Support-vector classification of downstream problem effects during physical product development and ramp-up

Abstract During manufacturing of products at low maturity levels (referred to as physical product development, or production integration), nonconformance problem solving constitutes a key activity to prevent failure cost, to accelerate time to market and to fulfill quality expectations. With increasingly complex products and shortened innovation cycles, it even becomes a competitive factor. However, empirical studies show that the accompanied increase in problem dynamics and complexity exceed the problem assessment capabilities of human experts. In this work we present a machine learning based approach, focusing on the early symptom-based anticipation of critical problem dimensions. This is accomplished through the support vector classification of general quality management data such as required by ISO 9001, with regard to historic target values from the order-to-delivery process. The approach is validated for an early anticipation of problem relevance regarding ramp-up and series production through a design science study within the automotive industry. It can be shown that a dedicated agent outperforms human problem appraisal not only in terms of efficiency but also effectiveness. Furthermore, a naturalistic evaluation confirms usability and acceptance of the developed artifact within the application domain.