Study on Life Prediction Method of Ball Screw Base on Constructed Degradation Feature and IGWO-BiLSTM

Abstract

Regarding the problem of reduced remaining useful life (RUL) due to wear of the ball screw in the feed system of CNC (computer numerical control) machine tools, a prediction method based on constructing the degradation feature vector of the signal data and the improved gray-wolf optimization with bidirectional long short-term memory (IGWO-BiLSTM) neural network regression model is proposed. Firstly, a time-domain analysis and the complete ensemble empirical mode decomposition with adaptive noise analysis (CEEMDAN) were carried out based on the collected life cycle signal data of a ball screw. The time-domain feature vector and the energy feature vector of each IMF (intrinsic mode function) component after CEEMDAN decomposition were constructed. The Pearson correlation coefficient was used to filter feature vectors and construct the multivariate feature vector. Secondly, this paper improves the traditional gray wolf optimization algorithm, adds a search strategy based on dimension learning, and combines the improved algorithm with the BiLSTM model, based on the IGWO-BiLSTM theory. A regression model between feature vectors and the remaining life of a ball-screw system was established. Finally, the prediction model was established according to the proposed method and compared with the other five neural network models: LSTM, BiLSTM, BO-LSTM (Bayesian optimization of LSTM), BO-BiLSTM, and IGWO-LSTM. The results indicate that this method has high accuracy and good generalization ability for predicting the remaining life of a ball-screw system.