Capability Analysis And Quality Diagnosis Of Multivariate Manufacturing Process

In manufacturing practice, the quality of a process or product is usually characterized by multiple correlated critical characteristic which are correlated. It is critical to assess the process capability and to diagnose the assignable causes in these cases in quality engineering. This dissertation mainly studies the multivariate process capability analysis (MPCA) and multivariate process diagnosis in manufacturing process with multivariate characteristics. Aiming at this objective, two multivariate process capability indices (MPCIs) are proposed to assess the process capability and a monitoring and diagnosis scheme is proposed and applied in multivariate manufacturing application.With the introduction of basic theories and methods of MPCIs and multivariate diagnosis approaches, the proposed MPCIs and the process monitoring and diagnosis scheme are proposed as follows:Firstly, a monitoring and diagnosis scheme is constructed based on statistical process control and multivariate quality diagnosis methods. The proposed scheme is well developed to detect the abnormal signal when the process is out of control, and to estimate the assignable causes which lead to the out-of-control of the manufacturing process.Secondly, two multivariate process capability indices based on principle component analysis (PCA) are proposed to estimate the process capability. Before measuring MPCIs, PCA is used to reduce the data dimension, and the tolerance of principle components (PCs) can be obtained. Then the MPCI proposed by Taam et al. is modified based on the PCs'tolerance, and the confidence interval is estimated through hypothesis test. Moreover, the estimation of non-conforming proportion on the basis of the probability density function of PCs is proposed to analyze the process capability. Case studies show that the two methods of MPCA are both convenient to assess multivariate process.Thirdly, the application of multivariate process control and diagnosis is studied in automobile manufacturing. The diagnosis approach based on support vector machine (SVM) is proposed to find the assignable causes after the multivariate control chart gives an alarm. The classification model based on SVM is suggested to detect the shift for each variable, while considering the effects of other variables. Therefore, different combinations of mean shifts of multivariate process are obtained and used to find the assignable causes. Case study shows that the identification of assignable causes is effective in diagnosing multivariate process.