On Anomaly Detection for Autonomous Transfer Vehicles in Smart Factories

Today, autonomous transfer vehicles (ATVs) have important roles in many smart factories. Therefore, flawless and uninterrupted operation of ATVs is required for the sake of effective production in smart factories. For this reason, it is important to detect anomalies (or, abnormalities) regarding ATVs during the operation. Therefore, this study aims to detect anomalies regarding ATVs so that possible losses during production can be prevented. For this purpose, two novel methods are proposed to detect anomalies for ATVs. The first method employs exhaustive feature selection to obtain the optimal subset of features for detecting anomalies. The other method utilizes a 2-stage hybrid approach for anomaly detection. Four types of anomalies (overdue pick-up delivery activity, unexpected pedestrian density, unexpected vehicle slow-down, and unexpected vehicle behavior) are considered for this work. During the experimental work, a test environment has been established for simulating a smart factory. The experimental results indicate that the first method provides a higher accuracy whereas the second one offers a better false-negative rate in detecting anomalies regarding ATVs.