Characterizing Sequential Patterns of Human Behavior in Advanced Manufacturing

Abstract Understanding sequential human behavior in manufacturing is essential for improving productivity, safety, and human-machine collaboration. However, little is known about how the temporal structure of these behaviors varies across time scales or how such patterns can be systematically modeled to support adaptive, human-centered manufacturing systems. This paper investigates the temporal structure of human interactions involved with Wire Arc Additive Manufacturing (WAAM), which serves as a representative context for advanced manufacturing environments. Using a large-scale dataset of annotated human activity in an advanced WAAM environment, we apply generative sequential models to systematically analyze how behavioral predictability and structure vary with sampling frequencies. This analysis reveals that human behavior in WAAM is highly sequential and temporally persistent, with a small number of latent modes sufficient to describe complex workflows. We show that finer temporal resolutions capture deterministic self-transitions, while coarser resolutions uncover broader procedural patterns. These insights show the potential value of classical generative models as interpretive tools in advanced manufacturing and provide a foundation for designing adaptive manufacturing systems. Future work will extend this framework to other manufacturing contexts and explore advanced sequence modeling approaches to further understand human behavioral patterns in complex industrial environments.