Network-based Root Cause Identification to Improve OEE in High-Precision Manufacturing

Abstract

Small deviations in the production cycle can cause expensive downtime or quality deviations in high-volume, high-precision production lines. If no precise root cause can be identified, only the symptoms are eliminated, resulting in a pattern of repetitive failures and temporary remedial measures. This paper presents a knowledge-based framework and algorithm that combines network science and graph theory to detect anomalies and identify root causes. The approach converts multivariate time series data into temporal multiplex recurrence networks and uses eigenvalue-based anomaly detection in addition to causal process graphs (CPGs). The framework is evaluated on a simulated pick-and-place production line with four failure scenarios. This contributes to a causal and transparent identification of root causes, which will be benchmarked against other methods in future work using real manufacturing data.