Research On Process Monitoring Of Electro-Fused Magnesia Furnace Based On Multi-View Data

The purpose of industrial process monitoring is to discover the abnormal situation of the production process in time and effectively,to ensure the safe production,to improve the product quality and to reduce the cost.In recent years,the industrial processe monitoring methods based on data-driven have become a research hotspot and are widely used.Among these methods,independent meta-analysis methods and non-negative matrix factorization methods are widely used due to the characteristics of extracting the valid information of data.At the same time,with the continuous complexity of industrial processes,it needs to monitor in a comprehensive way with data from multiple perspectives to achieve better monitoring results in the process of fault detection and diagnosis.Based on the previous work,this dissertation applies the independent element seeking method based on Hilbert-Schmidt independence criterion to industrial fault detection and diagnosis based on multi-view data,and puts forward the MFFD multi-view data dimension reduction method.The algorithm is applied to the fault detection and diagnosis of the fused magnesia furnace.In view of the above problems this thesis mainly did the following research:(1)The traditional kernel independent element is obtained by the method of KPCA and ICA.There are obvious shortcomings in using the ICA method to optimize the objective function.In this thesis,the independent element method of HSIC is applied to fault detection and diagnosis of fused magnesium furnace.By introducing the sliding average of the monitoring statistics,the simulation part makes it more suitable for multi-view data and improves the visual effect and detection effect of the curve.That is to say,when monitoring statistics of multi-view data have a large degree of transition,the addition of smoothing can restrain this uncertain transition.With the help of GBDT,the control limit is improved to achieve multi-fault diagnosis.It is possible to not only judge the fault category but also the severity of the fault by monitoring the statistics.At the same time,the data of the three perspectives of the grayscale data,the color data of the image and the physical variable data are modeled together to improve the accuracy of the final fault detection and diagnosis.(2)The sources of multi-view data are different,and the contribution of data from different perspectives to the final fault detection effect is different,and quantify the importance of each perspective by optimizing the perspective weights.In this thesis,a method of dimensionality reduction based on multi-view data is proposed.NMF is used to mine the local information of each view.LDA is used to restrict the classification ability of the basis matrix after synergetic reduction.Finally,this method is applied to the fault detection and diagnosis of the melting magnesium furnace,and achieved the desired result.(3)Different sources of data have different ways of use,this thesis presents two ways to establish multi-view data model:one is a unified modeling method for multi-view data,which improves the accuracy,robustness and comprehensiveness of fault monitoring and diagnosis;the other is to reduce the dimension of each perspective data,and then calculate the weight of each perspective.Finally,the fault detection and diagnosis are carried out according to the weight of the view.