| MgO single crystals have been used for many kinds of applications, due to its chemical stability, low dielectric constant and thermal expansion coefficient. In the high temperature superconductivity applications area, MgO crystals can be used with yttrium-barium-copper oxides for superconducting thin films. Other applications include the deposition of ferrous electric thin film coatings on magnesium oxide substrates. And there are also growing applications for MgO in the field of plasma display panel (PDP) technology.Crystal growth by the arc-zone melting technique has been used over the past decades to prepare single crystals of refractory metals and their alloys. Such a technique can easily generate temperature up to3600℃in a short time. Because Magnesium oxide is well known refractory material which has very high melting points up to2800℃, nowadays the arc-fusion melting technique become the main method applied to grow large magnesium oxide (MgO) single crystalsThe size of the MgO melt pool in the electric arc furnace is important to grow high-purity MgO single crystals. For the MgO single crystal furnace is a system which is multi-variable, nonlinear, intense coupling, diverse and with inferior working condition and frequent random disturbance, the practical measurement methods are so limited that the furnace design and the power control strategy have been the greatest challenges for the engineers. In order to control the size of the MgO melt pool precisely, the theory of soft-sensor was studied according to the analysis of the heat transfer theory of the electric arc furnace.Soft-sensing was a technique that was used to estimate nonmeasurable process states based on available measurements of input and output values. Taking the size of the MgO melt pool as research background, in combination with the mechanism of MgO melt process, deep research and discussion on the soft-sensing modeling methods and related problems funded by the863project of "Research on the Material Technology and Industrialization of MgO Crystal for PDP" and "Research on the Environmental Melt Technology and Equipment of Low-grade Magnesite" were performed.The main contributions of this paper are summed up as follows: 1. A model of the furnace has been built to estimate the size of the MgO melt pool in MgO single crystal furnace in terms of its consumed energy by using finite element method. In the investigation of this method, the growth of the MgO melt pool was calculated after the start of operation for5hours,10hours and15hours. The representation method of the sharp and the size of the MgO melt pool were indicated based on the simulation.2. In order to simplify the soft sensor model, some secondary factors were ignored by finite element method and operation parameters were treated as constant values, these would cause the prediction result were inaccurate. So a soft-sensor based on a partial least-squares extreme learning machine (called PLS-ELM) was proposed according to the analysis of the heat transfer theory of the MgO single crystal furnace to control the size of the MgO melt pool precisely, The learning capability and generalization performance of the model was examined by comparison to the model based on the support vector machine (SVM). The comparison results show that PLS-ELM has similar accuracy compared to SVM, but it has obvious advantages in parameter selection and learning speed.3. If the model of the soft sensor remains unchanged when the working condition changed, the performance of the intelligent algorithm would be restricted. So a new updating method for soft sensor model based on incremental learning was proposed. In the new method, the new data was used to train new model firstly. And then the new model was combined with the old models according to their weights. The new updating method adopted the latest training sample and weights the old training samples iteratively to insure that the influence of the old training samples was weakened. So it was benefit in the sense of computational cost and prediction accuracy by using the new updating method.4. A hybrid intelligent control structure based on soft sensing was proposed, and a soft sensor model of the size of the MgO melt pool was developed in the DCS which has been used to control the process of MgO single crystal growth in Liaoning Zhongda Superconducting Material Company. The modules such as data communication, data acquisition, data pre-treatment and soft sensor algorithm were developed in the sever computer. The operation results of the soft sensor were also showed on the screen of DCS, which would provide a good reference to the operators.The application of the soft sensor system in the Liaoning Zhongda Superconducting Material Company proved the effectiveness of the method to measure the size of MgO melt pool in MgO single crystal furnace, and the crystal yield were improved, energy was saved. |
PDF Full Text Request