Study On Nonparametric Quality Control Methods For Ordinal Profile

In the practical process of manufacture and service,the quality characteristics of some products and services are represented by a profile,which describes the functional relationship between response variables and explanatory variables,such as the relationship between product failure rate and service time.In the manufacturing and service industries,the response variables may be attribute data,such as the service is good,marginal or bad,the product is qualified or unqualified.For response with three or more attribute levels,parametric models are usually used to describe the function relationship of ordinal profiles.Many control charts based on these parametric models have been proposed to monitor ordinal profiles.However,the robustness of these parameter control charts is poor,and most monitoring methods do not consider the ordinal information between ordinal responses.This thesis aims to fill the gap by discussing the nonparametric quality control methods of ordinal profiles to solve the monitoring problems as follows:First,in order to reflect the order information between response variables,this thesis uses a nonparametric proportional odds ratio model to fit the functional relationship of the ordinal profile.Based on this model,this thesis constructs two nonparametric control charts.Second,when the in control model is known,this thesis uses the local linear kernel estimation method to estimate the model,and compares the performance of the nonparametric and parametric control charts under different simulation scenarios.Third,when the in control model is unknown,based on two nonparametric and one parametric model estimation methods,this thesis mainly analyzes the influence of different model estimation methods on the in control and out of control performance of the phase II control chart.Fourth,apply the nonparametric control charts constructed in this thesis and one parametric control chart to a practical case,illustrate the specific application method of the control charts,and analyze the performance of the control charts.Assuming that the in control model is unknown,the three model estimation methods mentioned in this thesis are used to estimate it,verify the estimation effect of nonparametric and parametric model estimation methods.This thesis can provide new ideas and methods for the in-depth study of ordinal profile monitoring methods,and could also help to improve the quality of products and services in the manufacturing and service industries.