Mass production manufacturing requires advanced quality control procedures in order to satisfy the increasing demands for productivity and product quality. Automated and performed online, these procedures are often based on vision systems and predictive models, while their design highly depends on the manufacturing process specificities, product characteristics and customer requirements. This talk presents a collaboration between academic and industrial teams in designing a quality control procedure for automotive components. The task is to monitor physical properties of mass-produced parts efficiently and reliably, and detect defects. We apply data mining to build models for predicting product properties from their digital images. In addition, we tune both the image processing and model building steps using evolutionary computation. The talk gives an overview of this case study and exploits it to provide general observations on establishing and maintaining an academic-industrial partnership.