Control Chart Patterns Detection Using COA Based Trained MLP Neural Network and Shape Features

Statistical process control (SPC) is widely applied as a potent tool to measure, recognize, analyze and interpret process data to enhance the quality of products and service by detecting instabilities and justifying possible causes. In this paper, fast and accurate system is proposed to detect the control chart patterns (CCP). The proposed method includes three main parts: the feature extraction, classifier and training parts. In the feature extraction part, we used shape features as effective inputs. The dimension of input vector is reduced from 60 to eight by using these features. In the classifier part, we used multilayer Perceptron neural network (MLPNN). In order to improve the MLPNN performance, we used cuckoo optimization algorithm (COA) to train the network. In test stage, 10-fold cross validation method was applied to the synthetic control chart time series dataset to evaluate the proposed system accuracy. The recognition accuracy of proposed system is 99.21%. This research demonstrated that the proposed hybrid system can be used to obtain fast automatic detecting systems for control chart patterns.

کلیدواژه ها:

COA ، CCP ، Training ، Feature Extraction