In the high-stakes world of nuclear power, ensuring the safe and efficient operation of pressurized water reactors (PWRs) is paramount. A breakthrough in fault diagnosis technology, developed by researchers at the University of South China, promises to revolutionize how we maintain and operate these critical energy assets. Led by Zhaohui Liu from the School of Computing/Software, the team has devised a novel framework that significantly enhances the accuracy and reliability of fault detection in PWRs, even under the most challenging conditions.
The stable operation of nuclear power plants is crucial for energy security, carbon emission reduction, economic benefits, and nuclear safety. Fault diagnosis, which involves analyzing operational data to detect and address potential anomalies in time, is a key technology for ensuring the safe operation of NPPs. Current fault diagnosis approaches can be broadly classified into knowledge-driven and data-driven methods. Knowledge-driven approaches rely on expert experience and physical models, yet their applicability is often constrained in complex systems. In contrast, data-driven methods leverage deep learning to automatically extract features, making them well-suited for handling complex nonlinear problems and demonstrating significant potential.
The challenge lies in the limited availability of labeled data and the severe class imbalance, particularly during Design Basis Accident (DBA) scenarios. “The high cost of data annotation in nuclear power plants results in limited labeled datasets, leading to the small-sample problem,” explains Liu. “Additionally, operational data from NPPs significantly outweigh fault data, causing a severe class imbalance issue.”
To tackle these hurdles, Liu and his team have developed a three-stage framework. First, they use a signed directed graph to select key parameters that significantly influence system behavior. This dimensionality reduction step ensures that only the most relevant data is analyzed. Next, they employ Gramian Angular Difference Field (GADF) imaging to encode temporal features, transforming one-dimensional vector groups into meaningful visual representations. Finally, they introduce an improved Deep Subdomain Adaptation Network (DSAN) that uses weighted Focal Loss and confidence-based pseudo-label calibration to enhance sensitivity to minority-class samples and improve fault classification under small-sample and class-imbalanced conditions.
The results are impressive. On a transfer task from the NPPAD source set to the PcTran-simulated AP-1000 target set across five DBA scenarios, the framework raised the overall accuracy from 72.5% to 80.5%, increased macro-F1 to 0.75 and AUC-ROC to 0.84, and improved average minority-class recall to 74.5%. This performance outperforms the original DSAN and four other baseline models, explicitly prioritizing minority-class samples and mitigating pseudo-label noise.
The implications for the energy sector are profound. Enhanced fault diagnosis capabilities mean better predictive maintenance, reduced downtime, and increased safety. For nuclear power plant operators, this translates to improved operational efficiency and significant cost savings. “Our framework shows significant advantages in multi-class classification tasks, especially under conditions of small sample sizes and class imbalance,” Liu notes. “This indicates that the weighted Focal Loss and confidence-based pseudo-label calibration strategies effectively enhance the model’s classification ability.”
The research, published in Energies, opens new avenues for developing more robust and reliable fault diagnosis systems. Future work will focus on validating the framework on actual plant operational logs and exploring more robust pseudo-label self-correction methods, optimizing domain alignment with adversarial learning, and enhancing class balancing strategies.
As the energy sector continues to evolve, the need for advanced diagnostic tools becomes ever more critical. Liu’s work represents a significant step forward, offering a glimpse into a future where nuclear power plants operate with unprecedented levels of safety and efficiency. The commercial impacts are vast, promising a more secure and sustainable energy landscape for all.
