Anomaly Detection in a Production Line: Statistical Learning Approach and Industrial Application

This paper focuses on industrial engineering, specifically discrete manufacturing processes, and emphasizes controlling production lead time (PLT) of production orders (POs) through the various workstations in a production line (PL). It highlights the use of dashboard control systems for monitoring PO flow, enabling managers to efficiently oversee and regulate PO progression. The research introduces a novel mathematical model for anomaly detection, identifying significant deviations from the expected time required for POs to flow through a PL’s designated workflow. The model aggregates multiple individual classification models, providing profound insights into data patterns and enhancing overall robustness. Applied to real-world data, it demonstrates practical utility and employs information theory, particularly Shannon’s entropy, to quantify uncertainty. As a result, the model exhibits a strong ability to distinguish between outliers and anomalies. Using statistical learning techniques, the model is scalable and can be integrated into dashboard control systems for real-time anomaly detection, aligning with Industry 4.0 principles of smart manufacturing and data-driven decision-making.