Research On Non-gaussian Process Fault Detection And Diagnosis Methods

Non-Gaussian process fault detection and diagnosis is of great significance forsecurity of industrial production. In recent years, more and more explosions or spillaccidents have happened in the petroleum chemical industries and smeltingenterprises. Absolutely, the research on complex process of fault detection anddiagnosis of has been carried out vigorously. Among them, non-Gaussian processfault detection and diagnosis occupies a large proportion. It is because that mostindustrial process is nonlinear and non-Gaussian. Three-tank as an example, thisthesis mainly research the non-Gaussian process fault detection and diagnosismethods, on the basis of collecting relevant literature review.Firstly, non-Gaussian process dynamic model is set up. This thesis modular formsthree-tank dynamic process, using Simulink of Matlab tool, three state equations andconstraint conditions of water. The fault model is set up at the same time because thisthesis mainly studies non-Gaussian process fault detection and diagnosis. In view ofthe three-tank single fault, it sets up four categories of fault, including tank leaking,pipes clogged, motor fault and sensor fault. Then, it analyzes the data collectedaccordingly. It puts forward a discriminate method based on Matlab function, whichtells us how to determine whether a process is a Gaussian process.Secondly, research on fault detection methods of the non-Gaussian process.Principal Component Analysis (PCA) and Independent Component Analysis (ICA)are now commonly used fault detection methods, which is why this thesis discussesthen uses their and, or and for process monitoring. Once a failure occurs, the system isimmediately to diagnose the fault data. In this thesis, the ICA and PCA methods arecompared. Through the experiments, it shows that the ICA is more suitable fornon-Gaussian process fault detection.Thirdly, research on fault diagnosis methods of the non-Gaussian process. For thesake of contrast, this thesis first to experiment the traditional neural network method,icluding BP Neural Network, Probabilistic Neural Network and Self-Organizing Mapping Neural Network. Based on this, it puts forward two new improved methods,respectively, Clustering Probabilistic Neural Network method and IncrementalSelf-Organization Mapping Neural Network method. They can improve the validityand rapidity of fault diagnosis.Finally, for the convenience of application in the actual process, using of VisualBasic programming language designs a platform for fault detection and diagnosis.Among them, it includes data acquisition, fault alarm, data storage, and otherfunctions, which verifies the reliability of fault detection and diagnosis methodsresearched by this thesis. At the same time, this thesis designs a set of networkarchitecture applied in the industrial process. It can get the fault detection anddiagnosis and the process control together. What is more, it can achieve remote faultmonitoring, which makes security much stronger.