Data-driven Digital Twin approach for process optimization: an industry use case

In this paper we present a novel approach for the process improvement based on the data-driven modelling. The idea is that by performing Big data analytics on the past process data we can model what is (statistically analyzed) usual/normal for a selected period and check the variations from that model in the real-time (as Six Sigma requires). Additionally, these data-driven models can support the root- cause analysis that should provide insights what can be eliminated as a waste in the process (as Lean requires). However, due to the above mentioned variety and volume of data, the analytics must be a) robust - dealing with differences efficiently and b) scalable - realized in an extremely parallel way. We propose a novel method for process control that uses big data analytics approaches to deal with the multidimensionality and the large size of the process space. In order to realize this idea we develop a new concept of self- aware digital twins which are able to reason about own behaviour and react if needed. Indeed, we revolutionize the concept of digital twins by extending their "virtual replica" (of physical objects) nature into "digital self-awareness" of physical objects (assets, systems), leading to the new generation of digital twins, so called self-aware DTs, which can "reasons" about the behaviour of an object (and not only mimic it) and actively participate in its improvement. We present the outcomes from the case study related to 3D laser cutting process.