The Detection Method Of Component Content Based On Color Characteristic In Pr/Nd Countercurrent Extraction Process

The rare earth elements are necessary to be extracted from the paragenic ore of the nature world before they are widely applied in industry products. Currently the method that the rare earth is extracted by the P507-HCL solvent is the most commonly rare earth separation system to obtain a single, high purity rare earth oxides at home and abroad. The component concent distribution of the rare earth elements is the significant indicator to measure extraction separation effect in the rare earth extraction separation process. During the production of the rare earth with ion characteristic color, the deviation degree of "the ion color features belt" to judge component content distribution is usually observed based on the operator’s experience. However, this way has many shortcomings, such as, large error due to the subjectivity in the judgment of component content, low production efficiency, irrealizable quantitative description of color information in mixed solution in the tub which should provide accurate information for subsequent process parameters adjustment and blind manual operation. Thus, this way leads to huge resource consumption in the extraction process, unstable product quality, and weak market competitiveness. It is necessary to research the measurement method of component content based on the color features during Pr/Nd extraction process, which is an effective method to measure rapidly the component content and can provide data support for the optimal control of the rare earth extraction process. Therefore, the measurement method research is of great significance for the sustainable development of the rare earth industry. The main researches and contributions of this thesis are as follows:(1) According to the characteristics of the Pr/Nd extraction solution, the extraction method of the solution color feature is analyzed in the common RGB/HSI color space. It is obviously indicated that the first moments of H/S/I features component are adopted to describe color information of rare earth solution.(2) Taking standard Pr/Nd mixed solution as a benchmark which is configured in laboratory, the first moment of H/S/I component based compensation model of Pr/Nd extraction mixed solution image is set up to eliminate image error caused by different kinds of noise in the production site. And then, the rapid forecast model of component content is established based on LSSVM in the Pr/Nd extraction process, by using principal component analysis to select the first moment of H and S components as input variables of model, which is close to component content. A self-adaptive genetic algorithm is employed to optimize combined parameters of the model. The test data shows high precision and good generalization ability of the developed model.(3) According to the particularity of the color of the Pr/Nd mixed solution, a quick forecast model of component content based color feature fusion by using LSSVM is proposed. Firstly, it puts forward a relative green component and the color vector angle in RGB color space, then describes the image information of Pr/Nd mixture solution with combination of the feature components of H/S, next uses the partial least squares method to eliminate the correlation among the four color components, and finally establishes the component content’ LSSVM model by selecting the fused characteristic components as input variables. The testing data shows that the component content based modeling method has higher prediction precision and better generalization ability by considering the universality and particularity.(4) Since there exist many factors influencing distribution of the element component content in the rare earth extraction site, a fixed model of component content has better generalization ability to the historical data but causes the prediction accuracy getting worse easily when it’s running at the long-term or continuous period. To solve this issue, an adaptive correction method for LSSVM model of component content based on color feature is proposed in the Pr/Nd extraction process. Such a method establishes LSSVM model of component content which has the high precision to historical data samples. Based model correction criteria, self-adaptive iteration method is adopted to update the model of the component content. Compared with conventional LSSVM, SVM and BP neural network models of element component content, the running data test, which is operated in the working condition change, shows that the method has higher prediction precision and good adaptive ability, and it is suitable to monitor REEs component content at the long-term and continuous period in rare earth extraction site.(5) Taking a rare-earth company’s Pr/Nd extraction production line as the application object, a machine vision based rapid detection system of component content of Pr/Nd mixed solution is developed in laboratory environment through designing hardware system and developing software system by using the virtual instrument (LabVIEW) and Matlab tools software. By analyzing and testing the running data in the extraction field, it is shown that the accuracy, repeatability and quickness of the detection system can meet the detection requirements of component content during the Pr/Nd production, which provides accurate and reliable reference basis for each process parameter adjustment of Pr/Nd extraction separation process and guarantees the stable operation with the extraction process at a long time and continuous period.