci, A. Guclu, H. Yilboga, M. Sevkli, and S. Baskan, "A SimpleState-Based Prognostic Model for Railway Turnout Systems," Industrial Electronics, IEEE Transactions on, vol. 58, pp. 1718-1726, 2011.。