A Sparse Convolutional Autoencoding Fault Diagnosis Method Integrating Multiple Regularizers

Deep learning models with powerful autonomous learning and feature extraction capabilities were applied with success to bearing failures diagnosis. However, the single deep learning model is limited by its own structural characteristics, and there are still some problems such as weak generalization ability, insufficient overfitting and stability. In response to such problems, based on the idea of decision-level fusion, a sparse convolutional autoencoding fault diagnosis methodology integrating multiple regularizers is proposed in current research. Firstly, to enhance the robustness and stability of the model, six independent sparse convolutional autoencoder base models are constructed by using six different regularizers. Secondly, to enhance the recognition properties of this model, the weighted voting strategy is used to combine the six sparse convolutional autoencoder base models to construct the final ensemble learning model. Finally, in order to prove the validity of the suggested approach, a large number of tests are conducted based on the bearing dataset, and the experimental findings indicate that the suggested approach could efficiently increase the learning recognition precision.