The dissertation presents neural network models for application in inline Statistical Process Control. It addresses three key pitfalls that previous research efforts in this area have encountered. Firstly, previous models were designed to detect only one pattern on the control chart and fail to detect multiple patterns that occur concurrently. Secondly, the applications restrict the number of data points that can be plotted on a control chart to the number provided by the NN architect. Thirdly, the number of training samples needed to train the NN is often huge. The model presented in this dissertation addresses these problems. In addition, the model introduces the use of neural networks in developing shorter and non-biased intervals for process capability estimators. |