Using LSTM neural network to predict remaining useful life of electrolytic capacitors in dynamic operating conditions

A critical aspect for prognostics and health management is the prediction of the remaining useful life (RUL). The existing RUL prediction techniques for aluminum electrolytic capacitors mostly assume the operating conditions remain constant for the entire prediction timeline. In practice, the electrolytic capacitors experience large variations in operating conditions during their lifetime that influence their degradation process and RUL. This paper proposes a RUL prediction method based on deep learning. The proposed framework uses the original condition monitoring and operating condition data without the necessity of assuming any particular type of degradation process and, therefore, avoiding the requirement of establishing link between model parameters and operating conditions. The proposed framework first identifies the degrading point and then develops the Long Short-Term Memory (LSTM) model to predict the RUL of capacitors. The LSTM-based method can reduce the computational time and complexity while ensuring high prediction performance. Its effectiveness is demonstrated by utilizing the simulated degradation process and temperature condition time-series of aluminum electrolytic capacitors used in electric vehicle powertrain.

Automated summary

This paper proposes a RUL prediction method for aluminum electrolytic capacitors based on deep learning. The method uses original condition monitoring and operating condition data, avoiding the need to assume a particular degradation process, while ensuring high prediction performance.