Research On Process Monitoring Method Based On CRPS

As a key technology of process monitoring,statistical process control(SPC)has a great significance in enhancing the process quality of products.SPC is mainly based on control charts,and aims to distinguish the normal and abnormal fluctuations in process.So,managers can find abnormalities and take measures as soon as possible,which helps to improve production efficiency.However,most of the existing methods apply to normal-distributed process only,or can't monitor the location and scale parameters simultaneously.In addition,with the development of modern measurement and inspection technology,the applications of various high frequency sensors,machine vision systems and other technologies make process data more abundant,which poses a great challenge to traditional process control.This thesis aims to monitor the location and scale parameters of any continuous process.It introduces the continuous ranked probability score(CRPS)method which is widely used in the ensemble prediction field.Meanwhile,the proposed method also considers the richness of process data,and is further applied to monitor image data.The CRPS-based monitoring method for image data is given in this thesis.First,the basic theory and characteristics of CRPS are explored.It is illustrated that CRPS values can reflect the fluctuation of process location and scale parameters,through calculating the CRPS values of both normal and non-normal process where there are different types of shifts.Then,the specific method of constructing the CRPS chart is given.Control limits are determined by approximating the CRPS distribution using parameter approximate method.Meanwhile,the chart performance is studied by calculating the average run length,and is also compared with other parameter and nonparametric charts.Next,the CRPS chart based on image data is further explored.Since each image is divided into a number of regions of interest,the charting statistic is established by the maximum CRPS value of the mean intensity for each region.Finally,an industrial produced non-woven fabric image is studied to evaluate the chart performances by simulations.The performances are also compared with other existed charts to show that the CRPS chart is pretty effective and useful in practice.