A study of the learning of three-layer back-propagation pattern recognizers for statistical quality control charts

Summary form only given, as follows. Back-propagation pattern recognizers are developed to identify unnatural patterns exhibited on statistical quality control charts. These unnatural patterns can provide valuable information for real-time process control. The selection of training patterns is iteratively determined by a training refinement procedure. The learning speed of these recognizers is studied by the rate of convergence and the actual training time. Nine different network configurations were investigated. The actual training time was characterized by known parameters such as the number of processing elements in each layer of the network, and was illustrated by computer processor time as well. The results indicate that the training time is on a manageable and affordable scale for problems of such magnitude