CHEMICAL ENGINEERINGTRANSACTIONS VOL. 51, 2016 A publication of The Italian Association of Chemical Engineering Online at www.aidic.it/cet Guest Editors:Tichun Wang, Hongyang Zhang, Lei Tian Copyright © 2016, AIDIC Servizi S.r.l., ISBN978-88-95608-43-3; ISSN 2283-9216 Silver Ore-dressing Model and Performance Analysis Based on ANFIS Ping Huanga, Ji Geb, Zaipeng Duan*a, Jingjing Xua aCollege of Environment and Resources, Fuzhou University, No. 2 Xueyuan Road, Minhou County , Fuzhou, Fujian, China bCollege of Resources and Environmental Engineering, Jilin Institute of Chemical Technology, No.45, Chengde street, Longtan District, Jilin, Jilin, China duanzaipeng@163.com 1. Introduction The aim of mineral ore-dressing technology research and ore-dressing production technology management is to improve the utilization rate of resources, reduce environmental pollution and costs, and improve efficiency and get higher profits. The main purpose of the mineral ore-dressing is the pursuit of higher resource recovery, in order to make maximum use of resources. Ore recovery rate is determined by analyzing ore, concentrate grade is calculated, and ore dressing recovery rate of high and low, the concentrate grade is related.Arhada Lead and Zinc concentrator, for example, analyze the influence on ore-dressing recovery rate, find what factors influencing the ore-dressing recovery rate of the larger, get the effect and correlation of general law, to guide the reasonable control of mineral ore-dressing technology research and Industrial production control, determine the beneficiation technology in the economic indicators plays an important role. Therefore, it is very necessary to explore analysis correlation between grade and ore-dressing recovery rate. In this paper, by integrating a total of 6 months of daily report from Arhada Lead and Zinc concentrator, 484 sets of data, based on controlling screening of these data and eliminating abnormal data, carrying out based on the mathematical relations between common 11 kinds of a metadata model of data statistics, correlation analysis and statistical test, through the comparison of all kinds of models, silver ore-dressing recovery rate and concentrate grade is established the unprocessed ore grade optimal data model. 2. The recovery rate model of silver mineral ore-dressing Arhada Lead and Zinc ore-dressing plant silver ore grade and ore dressing recovery rate correlation curve regression analysis, modeling results are shown in Table 1 below. The adaptive fuzzy inference system (ANFIS) of Arhada Lead and Zinc ore-dressing plant silver ore grade and ore dressing recovery rate correlation curve regression analysis. Importing 484 sets of data by sliver rate of recover and sliver grade to ANFIS edit, it would generate ANFIS automatically by Grid partition. After 500 times, training of the model of ANFIS would be got. The adaptive fuzzy inference structure diagram, the recovery rate of silver after training, see Figure 1, Figure 2. From the training output of silver recovery rate relations in Figure 2, It can be seen that with the increase of silver ore grade, the silver recovery rate tended to decrease. When silver ore grade exceeds 26 g/t, the recovery rate of silver showed a trend of fluctuation. Because the original design of the beneficiation process of silver ore grade design is about 20 g/t, due to the limitation of the beneficiation process design, with the increase of silver ore grade and silver recovery rate has shown a decreasing trend. Initial membership function and membership function after training, comparison of model output and original output and training error, see Figure 3, Figure 4. DOI: 10.3303/CET1651035 Please cite this article as: Huang P., Ge J., Duan Z.P., Xu J.J., 2016, Silver ore-dressing model and performance analysis based on anfis, Chemical Engineering Transactions, 51, 205-210 DOI:10.3303/CET1651035 205 Table 1: Based on the raw data the overview and parameter estimation table of silver mineral ore-dressing rate of recover model Model review and the parameter estimates Dependent variable: silver rat of recover Equation Overview of model Parameter estimation R2 F df1 df2 Sig. Constant b1 b2 b3 Liner 0.060 30.496 1 476 0.000 71.469 -0.267 Logarithmic curve 0.101 53.258 1 476 0.000 96.466 -9.989 Quadratic curve 0.123 66.619 1 476 0.000 53.266 262.940 Cubic curve 0.140 38.815 2 475 0.000 85.849 -1.222 0.014 Blend curve 0.152 28.398 3 474 0.000 99.331 -2.496 0.049 0.000 Power function curve 0.057 28.584 1 476 0.000 71.029 0.996 S-curve 0.094 49.621 1 476 0.000 104.471 -0.154 Growth curve 0.115 61.696 1 476 0.000 3.982 4.052 Exponential curve 0.057 28.584 1 476 0.000 4.263 -0.004 Logistic curve 0.057 28.584 1 476 0.000 71.029 -0.004 Independent variable: silver raw ore grade Figure 1: The adaptive fuzzy inference structure diagram 206 Figure 2: The recovery rate of silver after training Therefore, recommended Arhada Lead and Zinc ore-dressing plant is necessary for effective