Designing a Resilient Hybrid-Modeled Digital Twin for Reliable Manufacturing Processes

Digital twins expand production systems to include a virtual space for analysing, monitoring, and predicting production processes. Hybrid Modeled Digital Twins (HMDT), which combine physics-based and data-driven models, hold significant potential for improving key performance indicators in manufacturing processes. In a wide range of applications, hybrid models have shown their ability to, for example, improve environmental sustainability, reduce waste, and predict process outcomes. However, a common assumption in existing approaches is the availability of all model inputs for HMDT. In real-world scenarios, there may be disturbances in data transmission, such as incorrectly connected cables to machines or sensors, or failures in wireless communication. In such cases, traditional HMDT become incapable of providing instructions to the real system. This paper, therefore, introduces a novel concept: the design of resilient HMDT capable of delivering dependable feedback to real systems in the absence of inputs during short periods. This resilience enables the continued utilization of these models under restricted input conditions, albeit with a modest reduction in functionality. To exemplify this concept, we present a use case of a resilient HMDT in the field of industrial polishing of stoneware floor tiles.