A sample path-based method for the digital twin prediction update synchronization problem of unreliable production lines

The adoption of IoT technologies enables easy and cheap data collection and more efficient digital twin operations. The available data are used to align the digital twin with its physical system. However, too frequent information exchange between physical and digital systems leads to high latency and costs and affects production decisions. The digital twin synchronization problem determines whether or not to synchronize the digital twin at each observation period depending on the observed state of the physical system. Existing approaches provide solutions for the prediction update synchronization problem, but they can be applied only to simple systems. In this study, we propose a sample path-based method to solve the prediction update synchronization problem for unreliable production lines composed of multiple machines and finite buffers. The method requires estimating the performance measure for each synchronization decision by partially observing the state of the system. An optimal state-dependent synchronization policy is determined based on the observed state to balance the prediction bias and the synchronization cost. The results show that observing the state of the most crucial machine in the production line, i.e., the bottleneck machine, rather than fully observing the state of the system, is efficient for solving the problem without requiring a very long sample path.