A Method Based on an Autoencoder for Anomaly Detection in DC Motor Body Temperature

Abstract

Anomaly detection has an important role in industrial systems. Abnormal situations occurring in a system cause anomalies, and the anomalies reduce system performance over time, and may also make the system malfunction. Therefore, the correct and timely detection of anomalies is of critical importance for predictive maintenance. In this study, an autoencoder-based method is proposed for anomaly detection in DC motor body temperature. The performance of the method was examined on a dataset that was created specifically for this study. In the experiments, the three-sigma outlier method was also applied on the same dataset for the same purpose and its performance results are used for comparison. The performance results of both methods are represented in terms of three measures, namely, accuracy, recall, and precision. The experimental study showed that the proposed method achieved over 96% ratios for all three measures, and it can be successfully used for anomaly detection in DC motor body temperature. Additionally, it can be concluded that the proposed system can be preferred for anomaly detection in time series data collected from different types of sensors when the performance results are taken into consideration.