Convenient intelligent diagnosis for rotating machinery: An improved deep forest method based on feature...

Due to the advantages of feature autoextraction and deep architecture, deep learning-based intelligent fault diagnosis has attracted increasing attention. However, a variety of complex hyper-parameter settings greatly limit its practical applications. Moreover, it is more critical and difficult to diagnose multiple mixed faults of rotating machinery under small training samples. To bridge these gaps, this paper proposes a convenient intelligent diagnostic method based on the improved deep forest, where a feature reconstruction algorithm is used to address the high computational cost and feature submergence caused by the long time series characteristics of vibration data. Comparison experiments with typical deep neural network-based methods are implemented, and the results validate the effectiveness and superiority of the proposed method, as well as the robustness of the hyper-parameters.