improvement of the relevant part of the sorting and recovery of silver, in order to facilitate the recovery of silver, and avoid the loss of silver resources and waste. Finally with the increase of silver ore grade and silver recovery rate in a certain extent also with corresponding increase, to silver resources recovery and utilization can be achieved. (a) (b) Figure 3: Initial membership function and membership function after training 207 Figure 4: Comparison of model output and original output and training error 3. The grade model of silver concentrates The modeling results are presented in Table 2 through conducting fitting analysis of correlation between the grade of silver raw ores and the grade of sliver concentrates in Arhada Lead and Zinc Mine. Table 2: The grade model of silver concentrate review and the parameter estimates which are based on the original data Model review and the parameter estimates Dependent variable: grade of silver concentrates Equation The model review The parameter estimates R2 F df1 df2 Sig. Constant b1 b2 b3 Liner 0.155 87.226 1 476 0.000 1.265 8.199 Logarithmic curve 0.129 70.197 1 476 0.000 785.063 216.113 Inverse curve 0.111 59.435 1 476 0.000 1.680 -4.786 Quadratic curve 0.187 54.739 2 475 0.000 1.440 -3.429 0.165 Cubic curve 0.254 53.823 3 474 0.000 828.100 54.373 -1.421 0.012 Blend curve 0.126 68.442 1 476 0.000 1.298 1.005 Power function curve 0.114 61.377 1 476 0.000 966.893 0.130 S-curve 0.107 56.795 1 476 0.000 7.417 -2.992 Growth curve 0.126 68.442 1 476 0.000 7.168 0.005 Exponential curve 0.126 68.442 1 476 0.000 1.298 0.005 Logistic curve 0.126 68.442 1 476 0.000 0.001 0.995 Independent variable: silver raw ore grade When the coefficient of fitting degree, R2 equals to 0.254, it means that the preselected model has a big difference from the real model. Through observation of silver concentrate grade and silver ore grade diagram, we can see that the relationship of silver concentrate grade and silver ore grade is relatively more concentrated. Thus, the method of fuzzy clustering can be applied for finding out the clustering center in the relation. Through using the command of find cluster to open the graphical fuzzy clustering tool-MATLAB, importing the 484 groups of data formed by the silver concentrates and silver raw ores into it, selecting the subtractive calculation method, importing corresponding parameters into it and then choosing the Start button, the clustering center of the silver concentrates and the silver raw ores can be found. The clustering center is presented in Figure 5. The coordinate value of the clustering center is (x=23.2, y=1445). It is better that the mineral ore-dressing flow sheet and the flotation reagent can be controlled to get a relatively accurate silver concentrate grade in the silver ore-dressing recovery. And then the silver resources can be realized an efficient utilization if the recovery rate can be increased to an extreme on this basis. Thus, after clustering the silver concentrate model into a clustering center through adopting the method of fuzzy clustering, This point corresponds to the quality of the concentrate grade is the optimal target value The target value of the silver concentrates obtained from the fuzzy clustering analysis is 1445 g/t. Thus, the mineral ore- 208 dressing flow sheet and the reagent adopted in the flotation had better be controlled in making the silver concentrate grade fluctuate around 1445 g/t. And then on this basis, through increasing the recovery rate of silver to an extreme to realize efficient recovery and utilization of silver resources, the ore-dressing plant can get the maximum economic benefits. Figure 5: The clustering result of the relation between the silver concentrate grade and the silver raw ore grade 4. Analysis of the Model of Silver’s Ore-dressing Recovery Rate As is shown in Table 3, the optimal model of silver’s ore-dressing recovery rate can be built by checking and screening the original data set, removing the abnormal ones, calculating the standard deviation and at last with the regression analysis. Table 3: Summary of the Optimal Model of Silver’s Ore-dressing Recovery Rate and Parameter Estimation Equation Model Review Parameter Estimation R2 F df1 df2 Sig. Constant b1 b2 b3 Cubic Curve Mode 0.8125 236.87 3 164 0.000 115.03 -4.2269 0.1017 -0.0008 Table 3 indicates that the optimal model of silver’s ore-dressing recovery rate is as follows: 3 2 0.0008 0.1017 4.2269 115.03y x x x     (1) Where y—silver’s ore-dressing recovery rate, %; x—raw silver grade, g/t. The results of goodness-of-fit test of the model: R2=0.8125, Sig=0. Based on selected silver’s grade in the original data set, the data is evenly divided into 12 groups from 0 to 48 with the gap of 0.4 between each group, followed by elimination of abnormal data. The fuzzy clustering center points for each data group can be obtained with every data group analyzed in the means of fuzzy clustering. Finally, with the regression