Monitoring The Out-of-control Of Nonlinear Profile Based On DHNN

Profile monitoring is a real-time control of a process(or product)which the key quality is characterized by profile.In the case of small sample size and very complex data relationships,using the traditional method to build nonlinear profile model may be easily appear overfitting phenomenon.If the model form established seems very complex,the estimation of the relevant parameters will have a big error and the performance of the control chart is bound to be low,so it is difficult to monitor the out-of-control state of profile effectively.Therefore,under the premise of small sample size,the method of establishing control chart to monitor out-of-control profile has some limitations.In order to solve the above problems,we suppose to use the support vector regression to establish a profile model and use the discrete Hopfield network to monitor the out-of-control state of nonlinear profile.The main contents of this thesis are as follows:(1)Aiming at the problem of fitting nonlinear profile model with small sample size,the support vector regression machine is introduced to fit nonlinear function relationship.Firstly,the complex production process is sampled at intervals to obtain sample profile data set.Then,the suitable kernel parameters are selected and using the support vector regression machine to establish nonlinear profile model.This supposed method can overcome the problem of slow convergence speed and overfitting phenomenon exist in the traditional modeling method.Finally,establish the fitting mean and variance surfaces according to the measured points and their response values to illustrate the applicability of using support vector regression machine to fit nonlinear profile model.(2)Aiming at the problem of low performance of profile control chart,we suppose the discrete Hopfield network to monitor the out-of-control profile.Firstly,the standard profile under controlled state is stored as the attractors in the network.And then input a sample profile to the network,it will be monitored through iterative learning.At last the out-of-control state of profile is monitored by comparing sample profile and standard profile using the associative memory function of the discrete Hopfield network.(3)Simulation and empirical research are conducted on the method of support vector regression to fit the profile model and the discrete Hopfield network to monitor the out-of-control profile.Firstly,introduce an example,using a nonlinear function to represent the quality characteristics of a specific complex product manufacturing process,and use the discrete Hopfield network to monitor the out-of-control state of profile.Secondly,we give an example of an automobile engine and use the support vector regression machine to establish a profile model,and then use the discrete Hopfield network to monitor the out-of-control state of profile.Simulation and analysis of the empirical show that the established model has better global description ability using the proposed method,and the performance of monitoring the out-of-control state of profile is strong.With small sample size as the premise,the proposed method can reduce the cost of quality optimization process effectively.