Pattern recognition of control charts using artificial neural networks—analyzing the effect of the training parameters

Abstract

In this paper, we have utilized artificial neural networks (ANN) for pattern recognition of the most common patterns which occur in quality control charts. After detecting such patterns, it is possible to relate these patterns to their causes. This could find extreme importance for on-line quality monitoring and on-line trouble shooting. It could be possible to detect problems before they become serious and the operator has to shut the line down or the process may produce defective parts. In this work, we have attempted to explore the effect of the training parameters on the performance of the neural network. The training parameters are important because they emphasis the required performance and the accuracy required from the neural network. A resolution IV fractional factorial experiment is utilized to explore a portion of the range of selected parameters to obtain better performance of the neural network. The results showed that many parameters usually assigned by experience such as minimum shift, shift range, population size and shift percentage, have significant effect on the performance of the ANN, while others such as network size and window size do not have major significance on the performance of the net.