On the Use of the FuzzyARTMAP Neural Network for Pattern Recognition in Statistical Process Control using a Factorial Design

Abstract

Time-series statistical pattern recognition is of prime importance in statistics, especially in quality control techniques for manufacturing processes. A frequent problem in this application is the complexity when trying to determine the behaviour (pattern) from sample data. There have been identified standard patterns which are commonly present when using the X chart; its detection depends on human judgement supported by norms and graphical criteria. In the last few years, it has been demonstrated that Artificial Neural Networks (ANN's) are useful to predict the type of time-series pattern instead of the use of rules. However, the ANN control parameters have to be fixed to values that maximize its performance. This research proposes an experimental design methodology to determine the most appropriate values for the control parameters of the FuzzyARTMAP ANN such as: learning rate (β ) and network vigilance (ρ a , ρ b , ρ ab ) in order to increment the neural network efficiency during unnatural pattern recognition.

SPC without Control Limits and Normality Assumption: A New Method

“…The upward trend pattern was generated with d > 0.1σ and for the downward shift d < −0.1σ. A detailed explanation of the simulation and the selection of the network parameters it is given in our previous work [11]. The best parameters values determined after the experimental design are shown in the Table 1.…”

“…To overcome this limitation, a novel method to recognise and analyse statistical quality patterns using the Fuzzy ARTMAP (FAM) Artificial Neural Network (ANN) is proposed. The FAM network parameters are determined off-line using experimental design and the Monte Carlo method which constitutes a novel method to increase the FAM efficiency eliminating the trial and error procedure commonly used [11]. During testing, the FAM Learning parameters are selected automatically depending if special or non-special pattern is encountered.…”