Since industrial processes become more and more complex,the requirements of the sta-bility,efficiency and safety of production are increasing.Among the methods,the Multivari-ate Statistical Process Control(MSPC)methods have been widely applied to solve the prob-lem of process monitoring.But the non-Gaussianity of modern industry process data often limited the using of traditional MSPC method which under the assumption of the samples obeyed Gaussian distribution.Independent component analysis(ICA)method has significant advantage on the handling of non-Gaussian data.It is promising to apply ICA to industry process monitoring.However,most of the MSPCs are unsupervised algorithms.A fault can only be detected in the process of monitoring but cannot be classified.In industrial production process of expert experience is very important to solve this problem,but is rarely used in the above methods.So the supervised learning algorithm with expert knowledge is usually better than that of unsupervised learning algorithm.This dissertation develops the research based on the predecessor's work to solve problems,the main research contents are listed as follows:(1)Based on the traditional kernel independent component analysis(KICA)algorithm,an improved KICA algorithm is put forward in this thesis,namely fault relevant kernel inde-pendent component analysis(FKICA)algorithm.In order to focus on the industrial process monitoring and fault diagnosis problem of product quality,on the basis of the previous meth-od,a quality related fault relevant kernel independent component analysis(QFKICA)algo-rithm is proposed.The proposed methods are applied to fault diagnosis of fused magnesia furnace smelting process.The improved method can make full use of the historical fault data,improving the ability of fault monitoring.The effect of fault monitoring of non-gaussianity process is better.The experiment of fused magnesia furnace smelting process shows that the proposed methods have good fault monitoring ability.(2)By using manifold algorithm(LLE)and semi-supervised algorithm(Margin Fisher)to calculate the transformation matrix,and combined with the idea of the label propagation,a knowledge information process monitoring method based on a manifold clustering is pro-posed in this thesis and used for industrial process monitoring.This method is used in Ten-nessee and fused magnesia furnace smelting process for the experimental study.By analyzing the simulation results,its effectiveness is verified.The experimental results show that the proposed method can make full use of the knowledge information on the unlabeled data.It is a good way to monitor the state of Tennessee process and fused magnesia furnace smelting process,and has a better ability for fault monitoring. |