Surface Variation Modeling by Fusing Surface Measurement Data With Multiple Manufacturing Process Variables

Abstract

Controlling surface shape variations plays a key role in high-precision manufacturing. Most manufacturing plants rely on a number of multi-resolution measurements on manufactured surfaces to evaluate surface shapes and resultant quality. Conventional research on surface shape modeling focused on interpolation and extrapolation of spatial data using sampled measurements based on presumed spatial relationship over entire surface locations. However, the prediction accuracy is heavily restricted by the density of sampled measurements, preventing cost-effective evaluation of surface shape in high precision. New opportunities emerge for cost-effective high-precision surface manufacturing when the industry begins to extensively collect in-plant process information. This paper explores the opportunity by investigating strategies for fusing surface measurement data with multiple process variables. The fusion is achieved by characterizing the relationships between surface height and process variables using (1) linear regression based co-Kriging and (2) fuzzy if-then rules as well as considering spatial correlations. Under (3) Bayesian sequential updating frameworks, a generic surface variation model is updated sequentially using different process information. Case studies are conducted for comparisons and demonstrate the advantages of the fuzzy inference based spatial model.