Decision-making for the anomalies in IIoTs based on 1D convolutional neural networks and Dempster-Shafer theory (DS-1DCNN)

Abstract

The main motivation of the Internet of Things (IoT) is to enable everyday physical objects to sense and process data and communicate with other objects. Its applications in industry are called industrial Internet of Things (IIoTs) or Industry 4.0. One of the main goals of the IIoT is to automatically monitor and detect unexpected events, changes, and alterations to the collected data. Anomaly detection includes all techniques that identify data patterns deviating from the expected behavior. Deep learning (DL) can search for a specific relationship in billions of corporate IoT data and reach a meaningful goal by analyzing and classifying collected data, leading to making the right decisions. The realization of the IoT is entirely dependent on making the proper decisions. However, the conventional methods for processing voluminous IIoT data are not qualified. Hence, DL is indispensable for making the intended inferences through big IIoT data. Likewise, due to the advancement of sensor technology, various sensor resources such as sound, vibration, and current can be used to obtain appropriate inferences. Accordingly, the decision fusion theory can be used to make optimal decisions when there are multiple sources of information. Therefore, this paper proposes a method that combines one-dimensional convolution neural networks (1DCNNs) and the Dempster-Shafer (DS) decision-fusion method (DS-1DCNN) for decision-making on IIoT anomalies. According to obtained simulation results, this proposed method increases the decision accuracy and significantly decreases uncertainty. The proposed method was compared with long short-term memory, random forest and CNN models, which obtained better performance than these algorithms. The proposed method on the Mill dataset got an average recall of 0.9763 and an average precision of 0.9899, which is an acceptable and reliable result for decision-making.