An Intelligent Fault Diagnosis Framework for Rolling Bearings With Integrated Feature Extraction and Ordering-Based Causal Discovery

Recent advancements in data-driven deep learning methods have significantly improved rolling bearing fault diagnosis. However, these frameworks face limitations due to data defects and inadequate feature extraction. To overcome these obstacles, this paper proposes a novel feature extraction and causal model-based diagnosis framework. This framework leverages causal discovery to derive causal models from real data, accurately estimate causal effects, and enhance fault diagnosis performance. Specifically, the proposed framework utilizes the Time-reassigned Synchrosqueezing Transform (TSST) to process collected rolling bearing signals, transforming them from the time domain to the time-frequency domain. This transformation enables the extraction of energy distribution and frequency components of vibration signals at different frequencies, effectively reducing noise interference. Subsequently, a Directed Acyclic Graph (DAG) is constructed using an ordering-based causal discovery method to identify potential interfering factors that may impact vibration signals, facilitating causal effects estimation. Furthermore, by integrating the Vision Transformer (ViT) network with causal effects estimation techniques, the framework achieves end-to-end rolling bearing fault diagnosis. Experimental results on laboratory-bearing datasets demonstrate the superior performance of this proposed fault diagnosis framework compared to existing methods.