A lifetime prediction model based on two-path convolution with attention mechanism and bidirectional long short-term memory network

With the continuous advancement of technology, modern industrial equipment is becoming increasingly complex, integrated, and automated. The complexity of industrial processes often involves multiple variables, strong coupling, nonlinearity, variable operating conditions, and significant noise, making the establishment of accurate remaining useful life (RUL) prediction models a challenging research direction. This paper proposes a lifetime prediction model based on two-path convolution with attention mechanisms and a bidirectional long short-term memory (BiLSTM) network. The model's front end employs two-path convolution scales and attention modules to extract key fault information from bearings, enhancing the model's noise resistance. It utilizes adaptive batch normalization and Meta-Aconc activation functions to adaptively adjust the neurons of the model, thereby enhancing its generalization capabilities. The model's back end uses a BiLSTM network to remember and process the degradation information of bearings, achieving the prediction of bearing RUL. Furthermore, the model's accuracy is evaluated using root mean square error and a scoring function assessment system. Comparative experiments demonstrate the model's higher predictive accuracy. Finally, robustness and generalization experiments have proven the model to adapt well in scenarios with noise interference and working condition transitions. This model provides a reference for the prediction of the life of rotating machinery in practical scenarios with strong noise and variable operating conditions.