System-level Quality Planning and Diagnosis for Complex Multistage Manufacturing Processes.

The performance of a multistage manufacturing process (MMP) can be measured by quality, productivity and cost. All these measures are influenced by the variation of the key product characteristics (KPC’s). To remain competitiveness, variation of KPC’s should be reduced to ensure efficient delivery of quality products. However, the unprecedentedly high requirements on quality make variation reduction a very challenging problem. To reduce KPC variation, massive data are generated and collected from different phases of product realization, including quantitative data and qualitative data. The heterogeneous data poses great challenges to traditional quality assurance methodologies, which emphasize monitoring of manufacturing processes but provide limited diagnostic information. Taking advantage of readily available data, this research focuses on system-level methodology for effective quality assurance of MMP’s in the following aspects: (i) A mathematical variation propagation model is developed to describe the process induced variation and its propagation along multiple manufacturing stages. The generic formulation makes it capable to model a wide variety of processes where 3-D dimensional variation is of interest. The modeling concept and techniques can be extended and applied in early phases of product realization to effectively evaluate product and process design alternatives. (ii) A quality assured setup planning methodology is developed to address the quality assurance in the process design phase of product realization. Setup planning is formulized as an optimal sequential decision making problem and is solved based on analytical evaluation. This research creates the potential for future works on concurrent development of system-level setup and fixture planning. The setup planning results can be further utilized for process diagnosis in the manufacturing phase of product realization. (iii) An engineering-driven factor analysis methodology is developed to diagnose an MMP based on qualitative rather than quantitative representation of product/process interactions. By using the qualitative indicator vectors to direct the estimation of true spatial patterns from multivariate measurement data, the variation sources are identified. The diagnostic results are robust to unknown process changes. The proposed methodologies represent the initial research efforts in a general framework of unified methodology for quality assurance of MMP’s. Based on them, future research directions are identified and discussed.PhDMechanical Engineering and Industrial and Operations EnginUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/60659/1/jliuzz_1.pd