Reliable counterparts: efficiently testing causal relationships in digital twins

The lack of testability of digital twins poses several difficulties when developing reliable systems. Intricate models complicate the definition of comprehensive testing criteria, and physical couplings make obtaining test data an arduous task. To alleviate these challenges, we explore the use of causal inference based testing and propose a technique to allow for correct behaviour of digital twins to be captured in causal diagrams, which are then tested with an efficient data set through the use of counterfactuals. We explore a motivating example of a robotic arm to show how this technique can confirm known causal relationships in a system, and even uncover a fault in the system which caused dangerous behaviour. Our technique localised this erroneous behaviour to a single causal relationship between two variables. Having shown this technique works with a case study, we explore its limitations and the challenges when approaching other industrial applications.