The health status of rotating machinery equipment in nuclear power plants is of paramount importance for ensuring the overall normal operation of the power plant system. In particular, significant failures in large rotating machinery equipment, such as main pumps, pose critical safety hazards to the system. Therefore, this paper takes pump equipment as a representative of rotating machinery in nuclear power plants and proposes a fault diagnosis method based on a multi-scale convolutional self-attention network for three types of faults: outer ring fracture, inner ring fracture, and rolling element pitting corrosion. Within the multi-scale convolutional self-attention network, a multi-scale hybrid feature complementarity mechanism is introduced. This mechanism leverages an adaptive encoder to capture deep feature information from the acoustic signals of rolling bearings and constructs a hybrid-scale feature set based on deep features and original signal characteristics in the time–frequency domain. This approach enriches the fault information present in the feature set and establishes a nonlinear mapping relationship between fault features and rolling bearing faults. The results demonstrate that, without significantly increasing model complexity or the volume of feature data, this method achieves a substantial increase in fault diagnosis accuracy, exceeding 99.5% under both vibration signal and acoustic signal conditions.

Failures in critical systems and equipment within nuclear power plants (NPPs) significantly threaten operational safety and reliability. Therefore, rapid and accurate root cause localization during the incipient stages of failure is critical to preventing escalation. Traditional modeling methods often fail to address the inherent structural complexity of NPPs, the diversity of failure modes, and the stochastic mapping relationships between symptoms and causes. To address these challenges, this paper proposes an intelligent fault diagnosis framework integrating knowledge graphs (KGs) and Bayesian networks (BNs). First, by analyzing failure modes and anomaly characteristics, we define discrimination criteria for typical faults. Second, a structured knowledge modeling approach is developed to transform unstructured fault information into a KG, which is subsequently mapped to a BN topology. Finally, to mitigate the subjectivity of expert priors, data-driven structure and parameter learning algorithms are employed to optimize the model, enhancing inference accuracy. Robustness was validated through experiments targeting three fault severity levels, using signed directed graphs (SDGs), support vector machines (SVMs), domain generalization softmax (DG-softmax) and long short-term memory (LSTM) as benchmarks. Experimental results demonstrate that the proposed method maintains high diagnostic precision across varying severities, outperforming traditional data-driven methods in accuracy and stability. This study enhances the interpretability and engineering applicability of intelligent diagnosis in nuclear power systems.