An Architecture to Integrate Digital Twins and Machine Learning Operations

Digital twins (DTs) and artificial intelligence (AI) have attracted significant attention and interest from research and industry communities in recent years. DTs have significant potential to support the widespread adoption of AI, in particular machine learning (ML), for more intelligent and autonomous decision making. The adoption and integration of ML into production environments are often difficult and complex, which results in ML models often underperforming relative to the development and testing environment. Machine learning operations (MLOps) overcome this underperformance by continuously deploying, integrating and (re)training ML models in production environments. This paper introduces an architecture that integrates a system of digital twins with an MLOps platform. This integration is achieved using a tight coupling between a DT (part of the DT system) and an MLOps counterpart (part of the MLOps platform). This tight coupling has the benefit of enabling each DT to have unique ML workflows, which reduces complexity within a system of many DTs and many ML workflows. The paper further introduces a Type DT Aggregate which is the aggregation of DT Instances (DTIs) of the same type. This improves scalability in large systems where many DTIs require ML.