Corrective actions of product service failures via surrogate modelling of dimensional variations

Corrective actions of product service failures such as warranty and No-Fault-Found is important. Field performance characteristics of products are related to Functional Requirements (FRs) such as gaps and inclinations between mating parts in mechanical assemblies. FRs are achieved through Design Parameters (DPs) and Process Variables (PVs). Root causes analysis (RCA) of a service failure identifies faulty FRs and translates them as geometric features in the assembly CAD. Next,corrective actions adjust DPs and PVs to minimize production yield of the faulty geometric features. However there are two major challenges in doing corrective actions: (i) RCAof service failures mayidentify FRs which arenot modelled during design due to incomplete information;and(ii) lack of information on critical DPs, PVs and their interactions which affect variations in faulty FRs. Therefore there is need for a generic methodology which allows modelling of aFR, which is translatable as a geometric featurein CAD,in terms of critical DPs or PVs, irrespective of whether the FR is present or missing in original design. This research proposes to develop surrogate models of faulty FRs, which are translatableas geometric features in CAD.The surrogate models expressfaulty FRs as an analyticalresponse of dimensional variations of critical DPs. Machine-learning based method trainsthe surrogate models from data generated by Variation simulation analysis. Based on the closed-form relation of FRs and critical DPs, given by the surrogate models, a two-step design adjustment process is proposed: (i) Mean shift of DPs; followed by (ii) tolerance reduction of DPs to minimize production yield of faulty FRs. Cost of warranty and cost of tolerancing are considered to find optimal mean shift and tolerance reduction ofDPs.The proposed methodologyfor corrective actions of service failures is demonstrated by an industrialcase study ofautomotive ignition switch