A CNN-BiLSTM-attention-based framework for predicting cavitation damage degradation in centrifugal pumps

Cavitation-induced degradation is a major failure mechanism in centrifugal pumps. To characterize the degradation trajectory of cavitation damage and assess its severity, this study proposes a deep learning-based method for predicting the progression of cavitation damage in centrifugal pumps. A dedicated experimental platform was developed to simulate severe cavitation scenarios, and multi-channel vibration acceleration signals are synchronously collected from multiple measurement points. A sliding window approach is used to segment the time-series data. A convolutional neural network (CNN) extracts rich, high-dimensional spatial features, while an attention mechanism dynamically assigns weights to different sensor channels to enable multi-source data fusion. Subsequently, a BiLSTM network captures temporal dependencies in both forward and backward directions, improving prediction accuracy. Experimental results demonstrate that the proposed model effectively captures the degradation trajectory of cavitation damage, achieving a mean squared error (MSE) of 0.0098.