Automatic Recognition and Analysis of Statistical Process Control Patterns Using Neural Networks

Abstract

Control charts are statistical process control tools notable in analyzing, if the operation of newlinea process is occurring in the intended mode or is being interrupted by unnatural patterns. newlinePatterns displayed on control charts provide the details of the ongoing process. The 2 main newlinepurposes of the control charts in order to maintain the process quality are; to supervise the newlinestability of the process and serve as an analysis tool. A built-in common cause variation is newlinedisplayed once the process in found stable and in control. The control of a process is newlineestablished based on the subsequent variability (within limits) of the process on the basis newlineof prior data. The instability of the process is indicated by the display of non-random newlinevariation of external factors (special cause variation). A process is said to be in control, newlinebased on the variability (within limits) of the process on the basis of prior data and newlinedetermines the behavior of the process in the future. The instability of the process is newlineindicated by the display of non-random variation of external factors (special cause newlinevariation). Elucidation of the process data which comprises of pattern recognition tasks newlinecontinues to be complex. Identification of the presence of unnatural pattern in the process newlinerequires well trained and proficient quality control personnel. newlineControl Chart Pattern Recognition is a critical task in Statistical Process Control (SPC). newlinePresence of abnormal patterns in control charts can be linked to certain causes impacting newlinethe stability of the process severely. A number of studies on CCPR have been carried out newlinekeeping in view the aims and hypotheses. These studies also aim at improving the newlinecompetency and simplifying the recognition model. The implementation of various CCPR newlineis evident in industrial activity so as to control the process parameters and achieve the newlinedesired goals. newlineA control chart pattern recognition system which has been automated to work with newlineefficacy will be able to bridge the gap and provide continuous and unbiased interpretation newlineof CCPs by reducing the number of false alarms and enhancing the role of control charts. newlineKeeping this goal in view many techniques for CCP recognition have been proposed and newlinedeveloped. A significant improvement in the automated recognition of control chart newlinepatterns can be seen with the implementation of computational intelligence. newline newlineThe following research study proposes to design a cost effective, multi-process CCP newlinerecognition using a single recognition model. .Here the recognition model has been newlineimplemented for the Feed forward neural architecture alongside gradient descent learning. newlineThe effects of the faults in the recognition model which is crucial with respect to real time newlineapplications and the efficiency of CCP recognition have been quite often disregarded in newlinemany of the past studies. newlineThe current study also proposes to engage a stochastic gradient technique based adaptive newlineversion of the radial basis function neural network to map the pattern features of the CCPs newlineto recognize their belonging class in several categories. . Adaptiveness is given over the newlinespreadness and centers of Gaussian basis function appearing in the hidden nodes of the newlineradial basis function neural network. The mixture of several abnormal patterns and the newlinenormal abnormalities in patterns are also taken into consideration to capture the most newlineunfavorable conditions of the abnormalities at real time. newlineThe merits of implementing the proposed method are achieving; very high recognition newlineaccuracy, minimum error in learning and generalized performance with a small training newlinedataset in control chart pattern recognition. An improvement in the recognition newlinegeneralization of the control chart patterns with simple design and high level of decision newlineconfidence can also be seen. The performance of individual recognizer is exposed to the newlinequality and quantity of training data and final outcomes that often have dynamics in newlineoutcomes quality. So in order to overcome this issue an ensemble based technique is newlineimplemented that integrates the individual quantitative decision outcome of the newlinerecognition. newline