Research On Fault Detection Method Of Fused Magnesia Furnace Based On Semi-supervised Manifold Learning

With the continuous development of modern industry,the industrial process has become more and more complicated._Once the fault occurs,i_t will bring huge propertylos s and even safety hazard.It becomes more and more important to find abnormalfaul ts in the industrial process duly that puts higher demands on the fault detection ofindu strial processes.In _view of the non-linearity,faul_t proneness and low degree ofautom ation control in the current production process of magnesia,this paper modelswith i mage data and current data collaboratively and makes the fault detectionexperi ment simulation of the data collected during the industrial process of the fusedmagnes ia furnace from feature selection,semi-_supervised learning methods andprevent ion of semi-supervised degradation.In vie_w of the above problems,this pa_permainly co mpletes the following research:(1)A fault detection method based on feature reconstruction for collaborativemodeling of heterogeneous data is proposed.By estab_lishing the heterogeneous datacomposed o f current data and image data in the industrial process of fused magnesiafurnace,a collabo_rative modeling method is adopted.The curren_t and image data playa role at th e same time;then the feature selection is performed based on the featurereconstructio n method;and the manifold structure of the data is maintained before andafter the feat ure selection;finally,the data of th_e feature selection is performed by thesupport vector m achine for fault detection.It reduces the _data dimension and theamount of calcula tion,then improves th_e sensitivity of fault detection,finally facilitat_esreal-time fault det ection by the feature selection of the original data.(2)A semi-supervised method is introduced to reduce the cost of data markingbased on the method of the previous part.A fault detection _method based on featurereconstruction for s emi-supervised collaborative modeling of heterogeneous data isproposed.Feature sel ection i_s performed by feature reconstruction,and the mark andma_nifold structure of marker data and the manifold structure of current data aremaintained before and after feature selection.Finally the regular t_erms of 1-norm and2-norm are discussed.(3)In order to prevent the degradation of semi-supervised methods and improve the accuracy of fault detection,the semi-supervised support vector machine,_graphme thod and disagreement-based methods are combined to improve the existingse mi-supervised method.A disagreement-based safe graph fault detection methods ofsem isupervised support vector machine in industrial process is proposed,wh_icheffe ctively improves the accuracy of fault detection and reduces the possibility ofsemi-supervised degradation.Finally,the algorithm proposed in this paper is used in the industrial process offused magnesia furnace for simulation analysis.The _fault detection effect and real-timedetect ion of the algorithm are analyzed respectively.Compa_red with some traditionalmethods,this p_aper verifies the reliability and effectiveness of the methods.