| The use of Neural Network (NN) models have recently been recommended as statistical quality control (SQC) tools. The advantages of NNs, particularly the robustness of the nonlinear modeling abilities, are appealing to quality control practitioners for use in process monitoring. Advances in computing power have also made the Neural Network Control Charts (NNCC) an alternative SQC technique.;The systematic Design of Experiment (DOE) methodology is employed to find near optimal NN topology for NNCC for Poisson data. A (2k) full factorial design is implemented and supplemented as needed to investigate NN topologies. The effect of the following factors were investigated through a simulation study: the number of the inputs "n", the number of nodes in the hidden layer(s), the training data size, and in-control mean for shift range 0-3 sigma. The guidelines and steps of constructing the DOE study for the NNCC is given, along with an example.;The NNCC is compared to C-chart, X-individual chart and EWMA charts in terms of average run length (ARL), median run length (MRL), 5th percentile, and 95th percentile for their Run Length (RL) distributions. The performances of the charts are considered for negative binomial, discrete uniform, and the Poisson distributions. The NNCC run length is simulated, while exact run length distributions are used to calculate the RL moments for the C-chart and X-chart. The EWMA charts' performances are obtained using the Markov chain method.;The body of literature is summarized in this research project. The quality control tools are divided into two main categorizes: univariate charts and multivariate charts. Within each category, the types of shifts are extensively studied. The successes of NNCC over traditional charts are detailed, while the shortcomings of NNCC for discrete data such as Poisson distribution data are noted. |
