[4F3-OS-30c] OS-30

[4F3-OS-30c-01] A Method for Improving the Accuracy of Equipment Degradation Diagnosis by Human-in-the-Loop approach

Reducing False Detections through Interaction between Humans and AI in Regression-Based Anomaly Diagnosis

Takaaki Nakamura¹, 〇Yuki Tanaka¹, Takahiko Masuzaki¹ (1. Mitsubishi Electric Corporation)

Keywords:Predictive Maintenance, Health Diagnostics, Human-in-the-Loop AI

Monitoring sensor data from equipment to perform maintenance at the appropriate timing, known as condition-based maintenance, is in high demand across various fields such as factory automation (FA), plants, and transportation infrastructure. Data-driven degradation diagnosis using AI requires interpretation and judgment by experts regarding the AI's output. However, for experts who are not familiar with AI, understanding the output and providing appropriate feedback can be challenging. This paper proposes a human-in-the-loop anomaly diagnosis method for equipment, aiming to improve diagnostic accuracy through interaction between experts and AI. Focusing on a degradation detection method based on regression models, we explored ways to reduce false detections through expert-AI interaction. By visualizing the trends in degradation detection, experts can identify the causes of false detections and subsequently improve the degradation diagnosis model. We applied the developed prototype to degradation diagnosis using vibration data from a model railway and confirmed that it effectively reduced false detections.