Domain-Knowledge-Informed Functional Outlier Detection for Line Quality Control Systems

Detecting defects in faulty products in advance is a critically important issue. To address this, modern production systems collect diverse sensor data at quality inspection stations to monitor the condition of the products. However, in many cases, the severity of defects tends to worsen gradually after production, making it challenging to identify minute defects at the quality inspection stations that correspond to the early stages of defects. Consequently, distinguishing such defects solely based on collected data becomes almost impossible. In this study, we demonstrate that state-of-the-art functional outlier detection methods using sensor data struggle to effectively detect these subtle defects. To overcome this limitation, we propose an approach that leverages domain knowledge to enable the effective detection of minute defects. To validate our proposed method’s effectiveness, we evaluated its performance using simulated data and real sensor data from refrigerator inspection lanes