A VAE Conversion Method for Heterogeneous Data Inputs to Create Uniform Outputs for Diagnosis

Abstract

With the development of Industry 4.0 and the Internet of Things, in order to provide a more effective and reliable monitoring system, industrial machinery is equipped with multiple sensors to form a multi-sensor system that provides high quality and large amounts of data for diagnosis. However, different components of machine may have different features, and the same type of sensor has different characteristic outputs due to different sensing positions, resulting in heterogeneous data. Therefore, it is necessary to establish a complex data preprocessing process to train an effective detection model. In addition, the trained model is immutable. When the configuration of the sensors is changed, a lot of time must be spent to retrain the model. It is difficult to efficiently adjust the model according to the actual situation. In view of data heterogeneity and model immutability of multi-sensor systems, this study proposes an architecture named two-stage anomaly detection system. In the first stage, the data conversion model called Converter is proposed in this paper to convert heterogeneous data to uniform, thus making multi-sensor diagnostic systems easy for training. The second stage considers the relationship among data input and integrates all the unified data for accurate diagnosis of the entire system. The experiment of this study uses the IMS rolling bearing data set to verify the effectiveness of this architecture. The results show that the two-stage architecture can effectively perform anomaly detection and help to deal with heterogeneous data and model variability in multi-sensor systems.