Power IGBT Remaining Useful Life Estimation Using Neural Networks based Feature Reduction

Abstract

Power converters (PCVs) are a fundamental part of wind energy systems (WESs). Generally, the PCVs are subjected to high failure frequency rate and consequently to the system downtime due to its structural complexity. Therefore, the main propose of this study is to develop an understanding prognostic approach dedicated to WES PCVs. The proposed method investigated the Insulated Gate Bipolar Transistor (IGBT) failure in PCVs. Therefore, it integrates the data-based methodology. Usually, data-based approaches based on a recorded data base. Therefore, in this paper, we use the collectoremitter voltage as a precursor signal. In this work, the prediction of the Remaining Useful Life (RUL) of the IGBT device is achieved through the Feedforward Neural Network (FFNN) technique. Moreover, we used the time-domain analysis to extract useful information used as an input in the health indicator generation task. In this paper, the health indicator generation task is achieved using the Principal Component Analysis (PCA) technique. The advantages of the proposed prognostic method in term of RUL prediction are illustrated using a real IGBT accelerated ageing database provided from NASA Ames laboratory Prognostics Center of Excellence.