Complex and big data handling and monitoring through machine learning towards digital-twin in high precision manufacturing

The digital transition faced in the Industry 4.0 framework is reshaping the complexity and volume of data used for discrete manufacturing quality monitoring and modelling. Advanced manufacturing methods are used to produce more and more complex products, while the wide adoption of in-line sensing tools enable big and fast streams of sensor data while the part is being produced. The result is a novel level of quality data complexity, which gives rise to industrial challenges, namely the lack of appropriate statistical process monitoring methods, the difficulty to manage such high dimensionality of in-process and post-process data, and the difficulty to deal with multi-sensor data stream in an efficient and effective way. This paper presents novel solutions based on machine learning (ML) algorithms suitable to fuse and make sense of big multi-channel data gathered in discrete manufacturing applications to anticipate the detection of anomalies, and to enhance the quality of the products, towards the implementation of digital-twin solution in production environments. A real case study is presented in the field of milling operations via flexible systems. ML methods are compared against benchmark statistical techniques to highlight when and to what extent the actual ML potentials may help bridge the gap between scientific research and industrial deployment, pointing out current barriers and challenges in real production environments.