IGBT Fault Detection for Three Phase Motor Drives using Neural Networks Marjan Alavi 1,2 Ming Luo 1 Danwei Wang 2 Haonan Bai 2 1 Singapore Institute of Manufacturing Technology 71 Nanyang Drive, Singapore 638075 mluo@SIMTech.a-star.edu.sg 2 Nanyang Technological University 50 Nanyang Avenue, Singapore 639798 alav0001@e.ntu.edu.sg Abstract Motor drives are widely used in industry for controlling the speed of three phase AC motors. Faults in motor drives degrade motor performance and can cause catastrophic failures. IGBT (Insulated Gate Bipolar Transistor) switch faults are one of the main roots of electrical faults in in- verters and motor drives. In this paper, a method based on neural network is implemented to detect and isolate switch faults in a three phase voltage source inverter. Only the output signals of the inverter are monitored. The entropy of the phase current and voltage is selected as the switch fault feature. Single and multiple short and open circuit switch faults are isolable with this method. 1. Introduction Nowadays, the ability to maintain healthy operation of any system is of great importance from efficiency, pro- ductivity, and safety aspects. Every machine is subject to faults. Although many types of faults can be prevented during the design stage, there always exist probabilities of system failure due to device ageing, overloading, and un- predicted situations. To prevent the fault propagation and prevent unexpected shut-down of the vital systems, indus- tries are highly interested in fault tolerant systems. The first step towards having fault tolerant systems is fault de- tection and isolation (FDI). Fault diagnostic systems also can reduce the maintenance cost of the machines by elim- inating unnecessary scheduled maintenance services. A large number of hybrid electric vehicles consist of three phase induction motor drives. The precise torque control of these motors has been made possible by power electronics with controllable solid state switches. How- ever, the solid state switches can fail by remaining open or short regardless of the control signal. Power electronic inverters are considered as the weakest link in the electro- mechanical systems because of the high probability of failure of the semiconductor switches. In current tech- nologies, given the high reliability required in almost all systems, the ability to detect a system fault at the earliest possible stage is of primary interest [15]. Over the last decades, a number of intelligent systems approaches have been investigated in signal fault diag- nosis. Model-based approaches [9], fuzzy logic, artifi- cial neural networks, and case-based reasoning (CBR) are popular techniques used in various fault diagnostics prob- lems in electrical systems [14]. There exist a good amount of work in the literature on fault diagnostics for power electronics inverter [2, 13, 8, 3, 5, 12, 16, 10, 4]. A novel method based on fault dictionary that uses entropy as a preprocessor to diagnose faulty behavior in switched cur- rent circuit was proposed in [18]. They used a neural net- work for diagnosing soft faults in electronic components of switch current integrated circuits. In this research, a fault diagnostic system using neural network is studied for detecting multiple switch faults in a motor drive. The system is modeled by a three phase voltage source inverter, and a three phase RL load is con- sidered. A fault dictionary is built to capture 73 different fault modes with the higher rate of occurrence. The volt- age and current signals are collected via output detectors. In order to reduce the dimension of the fault dictionary, entropy and mean value of the detected signals are ex- tracted. These features are used to train a neural network system as the fault diagnosis engine. Simulation results show these fault isolation method is capable of addressing major single and multiple faults switch faults with high accuracy. The model of the inverter and the switch faults are described in Section 2. Section 3 describes the entropy function and the feature extraction process. The neural network structure and the fault isolation method come in Section 4. The fault detection and isolation results come in Section 4.3 and limitations of the presented method are 978-1-4673-4737-2/12/$31.00 ©2012 IEEE