The Optimization Study of a CNN-LSTM Hybrid Model with Denoising Layer on the EngineFaultDB Dataset

Deep learning-based engine fault diagnosis methods have been widely applied in recent years. However, when dealing with sensor data that contains significant noise, existing models often struggle to maintain sufficient robustness. To address this issue, this paper proposes an innovative hybrid model that combines a Denoising Convolutional Neural Network (Denoising CNN) with Long Short-Term Memory (LSTM). In this model, a Denoising CNN is introduced at the front end of the network to effectively reduce the interference of noise on model performance, enhancing its robustness and accuracy in processing sequential data. Experimental results show that the proposed Denoising CNN-LSTM model significantly outperforms traditional CNN, LSTM, and CNN-LSTM models on the EngineFaultDB dataset, with notable improvements in evaluation metrics such as accuracy, recall, and F1 score. The model's superiority in complex noisy environments is thoroughly validated using various evaluation methods, including confusion matrices, ROC curves, F1 scores, and t-SNE visualizations. The experimental results indicate that the Denoising CNN-LSTM hybrid model can effectively mitigate the impact of noise, providing a more robust and accurate solution for engine fault diagnosis, and offering theoretical support for the future application and promotion of intelligent fault detection systems.