Singularity Issue in Dimensional Variation Diagnosis of Multi-Station Assembly Processes

This paper presents a new method of diagnosing variation components of process error sources in a manufacturing system. Quite often in a complex multi-station assembly system only limited numbers of sensors are present due to which complete information of fixture errors is unavailable. This makes the system of variance components singular and not solvable by using regular least-squares estimation. The method suggests reformulation of the original error propagation model into a variation relation by using a matrix transformation. With the development of a new variation estimator and its diagnosability condition, some singular systems that are not diagnosable using traditional least squares methods become diagnosable. Difference between the new approach and the traditional approaches has been elaborated. Modified procedures of the new estimator are also presented to enhance its estimation performance. The idea is presented in the specific context of panel assembly processes, but the application of the idea should not be limited therein. Conclusions can be extended to general discrete-part manufacturing processes where fixtures are extensively used to ensure dimensional accuracy of the final product.

Singularity Issue in Dimensional Variation Diagnosis of Multi-Station Assembly Processes Conference Paper