Fault Detection Based On Exergy Information And Support Vector Data Description

Faults in industrial process such as equipment aging,catalyst deactivation,and heat exchanger fouling can lead to degradation of equipment performance and product quality,which in turn have an important impact on the safety and reliability of the production process.If the fault can be detected in time,the maintenance plan can be formulated to avoid economic losses.As energy is a universal concept across physical domains(chemical,electrical,thermal,and mechanical)it follows that energy could be used as another means of achieving the desired dimensionality reduction.And exergy is a unity that reflects the “quality” and “quantity” of energy.Its essence contains information about the performance of the process,and reveals the fault information more than the general attribute variables.In this paper,a feature extraction method based on exergy information is proposed,which not only reduces the complexity of fault detection model,but also shows pretty fault isolability.The main contribution of this paper is to propose a fault detection method based on exergy information and support vector data description(SVDD).The effectiveness and applicability of the proposed method are verified by numerical simulation and industrial distillation case respectively.The main researches are as follows:(1)According to the literatures about fault detection,qualitative,quantitative and semi-quantitative and semi-quantitative fault diagnosis methods are reviewed,and the research status of fault detection based on support vector data description is analyzed.Then the related theories and basic methods are introduced,which mainly explains the basic theory of support vector data description and principal component analysis feature extraction method and feature extraction based on energy information.(2)Since the energy feature extraction method can reduce feature dimension and reveal the overall performance information of the process,a fault detection based on exergy information extraction and SVDD is proposed.Firstly,a group of variables that are most relevant to the exergy efficiency are identified as energy feature variables by mutual information,and feature selection is carried out accordingly.Then,the feature variables are used to establish the SVDD models in different states,and the relative statistics of the model are used to determine the fault state of the sample to detecting the sample faulty or not.(3)Faced with the problems of time-varying characteristics of system and incompleteness of the known sample sets in actual process,this chapter adopts incremental learning technology to realize incremental updating of the model.At the same time,to overcome the shortcomings that the selected energy features are difficult to reflect the dynamic characteristics of the process,an energy feature extraction based on variable forgetting factor is used to update the energy features adaptively.Then,an online fault detection algorithm based on exergy information and incremental SVDD is proposed to accommodate the dynamic change of the system.Firstly,variable forgetting factor is introduced to adapt the energy features extracted at each time to the dynamic process.Based on the historical SVDD model,the model structure is continuously incrementally updated using valid historical data and new energy feature sets corresponding to the incremental samples.Different fault state models are obtained from a variety of fault state samples,and the samples are classified according to the relative statistics of the model,so that the fault state of samples can be detected.(4)The two proposed fault detection methods are applied to numerical simulation and distillation process.The research results show that the fault detection based on exergy information extraction and SVDD can extract the performance degradation information in the process.The addition of the exergy efficiency concept not only improves the fault isolability,but also indicates the evolution direction of the fault;The online fault detection based on exergy information and incremental SVDD solves the problem of online updating of the model,and also increases the speed of fault detection while preserving the classification performance.