Fault Dignosis Based On The Integration Of Local Linear Embedding And Exponential Discriminant Analysis

Industrial process data have the characteristics of large scale,high complexity,multivariable and strong correlation.There is great significance on how to find out and deal with the fault from data accurately and quickly,which can help ensure the efficient operation of the process.Industrial process data have the characteristics of high dimensions and nonlinearity,so it is very important to extract the data feature for fault diagnosis.In addition,an accurate and reliable fault diagnosis system plays an important role in the normal operation of industrial processes.Based on this,this paper puts forward several improved fault diagnosis methods based on local linear embedding(LLE)and exponential discriminant analysis(EDA).It has been verified by the simulation platform of TE and the simulation platform of penicillin fermentation process.Firstly,an improved fault identification approach for batch process is proposed named as kernel exponential discriminant analysis(KEDA),in which a kind of performance index based on difference degree is given to identify fault classification.KEDA combines the advantage of the kernel technology and the EDA.The proposed KEDA method shows powerful ability in dealing with nonlinear,small sample size data and has a noticeable improvement in classification performance.During the real application to fault identification,the normal data model and fault data models for known faults are established according to the historical data first.Then online measurement data are fed into these models to identify the current operation status,that is,whether the system is in normal state or failure state.If it is a fault state,this method can determine what kind of fault occurred or a new fault has occurred.Secondly,this paper proposes two kinds of improved exponential discriminant analysis methods:local linear exponential discriminant analysis(LLEDA)and neighbourhood preserving embedding discriminant analysis(NPEDA).The two methods both combine the global discriminant analysis with the local structure preserving.LLEDA is a parallel strategy to find a trade-off projection vector between the local geometric structure preserving and the global data classification.NPEDA is a cascade strategy,and its dimensionality reduction process is divided into two steps:keeping local structure and discriminant analysis.The two methods emphasize the intrinsic structure of the data while utilizing the global discriminant information,so they have better discrimination power than the traditional EDA method.Finally,the paper proposes a hybrid multi-method process monitoring and fault diagnosis method for the complex industrial process.The method includes the data analysis,model library establishment,timely diagnosis.Firstly,the historical data are simply screened by conventional methods PCA to distinguish normal and fault information.Then the clustering method is used to classify the fault data set,and the fault model libraries are established by LLEDA method.Finally,the fault diagnosis is carried out.The LLEDA method based on supervised learning is extended to unsupervised learning,which facilitates the processing of a large number of unlabeled data in complex industries.The proposed method is verified by the typical Tennessee Eastman process,and the result proves the effectiveness of the method.