A Comparison of Anomaly Detection Methods for Industrial Screw Tightening

Abstract

Within the context of Industry 4.0, quality assessment procedures 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 unsupervised 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 using 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.

Bosch’s Industry 4.0 Advanced Data Analytics: Historical and Predictive Data Integration for Decision Support

“…Several machine learning models were trained for this demonstration case [23] but only the two best performing unsupervised models were selected and the predictions included in the BDW, namely the ones that used the Isolation Forest and the Autoencoder. The Isolation Forest (iForest) leverages a clear distinction of characteristics of anomalous points which are present in fewer quantities and numerically different to normal instances and isolates them from normal points.…”

An Empirical Study on Anomaly Detection Algorithms for Extremely Imbalanced Datasets

“…In effect, the early studies addressing this task dates back to the 1960s [15]. In more recent years, a diverse range of algorithms have been proposed for anomaly detection, including based on statistics [17], clustering [8,1], classification [8,33,29,34,26] and graph mining [2]. In particular, supervised classification approaches require labeled data that is often difficult to obtain (e.g., requiring human effort).…”

“…Within this context, Machine Learning (ML) algorithms have been widely applied to anomaly detection. A common approach is to employ supervised learning methods, such as Logistic Regression [29], Decision Trees (DT) [33] and Random Forests [34]. These supervised ML algorithms are often coupled with resampling techniques to balance the training data, such as SMOTE [9] or Gaussian Copula (GC) [32].…”

“…An alternative is to employ an unsupervised learning, in particular via a one-class learning approach, where the ML algorithms are only fed with normal examples (the majority class). This includes algorithms such as Isolation Forest (IF) [20] and deep AutoEncoders (AE) [26,34].…”

Deep autoencoders for acoustic anomaly detection: experiments with working machine and in-vehicle audio

“…In particular, AutoEncoders (AE) are becoming popular for one-class AAD [14,15]. When compared with other ML approaches (e.g., IF and OC-SVM), AE present the advantage of requiring a lower computational effort [16,17], which is a valuable asset for designing AAD intelligent systems capable of working with operating machines and vehicles. This work is set within a larger R&D project that addresses an unsupervised in-vehicle intelligence using multiple data sources (e.g., images, sound, particles) and that includes the Bosch Car Multimedia S.A. (BCM) company.…”