A remaining useful life prediction method based on CNN-BiLSTM feature transfer in a high-noise environment

Prognosis and health management (PHM) is a comprehensive technique for fault detection, prediction, and health management. However, achieving accurate predictions of remaining useful life (RUL) under complex working conditions such as is still High-Noise Environment a challenge. Therefore, this paper proposes a feature transfer model based on Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory Neural Networks (BiLSTM) to predict RUL. In the feature extraction stage, On the basis of signal decomposition using local mean values，CNN is used to extract the degradation features. Secondly, the health factors are constructed by monotonicity and correlation to filter the features again. Thirdly, it uses BiLSTM to model the time series data in the RUL prediction stage. Then, it introduces the transfer learning algorithm to solve the problem of different data distribution due to the inconsistent working conditions of mechanical equipment data and estimates the confidence interval of the RUL by the Monte Carlo simulation technique. Finally, the effectiveness of our constructed framework via CNN-BiLSTM model on a publicly available degradation simulation dataset of turbine engines.