Remaining Useful Life Prediction of Gearbox under Variable Operating Conditions based on KPCA and LSTM

In the field of gearbox research, the remaining useful life prediction for gears is a critical task. This paper investigates the essential technology for predicting the remaining useful life of gears. Firstly, the time synchronous averaging (TSA) algorithm is employed to reduce the noise in the vibration signal. The gear's vibration data is divided into six operating conditions based on torque changes. Secondly, to determine the variation rule of the remaining life of the gear, reduce the interference of different operating conditions, and extract appropriate fault parameters, healthy operating conditions are treated by different operating conditions, a vector autoregressive (VAR) model that considers the interaction of different independent variables is adopted for different operating conditions. Thirdly, after extracting features from the VAR data, kernel principal component analysis (KPCA) is used to reduce the dimensionality of the feature matrix while retaining as much original information as possible and reducing data volume. Finally, long short-term memory (LSTM) neural networks are employed to predict the remaining life of the gear, avoiding possible safety hazards and economic losses. The average error between the predicted output value of the model and the actual value is 0.15%. The results demonstrate that this technique can accurately predict the remaining useful life of gears and provide practical support for equipment maintenance.