Proposal for Concept Drift Detection in Factory Automation

This document is a proposal that addresses the issue of concept drift detection in the factory automation (FA) domain. It outlines the challenges faced in anomaly detection due to frequent changes in the 4M factors—Man, Machine, Material, and Method—leading to increased product defects. The proposal suggests a method that integrates multiple anomaly detection models, taking into account causal relationships to enhance the detection of product defects and concept drift. The paper discusses the limitations of conventional methods, such as moving averages and average-run-length techniques, which struggle with real-time performance and delayed detection. The proposed solution utilizes the Isolation Forest algorithm, which is noted for its high-speed inferencing capabilities and minimal operational skill requirements. The effectiveness of this method is verified through experiments conducted on an injection molding machine, demonstrating its ability to detect both product defects and concept drift in real-time, thus addressing the operational challenges in modern manufacturing environments.