Quality Process Diagnostic Modeling Based On Data Mining

Along with the rapid development of modern manufacturing, the important ways forenterprise to win in the market is to improve its product quality and guaranteeing itsconsistency. While a significant method to guarantee the product quality is to control thequality in the producing process and diagnose the abnormal variable if necessary.However, only using traditional statistical process control technology can not be fullysatisfied with demands. In recent years, Scholars have been busy studying this hot issuescombining traditional statistical process control method with data mining and artificialintelligence technology. Concerned with the features of univariate and multivariate of thequality process, this paper has improved the traditional statistical process monitoringmethod in various degrees, and offered a series of diagnosing algorithms.The main tasks and contributions of the text are as follows.1. According to the characteristics of the complex producing process, we havecomprehensively analyzed the course of development of quality process control fromStatistical Process Control (SPC) to Statistical Process Diagnosis (SPD) even up toStatistical Process Adjustment (SPA). At the same time, we have expounded thecharacteristics and using rules of W.A. Shewhart control chart and HotellingT2controlchart respectively from different angle of univariate statistical process control tomultivariable statistical process control.2. Considering the problem that the abnormal features exist great similarity so thatsimple structure and high precision modeling cannot be achieved. An univariate abnormaldiagnostic algorithm with control chart pattern recognition based on kernel principalcomponent analysis and neural network has been proposed. Firstly, the kernel method wasused to translate the nonlinear feature into a higher dimensional linear feature space.Secondly this feature was projected to lower dimensional feature space. Finally the BPneural network classifier was introduced to identify the control chart pattern. This methodwas verified through stochastic simulate experiment. The result demonstrates that thismodel can cluster each control chart pattern effectively and improve recognition accuracy.3. Considering the characteristic that pluralism, nonlinear, strong correlation andcoupling characteristics in the actual manufacturing process. A novel approach for diagnosing the out-of-control signals in the multivariate process is described in this chapter.The proposed methodology uses the optimized support vector machines (support vectormachine classification based on genetic algorithm) to recognize set of subclasses ofmultivariate abnormal patters, identify the responsible variable(s) on the occurrence ofabnormal pattern. We used GA to globally optimize the kernel parameters and penaltyfactors of the Support Vector Machine (SVM) Multiple sets of experiments are used toverify this model, The performance of the proposed approach demonstrates that this modelcan accurately classify the source(s) of out-of-control signal, even outperforms theconventional multivariate control scheme such as BP and SVM.