This paper presents a novel solution for bearing fault diagnosis based on a Broad Learning System with multi-domain feature selection. Currently, the diagnosis of bearing faults is beset with numerous challenges, including operation under complex and fluctuating conditions, the subtlety of early fault indicators, noise interference in measurement signals, variations in operational conditions, and a broad spectrum of potential fault types. These factors collectively contribute to the complexity of accurately identifying and classifying bearing faults. The Broad Learning System introduced in this paper, by integrating multi-domain feature selection, enhances the model's capability to extract and utilize relevant features from noisy or incomplete data, potentially increasing diagnostic accuracy in suboptimal sensor environments. By leveraging features across multiple domains, the method based on the Broad Learning System can capture a wider array of fault characteristics, thereby broadening its applicability across various types of bearing faults and operational conditions.