Study On Nonlinear Profile Monitoring Method Based On Wavelet Analysis

In the manufacturing process of complex products such as spacecraft and precision instruments,there is a nonlinear functional relationship between quality characteristics and response variables due to the multiple production processes and complex technology,the functional relationship can be described by profile data.In recent years,scholars at home and abroad are interested in the study of nonlinear profile monitoring.Because of the different sources and various sampling methods of the quality characteristics,profile data that reveals the nonlinear relationship between quality characteristics and response variables contains a lot of noise.Now,scholars mainly use control chart to monitor the profile data.However,in the manufacturing process of complex products,due to the noise interference and the nonlinear profile data,the recognition accuracy of the control chart is not high.So,It is an urgent problem that how to effectively overcome the impact of noise and build a monitoring model with high identification accuracy for nonlinear profile monitoring.At the same time,the cost of complex products is high and the rate of non-conforming products is low,so how to monitor the products in the case of abnormal samples missing is also a problem to be studied.Based on the theory of wavelet analysis and quality control,this paper systematically put forth a new method to monitor nonlinear profile through reading a lot of literature.Firstly,through introducing the basic principles of profile monitoring and wavelet analysis,this paper summarizes universal research and theories of profile monitoring.Secondly,based on the principle of wavelet threshold denoising and support vector data description(SVDD),this paper proposes wavelet threshold denoising for profile data which can solve the problem of high noise in the original profile data.Through interval sampling of complex production process,we obtain a set of profile sample,select appropriate wavelet function for wavelet decomposition and choose suitable threshold function for reconstruction.Using the denoising data and support vector data description,this paper build the profile monitoring model successfully.In addition to,the parameter and performance evaluation indexes of the model have been investigated.Finally,according to the actual background,this paper use simulation data to build a SVDD model for pofile monitoring that based on wavelet analysis.Experimental results show that the monitoring model is more efficient than the existing monitoring methods.The characteristics and innovation of this paper lie in the following three aspects:(1)In order to eliminate the influence of noise on profile data,this paper introduce wavelet analysis method into profile monitoring and use threshold denoising method to overcome the influence of noise on model recognition accuracy.(2)This paper uses support vector data description method to construct SVDD model for the pre-processed data,compares the influence of different parameters on the monitoring effect and provides the implementation steps for this method.(3)This paper compares the proposed method with?~2?T~2control charts and the simulation results shows that the monitoring efficiency is better than that of these methods through comparing the signal to noise ratio and mean square error.This study not only provides a set of operational monitoring methods for nonlinear profile data,but also provides a theoretical basis and analysis method for monitoring other abnormal quality data.