Automatic Model Generation and Data Assimilation Framework for Cyber-Physical Production Systems

Abstract

The recent development of new technologies within the Industry 4.0 revolution drives the increased digitization of manufacturing plants. To effectively utilize the digital twins, it is essential to guarantee a correct alignment between the physical system and the associated simulation model along the whole system life cycle. Data assimilation is frequently used to incorporate observation data into a running model to produce improved estimates of state variables of interest. However, it assumes a closed system and cannot handle structural changes in the system, e.g., machine breakdown. Instead of combining the observation data into an existing model, we aim to automatically generate the model concurrently with the data assimilation procedure. This can reduce the time and cost of building the model. In addition, it can generate a more accurate model when sudden operational changes are not reflected at the higher planning levels. Component-based model generation approach is used with the application of data and process mining techniques to generate a complete process model from the data. A new data assimilation method is proposed to iteratively generate new models based on the arrival of further data. Each model is simulated to obtain the system performance, which will be compared to the real system performance to select the best-estimated model. Identical twin experiments of a wafer-fab simulation are conducted under different scenarios to evaluate the feasibility of the proposed approach.