[3E2-GS-2g] Machine learning

11:30 AM - 11:45 AM JST(2:30 AM - 2:45 AM UTC)

[3E2-GS-2g-03] Early detection of anomalies for sluice gate facilities by dynamic selection of training data

〇Keitaro Shoji¹, Yusuke Motegi¹, Yoshitaka Suzuki¹ (1. IHI Corporation)

Keywords:

Machine Learning,XAI,anomaly diagnosis

In recent years, climate change has heightened the risk of water-related disasters, making advanced disaster prevention and mitigation essential. Reliable flood control depends on continuous, accurate monitoring of sluice gate facilities and early detection of possible failures. IHI is developing advanced technologies for anomaly detection and predictive maintenance using operational data from these facilities. Traditional methods mainly monitored steady-state operations, but data characteristics can vary depending on operational modes and environmental factors, reducing detection accuracy when analyzed collectively. To address this, we developed a method that dynamically selects optimal training data based on the current operation mode and environment when analyzing time-series data. Furthermore, we applied an index that quantifies the contribution of each factor to deviations from the normal state to the frequency components of vibration data during anomalies. This analysis demonstrated the potential for fault diagnosis. This paper describes the sluice gate monitoring technology that dynamically selects training data and presents validation results demonstrating its effectiveness.