Automating Digital Twin Creation for Humancentric Manufacturing Systems

In today's industrial landscape, manufacturing systems face challenges that have significant implications for productivity, efficiency, and overall performance. The growing complexity of these systems and the need for optimization, efficient resource allocation, and adaptability demand innovative approaches. One solution that has gained prominence are digital twins (DTs), which are virtual replicas of physical manufacturing systems that can address the challenges faced by modern manufacturing systems. DTs enable methods like predictive maintenance and simulation-based optimization. Consequently, companies have a strong motivation to develop DTs for their manufacturing systems. Our research focuses on an especially challenging case, i.e., humancentric manufacturing systems. These involve the active human participation, often driven by intrinsic motivation rather than strictly adhering to predefined protocols. This poses additional challenges in automating the creation of their DTs. On one hand, human behaviors introduce specific uncertainties, which must be appropriately considered in the model; not only are production cycles never identical, but factors like motivation, well-being, and energy levels constantly fluctuate. On the other hand, the processes of collecting data become more complex as human workers require additional hardware, such as wearables, to effectively gather relevant data. Consequently, there currently may be a significant imbalance in the data, with sparse representation for the human-related steps and abundant data available for the machine-related ones. In addition, the collection of worker-related data may lead to privacy concerns that will need to be addressed as well. Due to the aforementioned challenges, most DTs for humancentric manufacturing systems are currently created manually, which is a labor-intensive and time-consuming process. Moreover, the need to adapt production based on economic factors and derived goals leads to continuously adjusting the underlying models, significantly increasing maintenance costs. However, many companies possess large amounts of machine-related manufacturing data that can be leveraged to automate model extraction processes. Process mining is one such technique that automatically extracts underlying process flows in manufacturing systems from their event logs. Extracted process flows form the basis for the machine-related part of the DTs. To complete DTs, the human-related processes need to be integrated. This integration starts by defining the goals of the human-related portion of DTs and subsequently identifying or adjusting the relevant performance metrics. Our goal is to develop a general framework and methodology that links these performance metrics to data streams to enable data-driven DTs for humancentric manufacturing systems.