Defect prediction models to improve assembly processes in low-volume productions

Abstract

Assembly processes in low-volume productions, i.e., single-units or small-sized-lots, are often characterized by a high level of customization and complexity. As a consequence, the scarcity of historical data and the difficulty in applying standard statistical techniques make process control extremely challenging. Accordingly, identifying effective diagnostic tools plays a key role in such productions. This paper proposes an innovative method for identifying critical workstations in assembly processes based on defect prediction models. Starting from the level of complexity in terms of assembly process and design, the method allows identifying the workstations whose defectiveness deviates, at a certain confidence level, from the predicted average value. Once the causes leading to significant nonconformities have been detected, appropriate corrective actions may be promptly undertaken to improve the process. An example of implementation of the method in wrapping machines production is presented and discussed.