Why human actions matter in resilient production
Modern manufacturing must adapt quickly to shifting products, fluctuating volumes and changing materials. While digital tools support planning, shop floor reconfiguration still depends heavily on human expertise. Operators develop their own ways of adjusting machines, shaped by years of experience. These skills are difficult to document, teach or standardise, yet they are essential for resilient production. When expertise varies, reconfiguration becomes slow, inconsistent and prone to error.

A new approach to learning from expert behaviour
This recent research presents a method for capturing, analysing and predicting the sequence of human actions during machine tool (re)configuration. Using IMU motion sensors, eye tracking and video, the actions and intentions of an expert operator are recorded as they adjust a progressive forming press. The method then combines manual annotation with automated recognition based on principal component analysis and clustering models. This produces a clear, structured sequence of actions that can be compared, evaluated and stored as operational knowledge.

Turning motion data into transferable skills
The study shows that actions such as walking, picking up parts, screwing and placing components vary in duration, frequency and order across different reconfiguration scenarios. These variations highlight why inexperienced workers struggle to consistently follow the same process. By automatically predicting and annotating action sequences, the method lays the foundation for training tools that explain expert behaviour and support non-skilled workers in performing complex, unstructured tasks more reliably.

Towards human-centred reconfiguration support
Action prediction offers a pathway to future Operator 5.0 systems in which human expertise is captured, shared and augmented. Applied broadly, this approach could reduce errors, shorten changeover times and enhance the resilience of machine-intensive production environments.

This article is based on the peer-reviewed publication “Human Action Sequence Prediction for (Re)configuring Machine Tools”, published in Procedia CIRP (Volume 130, 2024) and available via ScienceDirect.