Study Of Monitioring And Fault Diagnosis In Industrial Process Based On The Theory Of Multivariate Statistics

Fault diagnosis is fault monitoring and diagnosis for monitor and control system.It also analyzes fault source,frequency,severity,tendency etc.,and provides scientific decision-making basis in order to confirm fault, take remedies,such as timely maintenance and defences.With the development of science and technology,the industrial production installment's structure is getting more and more complex , and develops gradually from the single variable system to the many-variable system primarily . Since it is usually highly nonlinear , time-varying , seriously coupling and its structure parameters are uncertain,traditional fault diagnosis method can't satisfy the demand.Once this kind of system and equipment comes about malfunction,it will take a long time to be solved and lead to a large amount of economic loss , even human injuries or environmental problems.Multiple Statistics Analysis originates from Mathematical Statistics Theory,and it is an important branch in the research of fault diagnosis based on data driven.On the basis of Multiple Statistics Analysis Theory,the paper does a research on the problem of fault diagnosis in two typical processes (continuous process and batch process).Contents in this paper are as follows:1.Summarizes the research circs and development trend of the typical methods in fault diagnosis field.And provides complete introductions based on Statistical Analysis Theory.2.Briefly introduces Principal mathematics tools applied to fault diagnosis method based on Statistical Analysis,including Principal Component Analysis (PCA) , Principal Component Recursion (PCR) and Partial Least Squares (PLS).Analyzes their characteristics used in fault diagnosis.3.To overcome the limitation of the SPE statistics adopted in PCA fault diagnosis,proposes the concept of the sub-space,study the fault detection and reconstruction of the super limit by the T 2 statistics.4.To overcome the shortage of the traditional recursive PLS algorithm in fault monitoring and diagnosis of batch process,a tracking recursive PLS (TRPLS) is proposed.The algorithm is practically worth in monitoring and diagnosis of batch process.5.According to the problem of unconformity between the assumption of measurable data in principal component analysis monitoring methods obeying independence and normal distribution and real production process data,a nonlinear kernel PCA model is proposed,it can distill nonlinear variable,calculate simply,furthermore,there's no need to consider data distributing.