An Enhanced Deep Forest Rolling Bearing Fault Diagnosis Method

Abstract

Deep forest(DF), as a new deep learning model, has superior performance in model tuning and training time and has been widely studied in recent years. However, there are still considerable challenges in reducing the disadvantages of the multi-granularity scanning process and cascading forest stitching. To this end, this paper proposes a novel deep forest-based rolling bearing fault diagnosis method, which enhances feature learning using the developed multi-scale-self-attentive strategy. The highlight of this method is the designed multi-scale self-attentive reinforced feature extractor and low-dimensional self-encoder. It significantly mitigates the feature swamping caused by interference signal features and cascading forest dimensional splicing, thus effectively learnings the bearing fault features. Benefiting from it, the proposed method can perform better. Diagnostic tasks built on the rolling bearing dataset validate the proposed method’s effectiveness and superiority.