School of Architecture and Transportation Engineering, Guilin University of Electronic Technology, Guangxi, China.
Guangxi Key Laboratory of Manufacturing System & Advanced Manufacturing Technology, School of Mechanical and Electrical Engineering, Guilin University of Electronic Technology, China.
Comput Math Methods Med. 2022 Jul 14;2022:7459354. doi: 10.1155/2022/7459354. eCollection 2022.
The insulated gate bipolar transistor (IGBT) is widely utilized in the transportation, power, and energy domains because of its high input impedance and minimal on-voltage drop. IGBTs are frequently used in industrial applications for lengthy periods of time, collecting fatigue damage and eventually aging and failing, which can result in system shutdown and financial losses in severe circumstances. As a result, a study into the IGBT's reliability is extremely important. Fault prediction technology, which is an important aspect of reliability research, may analyze device state through changes in terminal parameters, anticipate aging trends, and issue early warnings at thresholds to avoid significant safety issues caused by IGBT aging failures. Therefore, the appropriate end parameters are selected as aging characteristic parameters, and fault prediction is performed. Therefore, this paper has carried out research on the IGBT fault prediction technology that integrates the terminal characteristics and artificial intelligence neural network. The main research contents include the following: (1) this paper starts from the basic principle of IGBT and the structure of its device and analyzes its failure mode on the failure of IGBT. The characteristic parameter of collector-emitter turn-off peak voltage value is selected for IGBT fault prediction, and the aging data of NASA PCoE Research Center is used to verify that the characteristic parameter can be used for fault prediction. (2) In view of the shortcomings of traditional fault forecasting methods, this paper proposes to use deep learning time series forecasting methods for fault forecasting. The LSTM is theoretically analyzed, and the prediction network is built. The experimental results show that the LSTM network model can improve the accuracy of IGBT fault prediction, with fewer parameters and higher prediction efficiency.
绝缘栅双极晶体管（IGBT）由于其高输入阻抗和最小导通压降，广泛应用于交通、电力和能源领域。IGBT 常用于工业应用，长时间运行，收集疲劳损伤，最终老化和失效，在严重情况下会导致系统停机和经济损失。因此，研究 IGBT 的可靠性非常重要。故障预测技术是可靠性研究的一个重要方面，它可以通过终端参数的变化来分析设备状态，预测老化趋势，并在阈值处发出预警，以避免因 IGBT 老化故障而导致的重大安全问题。因此，选择合适的终端参数作为老化特征参数，并进行故障预测。因此，本文开展了基于终端特性和人工智能神经网络的 IGBT 故障预测技术研究。主要研究内容包括：（1）本文从 IGBT 的基本原理和器件结构出发，分析了 IGBT 失效时的失效模式。选择集射极关断峰值电压值作为 IGBT 故障预测的特征参数，并利用 NASA PCoE 研究中心的老化数据验证了该特征参数可用于故障预测。（2）针对传统故障预测方法的不足，本文提出采用深度学习时间序列预测方法进行故障预测。对 LSTM 进行理论分析，并构建预测网络。实验结果表明，LSTM 网络模型可以提高 IGBT 故障预测的准确性，具有较少的参数和更高的预测效率。
