Worker skill assessment in human-centric assembly systems

Abstract

Accurate assessment of human worker skills is essential for human-centric automation in modern manufacturing systems. However, modelling these skills remains challenging due to variability in human performance and limited high-quality open-access datasets. This study proposes an approach for evaluating individual worker skill levels in assembly systems and assessing the reliability of internal company metrics. It introduces worker-specific learning curves that capture individual skill progression through an exponential model that incorporates three parameters: b, representing worker-specific learning rate, xe, quantifying previous experience and, c, correcting inaccuracies in the NMV (Nominal Measured Value) calculations or pre-training effects. An algorithm based on the trust-region method is used to fit the model using task execution data, effectively capturing individualized learning trajectories and distinguishing skill progression across workers. The proposed approach was validated using a real dataset collected from an assembly line over two weeks, where four workers performed tasks at both manual and collaborative workstations. As this dataset can be seen insufficient, a Moving Block Bootstrap (MBB) stability analysis was conducted to assess the robustness of the fitted learning curve parameters under resampling. The MBB results show partial stability, with some parameters remaining consistent under resampling while others vary more substantially. Comparison with company-reported skill levels reveals notable discrepancies, highlighting the need for improved assessment methodologies. Furthermore, the proposed approach was benchmarked against classical learning curve models. The dataset is released as open-access, enabling transparent and reproducible research and fostering advancements in skill modelling.