Intelligent Fault Detection of Rolling Bearings Based on Bayesian Capsule Network and Self‐Calibrated Convolution

In critical sectors such as aerospace and automotive manufacturing, rolling bearings are required to operate under extreme conditions including high temperatures and high rotational speeds. Their failure can easily trigger safety incidents and cause chain reactions throughout the industrial supply chain. However, existing deep learning‐based diagnostic methods exhibit significant shortcomings in adaptability to varying operating conditions, robustness against noise, and diagnostic capability with limited samples. This study aims to construct an intelligent diagnostic model that balances both precision and environmental adaptability, addressing the challenge of bearing fault identification in complex industrial scenarios. By integrating Bayesian algorithms into the capsule network, hyperparameters and the iteration count of dynamic routing are optimized. Simultaneously, a self‐calibrating convolutional block is introduced to achieve multi‐scale feature extraction and noise suppression, while Local Maximum Mean Discrepancy (LMMD) is employed to align the distributions of the source and target domains. These components collectively form the optimized self‐calibrating convolutional capsule network (OSCC‐CapsNet). Experimental results demonstrate that the proposed model achieves an accuracy of 98.0% under constant conditions, an average accuracy of 93.0% under varying conditions, and a compound fault classification rate of 98.4%. The study confirms that the model effectively captures hierarchical relationships within the fault feature space, maintaining excellent diagnostic performance even in scenarios with small sample sizes, high noise levels, and varying operating conditions. This provides an efficient technical solution for rolling bearing fault diagnosis.