Recognition And Application Of Control Chart Patterns Using Fuzzy C-Means Algorithm

As an important quality diagnostic tool in modern industrial production,abnormal fluctuation in manufacturing process can be marked by control chart.Control chart pattern recognition plays an important role in the effective monitoring of machining process and the diagnosis of abnormal source.When the enterprise needs to ensure the flexibility of the manufacturing system and make it run naturally,it is necessary to carry out real-time monitoring of the quality process.Then the abnormality in the production and processing process would be recognized in time,and finally achieve the goal of reducing the loss and cost.Firstly,according to the research and summary of related literature,it is found that the control chart pattern recognition research in recent years mainly focuses on artificial neural network and support vector machine.The structure and parameter settings of artificial neural network is complex,and many training samples are required.Although support vector machine can be used in the case of small samples,its recognition performance depends on the kernel function to a large extent.Fuzzy C-means（FCM）clustering algorithm has the simple structure,needs less parameter settings,and is suitable for small samples and the clustering result is well,so control chart pattern recognition using the fuzzy C-means clustering algorithm is proposed in this paper.Secondly,the original data of the control chart pattern is generated by the Monte Carlo method.The characteristics of the control chart pattern is extracted by one-dimensional discrete wavelet transform,and it is used as the input signal to train and test the algorithm,then the performance of the control chart recognition is obtained.Due to the accuracy of control chart pattern recognition is affected by stopping threshold and the wavelet function and the number of training sample,each of the above factors will be optimized by controlling variates method.Naturally,get the best recognition performance of this method,it is 99.43%and the standard deviation is0.0028.Now the best parameters are as follows:the stopping threshold is equal to8.25×10^-5,wavelet function is Coif4,the training sample size of 1800.In addition,the theoretical and experimental results show that the stopping threshold has the greatest effect on the algorithm performance,the second is wavelet function,and the training sample size has the least effect.Finally,by comparing with the relevant study in recent years,it is found that the proposed method can not only achieve a high recognition accuracy,but also has a high degree of convenience in feature extraction.And the algorithm has fewer parameter settings and faster learning speed.Furthermore,the method is applied to a practical production case,and the abnormal fluctuation in the process is successfully identified,which proves that the method has good production application value to a certain extent.