Quality-critical production processes influence the final product quality significantly. There is an increasing demand for highly accurate product quality monitoring systems to reduce time- and cost-intensive product inspections. This paper proposes a data-driven product quality monitoring system for execution on devices with low computational power in production environments. Particularly for quality-critical processes, the developed monitoring approach promises to deliver high accuracy. It is based on building a regression model to describe a quality indicator dependent on sensor data. The developed approach is addressed to highly variable production processes with a minimal set of reference data in which the quality assessment must be available in a timely manner. This small set of reference data is used for model building. Therefore, it is estimated that the regression model tends to deliver limited predictive power. The authors consider semi-optimal models explicitly and design a quality classifier sensitive to the prediction model's predictive power. The presented approach is evaluated on historical data for a use case from powder metallurgy. Furthermore, the approach for product quality monitoring under consideration of semi-optimal regression models provides a one hundred percent accuracy in an exemplary test case. It is shown that the model's predictive power in quality monitoring must be considered to design monitoring systems with high accuracy.     «

Quality-critical production processes influence the final product quality significantly. There is an increasing demand for highly accurate product quality monitoring systems to reduce time- and cost-intensive product inspections. This paper proposes a data-driven product quality monitoring system for execution on devices with low computational power in production environments. Particularly for quality-critical processes, the developed monitoring approach promises to deliver high accuracy. It is...     »