Studying Of Manufacturing Process Quality Control Theory And Methodology Based On Intelligent Learning Model

The continuously stable product quality is an important and key objective in industry, and the manufacturing process quality control is one of the important factors that ensure final product quality and high productivity. Due to their important significances in theory and industrial practice, process quality control has been widely studied over the last a couple of decades, and continues to attract researcher interests both in academia and industry. Statistical process control (SPC) is very useful in maintaining an acceptable and stable level of quality characteristics in manufacturing. But SPC is unable to implement process status quantification and visualization, recognition of out-of-control sources, information fusion for high dimension dataset. Intelligent learning models have excellent noise tolerance, nonlinearity, information fusion, and learning capability in real time, requiring no hypothesis on statistical distribution of monitored measurements. These important features make intelligent learning models promising and effective tools that can be used to improve data analysis in manufacturing quality control applications. Eventually, intelligent learning models effectively solve these problems existing in SPC, and help to realize intelligent quality control for manufacturing processes. The typical manufacturing processes (i.e., independent and discrete process, auto-correlated process and multivariate process) and complicated multi input and multi output (MIMO) processes are studied in this research, which aims to provide general theories and methodologies of manufacturing process quality control based on intelligent learning models. This research lays a theory and practice foundation for applications of intelligent learning models in manufacturing process quality control.1. The monitoring of manufacturing process status is as the first research content in this research. We take the lead in proposing some intelligent monitoring models that provide quantification and visualization of manufacturing process status. In this research, a novel control chart called minimum quantifying error (MQE) based on self-organization mapping (SOM) is proposed to provide a comprehensible and quantitative assessment value for current process status. The modeling method for SOM is based on data-driving and unsupervised learning for the given dataset with complicated distribution. Thus, MQE control chart based on SOM significantly improves its utility in real-world applications in comparison with some monitoring models based on supervised learning. The usability, validity and universality of MQE chart are evaluated under the assumption that the predictable abnormal patterns are not available. The experimental results show that MQE control chart has the potential to become an effective monitoring tool for process quality control. Moreover, the studying of unsupervised learning-based monitoring model lays a foundation for the applications of other unsupervised learning models in process quality control. 2. This research further proposes a process monitoring model based on adaptive Gaussian mixture model (AGMM) for time-varying processes, and uses log likelihood of GMM as a comprehensible and quantitative assessment value for current process status. Some novel adaptive updating schemes like forgetting-factor, and component split and merge are proposed for AGMM, which aims to build an adaptive monitoring model for complicated time-varying processes. The problem that the traditional control charts with fixed model are unable to monitor time-varying processes is solved by AGMM-based monitoring model successfully. This research further approves the special capability of unsupervised learning models in monitoring complicated manufacturing processes.3. This research takes the lead in using an artificial neural network (ANN) ensemble for recognition of out-of-control sources in multivariate manufacturing processes. A selective ANN ensemble algorithm (named DPSOEN) is developed to improve the engineering utility and recognition performance. Through integrating MQE control chart and DPSOEN, a hybrid learning model is established for monitoring and diagnosis of out-of-control signals in the multivariate manufacturing processes. The immediate abnormal warning and location of the abnormal source(s) can greatly narrow down the set of possible assignable causes, facilitating rapid analysis and corrective action by quality operators.4. A knowledge discovery algorithm based on genetic algorithm (named GARule) is proposed for extracting the important process knowledge existing in the relationship between the process input parameters and the output product quality. Moreover, a seamless integration of GARule and knowledge-based ANN (KBANN) is realized for on-line intelligent monitoring and diagnosis of the manufacturing processes. The proposed hybrid learning model is capable to provide abnormal warnings, reveals assignable cause(s), and helps operators optimally set the process parameters for MIMO processes. This research lays an important foundation for applications of intelligent hybrid leaning models in complicated manufacturing process quality control.This research is an important part of modern quality control engineering. The applications of intelligent learning model-based quality control models will improve process capability, solve some problems existing in regular control charts, improve product quality, reduce manufacturing cost, and improve stability and efficiency of the manufacturing system. Moreover, the studying of intelligent learning-based quality control models introduces new theory and methodology for regular process quality control theory. This research also lays a foundation for the applications of others intelligent learning models in manufacturing process quality control.Acknowledgement This dissertation is supported by the National Naturaral Science Foundation of China (No.50675137), the Program of Introducing Talents of Discipline to Universities (No.B06012), the Program for New Century Excellent Talents in University of China (NCET 2006), Excellent Doctoral Dissertation Foundation of Shanghai Jiao Tong University.