Key points are not available for this paper at this time.
In the recent years, as an alternative of the traditional process quality management methods, such as Shewhart SPC, artificial neural networks (ANN) have been widely used to recognize the abnormal pattern of control charts. But literature show that it is difficult for a developer to select the optimum NN topology architectures in a systemic way, this kind of work was primarily done according to the developer's personal experiences and could not get desirable effect. This paper proposes to use probability neural network (PNN) to recognize the six kinds of control chart patterns (i.e. normal pattern, upward/downward mean shift pattern, upward/downward trend pattern, cyclic pattern) to improve the design effect of pattern recognition. Numerical simulation result shows that PNN has not only the feature of simpler topology structure but also the higher pattern recognition accuracy and faster recognition speed. As the PNN pattern recognition method can get the optimum classification effect in terms of the Bayesian criterion, it is a comparable way between different manufacturing processes and suitable to be generalized as an industry criteria.
Cheng et al. (Tue,) studied this question.