Research On Pattern Recognition Based On Wavelet Analysis For Hybrid Anomaly Of Control Chart

The control chart is an effective tool for the quality monitoring and diagnosis in the actual production process.Especially the application of monitoring have greatly improved the level of production quality assurance in the production process.Although it plays an important role in the quality monitoring of the production process,it is difficult to identify the hybrid anomaly pattern in the production process,which greatly weakens the recognition effect of the control chart.At present,many researches are based on the basic anomaly model,wavelet analysis and neural network control pattern recognition has become a research hotspot.In this paper,the combination of wavelet analysis and neural network is applied to the study of hybrid anomaly model of quality control chart.In this paper,the research content is as follows.First of all,making the foundation anomaly pattern characteristic data with Monte-Carlo monte method,which generates hybrid anomaly pattern characteristic data source through the data stack.Secondly using based anomaly pattern to train BP neural network and test the correct recognition rate of the network.Then using one dimensional discrete wavelet on the original data feature extraction to establish hybrid abnormal pattern wavelet analysis of control chart pattern recognition model.Finally,the reconstructed detail signal and the approximate signal are used as the basic data of the original data to input into the trained BP neural network.And come to the conclusion?How to select wavelet basis and the characteristics of the hybrid control diagram of anomaly model is the key to research in this paper.In this paper,according to the characteristics of wavelet function selection DbN,SymN and CoifN three wavelet function.And using the MATLAB toolbox to compare decomposition coefficient from three kinds of wavelet function series in different layers and selecting the best wavelet function and the characteristic of the original data decomposition layers.Then use the selected wavelet function and the wavelet decomposition layers to reconstruct and generate the data input into the trained BP neural network for identification.The research shows that when the coif4 wavelet is used to reconstruct the characteristic data,the recognition rate of network identification is higher than others.In this paper,we provide the theoretical support for the study of hybrid pattern recognition of control chart,and provide an analysis method for the identification of other control charts.