False sensor-data detection strategy for post-hazard condition monitoring of nuclear systems using statistical approaches and long short-term memory

Next-generation nuclear power plants are advancing toward autonomous Online Monitoring (OLM) systems to ensure operational safety and efficiency. A critical factor in the reliability of OLM systems is the integrity of sensor data collected from the facility. Erroneous sensor data can compromise the OLM's ability to accurately assess the plant's condition, potentially leading to severe safety risks. For instance, in post-hazard scenarios like earthquakes, undetected faulty data may obscure degraded conditions in piping and equipment systems. Such degradation, if unchecked, can escalate into catastrophic events, including loss of coolant accidents (LOCAs). This study introduces an innovative False Signal Detection and Correction Model (FSDCM) designed to safeguard OLM data integrity. The FSDCM operates through a two-step mechanism: first, it employs statistical correlation analysis to detect false sensor data; second, it uses deep learning algorithms to correct these inaccuracies. By analyzing historical data and learning patterns, the deep learning component can overwrite erroneous sensor readings with validated data, enhancing reliability. A case study on a nuclear piping system demonstrates FSDCM's effectiveness. Using finite element simulations, acceleration-time series signals are generated as sensor data, and random noise is introduced to simulate false signals. The FSDCM accurately identifies and corrects these anomalies across various test scenarios, showing robust detection and correction capabilities. This novel framework not only enhances operational accuracy but also plays a vital role in risk mitigation for nuclear facilities, paving the way for safer, more autonomous power plant management.