基于退化数据和DBN算法的IGBT健康参数预测方法

Prediction Method of IGBT Health Parameters based-on Degradation Data and DBN Algorithm

摘要:

绝缘栅双极型晶体管(IGBT)等电子元器件被广泛用于运输和能源部门,其健康状态对于设备安全和有效至关重要；在对IGBT的结构和损伤机制分析基础上,结合NASA艾姆斯中心开展的IGBT加速退化试验,选择集电极-发射极关断峰值电压作为失效特征参数,提出了一种基于深度信念网络的预测模型对其进行分析和预测；以Levenberg-Marquardt(LM)算法模型作为对比,实验结果显示文章提出的三隐藏层DBN模型相比于LM模型有更好的预测性能和更高的预测精度。

Abstract:

Insulated Gate Bipolar Transistor (IGBT) and other electronic components are widely used in the transport and energy sector, its health status for equipment safety and effectiveness is essential. Based on the analysis of the structure and failure mechanism of the IGBT, the peak of the collector-transmitter voltage is selected as the failure characteristic parameter combining with the accelerated degradation experiment data from NASA Ames Center. Then a prediction method based on the Deep Belief Network (DBN) is proposed for the analysis and prediction of the trend of the IGBT. Comparing with the Levenberg-Marquardt (LM) algorithm model, the experimental results show that the proposed three hidden layer DBN model has better prediction performance and higher prediction accuracy than LM model.