The improved CNN_BILSTM hybrid model for predicting compressor stall

Compressor stall is a destructive failure mode in aero-engines and poses a significant threat to aircraft safety. The timely diagnosis and prediction of such events have long been key areas of research. However, traditional stall early warning techniques suffer from inherent limitations. With the rapid development of artificial intelligence, the integration of deep learning approaches offers enhanced speed, accuracy, and reliability for stall prediction. This study first evaluates the advantages and limitations of various baseline deep learning models in predicting compressor stall. Based on this analysis, a hybrid model, CBLSTM (CNN_BiLSTM), is proposed, which combines a one-dimensional convolutional neural network (Conv1D) with a bidirectional long short-term memory network (BiLSTM). The Conv1D layer is used to extract local features from the input data, which are subsequently processed by the BiLSTM layer to capture temporal dependencies. Using the NASA Rotor67 transonic fan as the case study, prediction performance evaluation and sensitivity analyses were conducted on stall simulation datasets under different operating conditions. The experimental results show that the CBLSTM model can effectively extract precursor features of stall and provide timely predictions, demonstrating high accuracy and strong generalization capability. These results highlight the model's potential for real-world applications.