Based On The Improved BP Neural Network Application In The Autocorrelation Process Control

In the contemporary era, the core of product competitiveness is product quality. Based on computer aided quality management monitoring is a common method of industrial production. Recently conventional control charts, CUSUM control charts and EWMA control charts are the most common monitoring methods, which are based on the assumption that the observed values are independent and identical distributed. However, in the continuous production process, most of the collected data is auto-correlative, and continue to use the common control charts for monitoring will be issued a false alarm, resulting in a large number of losses. Therefore, it is very important to use the correct method to monitor auto-correlated processes. At present, the adjustment of conventional control chart control limit, residual control chart are used for monitoring the auto-correlated processes, but the effect is not ideal.This paper studies the use of BP neural network with pattern recognition function. This method is not restricted by the independent and identical distributed data, and can directly use the trained network to predict whether the data is out of control. However, because the initial weights and thresholds of the neural network are randomly assigned, in the number of training given conditions, the initial weights and thresholds directly affect the training results of the network. Aiming at the above problems, this paper mainly does the following aspects of the work:(1) The basic theory of the conventional control charts and residual control charts are described, and by the experiment further analysis of the several methods in monitoring the advantages and disadvantages of auto-correlated processes are given.(2) Aiming at the influence of the auto-correlated of the observation value on the recognition ability of the control chart, we study the method steps that using BP neural network to monitor the auto-correlated processes, including the training data selection and network structure determination, and verify that it is feasible to use BP neural network for recognizing by an example.(3) In view of the shortcomings of the slow convergence of standard neural network, genetic algorithm is adopted to improve the convergence speed of BP neural network. Besides, we propose using genetic algorithm to optimize the initial weights and thresholds of BP neural network, monitoring using the optimized BP neural network can greatly improve the network prediction accuracy, and propose the way of using the forecast results to determine the offset point position.(4) By the numerical experiments, the recognition ability of several auto-correlated processes monitoring methods are compared in detail. The results showed that stronger recognition capability and more accurate locking offset point position control with method proposed in this paper.