Machine learning models for Statistical Process Control must deliver not only high performance but also reliability, transparency, and workflow fit to earn trust in industrial use. Based on responses from industry professionals, this study identifies evaluation criteria shaping perceptions of ML tools in quality applications. Results show a clear preference for classification metrics over regression metrics, indicating limited understanding and perceived relevance of regression-based evaluation. Importantly, performance alone does not secure trust: users prioritize reliability, error minimization, and clear model behavior. Expectations for ML-based quality control often exceed those for manual inspection, with most users demanding that intelligent systems outperform traditional methods. The study also highlights the need to tailor digital tools to specific roles and workflows. To support adoption, we recommend emphasizing reliability and providing role-specific user interfaces. These findings add to research on ML integration in manufacturing and guide development of trustworthy industrial ML systems.
Mayer et al. (Tue,) studied this question.