Research On Process Monitoring Of Electro-fused Magnesia Furnace Based On Manifold Framework

In the actual productive process of magnesium oxide,the fact that the process is complex and it is difficult to control the process results in the dead zone with material far away from electrode.Additionally,the uneven heating leads to low yield and quality.The thesis proposed the method that improves the procreative process by using electromagnetic stirring.In this thesis,for productive condition of the low automation control and high frequence of occurring,the kernel flexible manifold embedding(KFME)mehod based on knowledge analysis belonged to the complex industrial fault diagnosis is proposed.The thesis,for the fact that the data collected on the practical industrial process contains noise and strong coupling features,and KFME is sensitive for noise,proposes the fault diagnosis method based on KICA with knowledge analysis plus KFME.According to what we have discussed above,the following research is implemented.(1)Through the analysis of the actual production process of fused magnesia furnace and the information of production,we use electromagnetic stirring to improve the effect of magnesium oxide production in view of the feature of material dead zone and the relation between the quality of magnesium oxide production and temperature.By using the Maxwell’s equations with the relation of electromagnetic field and the finite difference method,the simulation space and Maxwell’s equations are dispersed,so that partial derivative value can be approximated by the differential value.The simulation of the electromagnetic field distribution within the fused magnesia furnace is implemented.Finaly,the relation between string frequence,electromagnetic force and rotate speed is obtained.(2)For the characteristic of nonlinear,high-dimensional and few labeled data,the thesis proposes the fault diagnosis method based KFME with knowledge analysis.KFME uses unlabled data to look for the distribution space of data and uses a little labled data to adjust the distribution space of data based on manifold embedding and the kernel trick,so that the low-dimentional feature can be found from the high-dimentional space.Finally,the fault diagnosis model can be got.The experiment suggests that the fault diagnosis effect greatly dependent on the number of labeled data and balance parameters of model.What is more,the diagnosis effect about unknown fault is unsatisfactory.(3)For the characterisitic of the non-gaussian and strong coupling and the noise information,the thesis proposes the fault diagnosis method based on KICA with knowledge analysis plus KFME.The method uses KICA to cope with the non-gause and noise.Then the diagnosis model can be got by using KFME.The experiment shows that the method has both high accuracy for fault diagnosis and fast speed.What is more,the method proposed in this thesis can not noly achieve fault diagnosis for known fault but also monitor the unknown fault.In addition,the method can provide better guidance for pratical industrial process.