Validation Framework of a Digital Twin: A System Identification Approach

System Validation;AI-enabled systems;Model;Digital Twins;System Identification

The constant improvement and developments in Artificial Intelligence/Machine learning models coupled with increased computing power have led to the incorporation of AI/ML for simulating learning and problem-solving in simple and complex engineering systems. However, the advent of AI/ML-enabled systems possesses latent uncertainty and unpredictable characteristics compared to traditional systems. This reality challenges engineers and industry stakeholders who care about ensuring the right AI-enabled systems are built (system validation). Digital Twins is an excellent example of such AI-enabled systems whose system validation has not been well-researched. This study delves into existing research and frameworks for validating Digital Twins and proposes a novel model-centric validation framework based on system identification techniques. Since Digital Twins are data-centric, system identification offers an intuitive approach to uncovering the system dynamics of physical assets' data, which Digital Twins aim to replicate, monitor, and update for structural health monitoring and control, which can further help validate Digital Twins. As a case study, we apply this model-centric validation framework towards partially validating a Digital Twin for a single-heat-pipe test article in a Microreactor Agile Non-nuclear Experimental Testbed demonstrated at a national laboratory last year. The system identification method helped identify the best mathematical process model that best represents the dynamics of the heat pipe and provides a pathway towards improving future digital twin ML prediction capabilities with a promise of finally validating future ML forecast datasets for this heat pipe on the identified process model to complete the system validation process. The outcomes of this study will help improve trust and system-level assertion for Digital Twins in practice towards sustaining the operational health of physical assets for various industry applications.