Improved denoising autoencoder for crusher fault diagnosis

In response to the challenges of scattered and sparse data, as well as the difficulty in distinguishing noise and faults due to the limited number of fault samples in crusher fault diagnosis, an improved denoising autoencoder-based fault diagnosis method is proposed. Firstly, Bayesian optimization is used to fine-tune the hyperparameters of the CTGAN, identifying the optimal parameters to address the issue of insufficient data diversity under default settings. Secondly, polynomial feature transformation is applied to enhance the denoising autoencoder, reducing data noise while avoiding the loss of valuable information that can occur when directly using the denoising process. Finally, the proposed method is validated using four machine learning classification algorithms: Deep Belief Network (DBN), XGBoost, Gated Recurrent Units (GRU), and Deep Neural Network (DNN). The results show that the data processed using the proposed method reaches above 90% accuracy on DBN, XGBoost, GRU, and DNN.