| Due to worldwide competition, the achievement of high-quality products has become the key solution for manufacturers to survive in this dynamic global market place. Quality control plays an important in most industrial systems. Its role in providing relevant and timely data is vital to management for decision-making purposes. Statistical Quality Control (SQC) is the application of statistical methods to problems of interest to evaluate, establish, or verify the quality of a product. The basic area of SQC that has received the greatest attention in academia and the widest acceptance in industry is a method that uses statistical techniques to monitor and control product quality improvement; namely, Statistical Process Control (SPC).; SPC is a collection of tools that are essential to quality improvement activities. Central to SPC implementation is control charting. Control charting is a monitoring technique to identify unnatural behavior in a process. Shewhart control charts have been the most popular charts applied in practice for decades (Montgomery, 2000). When used properly, a Shewhart control chart's strength lies in its ability to distinguish specific disturbances from inherent variability in a process. A control chart can only indicate when a disturbance occurs; it cannot, by itself, identify the disturbance or the source of the disturbance.; In many industries, SPC techniques have become one of the most commonly used tools for improving quality. Among these techniques, control charts are the most important and frequently used due to their usefulness in identifying the presence of the special causes in the manufacturing process. To use control charts, samples of the products are drawn and measured, weighted, etc., during the manufacturing process, and the sample statistics are plotted on control charts. If special causes are present, the sample statistics are likely to plot outside the predefined control limits to trigger an out-of-control signal. Operators and engineers then search for the special causes and make necessary adjustments to bring the process back to an in-control condition.; In this dissertation, Neural Networks (NNs) were used as a tool to automate the implementation of control charts and to recognize process variability in a timely manner and predict the changes in the process before going out-of-control. The pattern recognizer, the Backpropagation Neural Network (BPNN), was chosen based on the advantages of layered networks. Automate the decision-making process by implementing and testing various forms of statistical and neural pattern recognizers. Various experiments were performed to select the appropriate NN structure for the control chart. Simulations and various experiments were conducted to determine the combination of parameters to optimize the BPNN model. The resulting BPNN model proved capable of detecting various patterns: sudden shifts (upward or downward), linear trends (upward or downward), stratification variations, mixture, systematic variables, and cycle to a level not previously accomplished in any other published work. |
