Towards Intelligent Monitoring and Control of Industrial Internet of Things Deployments with Causality-Aware Digital Twins

DTs are expert-designed digital replicas of the associated Physical Assets (PAs), built using top-down software engineering approaches. However, modeling the expected system behavior at design time and understanding it at runtime are both hindered by the cyber-physical complexity and heterogeneous, interdependent variables of IIoT systems. In this paper, we propose to augment DTs in a data-driven manner by learning a causal model of the key variables involved in PA and DT interactions. Our methodology applies causal discovery techniques to refine DT models based on data collected during PAs operation. The resulting 'Causal DT' enables (i) discovery of hidden causal relations, (ii) translation of insights into actionable knowledge, and (iii) planning of interventions to steer the system toward desired outcomes. We validate our approach using real hardware that emulates a scaled-down industrial Microfactory with DT-enabled machines.