A comparison of anomaly detection methods for industrial screw tightening

Abstract

Within the context of Industry 4.0, quality assessment pro- cedures using data-driven techniques are becoming more critical due to the generation of massive amounts of production data. In this paper, we address the detection of abnormal screw tightening processes, which is a relevant industrial task. Since labeling is costly, requiring a manual effort, we focus on unsupervised approaches. In particular, we assume a low-dimensional input screw fastening approach that is based only on angle-torque pairs. Using such pairs, we explore three main unsuper- vised Machine Learning (ML) algorithms: Local Outlier Factor (LOF), Isolation Forest (iForest) and a deep learning Autoencoder (AE). For benchmarking purposes, we also explore a supervised Random Forest (RF) algorithm. Several computational experiments were held by us- ing recent industrial data with 2.8 million angle-torque pair records and a realistic and robust rolling window evaluation. Overall, high quality anomaly discrimination results were achieved by the iForest (99%) and AE (95% and 96%) unsupervised methods, which compared well against the supervised RF (99% and 91%). When compared with iForest, the AE requires less computation effort and provides faster anomaly detection response times.