This paper investigates the application of Machine Learning (ML) techniques for the optimization of geometric alignment processes in industrial assembly, with a focus on automotive headlamp adjustment. The research objective is to evaluate the potential of data-driven models to enhance process accuracy, reduce variability, and minimize manual intervention. A diverse set of regression-based ML methods—including linear, kernel-based, tree-based, ensemble, and interpretable piecewise linear regression—is systematically compared. Models are trained and validated on real-world production data using 10-fold cross-validation and separate test sets. The results show that several ML models achieve high predictive performance, particularly for one of the three alignment targets, while prediction accuracy is more challenging for the other targets due to greater geometric variability. The study highlights both the benefits and limitations of applying ML in data-constrained assembly contexts. Beyond the specific automotive use case, the findings demonstrate the broader potential of adaptive, data-driven modeling to support intelligent process control in smart manufacturing environments.
Strasser et al. (Thu,) studied this question.