Multi-stream big data mining for industry 4.0 in machining: Novel application of a Gated Recurrent Unit Network

In Industry 4.0, the availability of signals from multiple sensors stimulates the investigation of novel quality monitoring and prediction methods. This paper tackles the in-line machining process monitoring by exploiting big data in the shape of multi-stream complex signals, eventually containing degradation and tool wear signatures. The proposed novel solution is fed by real-time multichannel data to identify anomalous states in machining applications. We investigate the effectiveness of a category of ANNs specifically conceived to predict process patterns based on time series of sensor signals, i.e., the Gated-Recurrent-Unit-Network. A real case study shows the efficiency of the proposed solution in predicting wild, complex and drifting patterns, typical of real productions, highlighting its provided benefits for in-line big data mining in industrial applications.