analysis of the data set composed of fuzzy clustering center points in each group, the optimal model of silver’s ore-dressing recovery rate will be built. The optimal model of silver’s ore-dressing recovery rate is as follows: 2 0.018 1.765 95.58y x x   (2) Where y—silver’s ore-dressing recovery rate, %; x—raw silver grade, g/t. The results of goodness-of-fit test of the model: R2=0.918, Sig=0. The index of judging fitting degree of the built model is the coefficient of determination R2. The closer R2 is to 1, the higher the fitting degree is. R2 of the two kinds of models built by curve fitting and ANFIS is 0.8125, 0.918 respectively. It shows that there is indeed high correlation and close connection between lead recovery rate and raw lead grade. From the two ways of modeling above, it can be concluded that the R2 value in ANFIS model is bigger than that of curve fitting model. It shows that the fitting degree of the ANFIS model is better 209 than that of the curve fitting model. Therefore, ANFIS modeling is more advantageous and representative than the curve fitting model. Based on the modeling results above, a t-test has been given to the regression coefficient in the prediction model of silver recovery rate. For SPSS model: t1 = 2.347 > 2.201, indicating that the constant term has practical influence on the forecast of silver recovery rate; t2 = 2.415 > 2.201, showing the great significance of raw silver grade upon lead recovery rate. For ANFIS model: t1 = 2.542 > 2.201, indicating that the constant term has practical influence on the forecast of silver recovery rate; t2 = 2.535 > 2.201, showing the great significance of raw silver grade upon lead recovery rate. As can be seen, the goodness of fit in ANFIS model is better than that of curve fitting. Therefore, it is more realistic to predict silver’s ore-dressing recovery rate with ANFIS fitting function (y = 0.018x2-1.765x+95.58). This result shows that the model can be used for the forecast of silver’s ore-dressing recovery rate for ore-dressing plants. 5. Conclusion On the basis of statistical analyses of silver samples’ grade frequency in various mining areas of Arhada lead and zinc mine, A model for geologic reserves and average grade can be built through modeling work such as data smoothing, reserves integration and reserves regression. By regression analyses and statistical tests, and comparisons between various models, Adaptive-Network-Based Fuzzy Inference System (ANFIS) is employed to build a model for raw silver grade, ore-dressing recovery rate and concentrate grade. It is considered, by analyzing the model, that ANFIS model is more applicable to the establishment of the silver- dressing model than curve fitting model. Silver-dressing recovery rate is closely related to the raw ore grade and the concentrate grade, of which the raw ore grade is most relevant, followed by the concentrate grade. This is why the raw ore grade should be reduced and controlled in searches and practice of ore-dressing. For the certain silver concentrate grade, ore-dressing recovery rate increases as the raw ore grade increases: ore- dressing recovery rate and the raw ore grade are positively correlated; for a certain raw silver grade, ore- dressing recovery rate decreased, ore dressing recovery rate and selected ore grade is negative correlation. Acknowledgments This research is supported by the China Postdoctoral Science Foundation (2016M592895XB). Special thanks to Xilinguole league Shandong Gold Arhada Mining Co., Ltd., which provided much research material and support during the research crew worked at the mine, and the discussion and suggestion also improved the study. Reference Cosenza B., Galluzzo M., 2011, Adaptive Type-2 Fuzzy Logic Control of Non-Linear Processes, Chemical Engineering Transactions, 2011: 235-240, DOI: 10.3303/CET1124040 Meng X.Y., Gui X.Y., Yu Z.M., 2008, Study on application of drill cutting weight andits gas desorption index in forecasting coal and gas outburst, Mining Research and Development 28 (2), 75 (in Chinese), DOI: 10.3969/j.issn.1005-2763.2008.06.001 Mitri H.S., 2007, Assessment of horizontal pillar burst in deep hard rock mines, International Journal of Risk Assessment and Management 7 (5), 695–707, DOI: 10.1504/IJRAM.2007.014094 Nie B.S., 2003. Present situation and progress trend of prediction technology of coal and gas outburst, China Safety Science Journal 13 (6), 40–43 (in Chinese), DOI: 10.16265/j.cnki.issn1003-3033.2016.01.002 Rocscience Inc., 2007. Online Help Manual: Phase 2 v6.0 – Finite Element Analysis for Excavations and Slopes. Yin G.Z., Li X.Q., Zhao, H.B, Li X.S., Li G.S., 2010. Insitu experimental study on the relation of drilling cuttings weight to ground pressure and gas pressure, Journal of University of Science and Technology Beijing 32 (1), 1–7 (in Chinese), DOI: 10.13374/j.issn2095-9389.2016.03.002 210 http://dx.chinadoi.cn/10.3969%2fj.issn.1005-2763.2008.06.001 http://doi.org/10.1504/IJRAM.2007.014094 http://dx.chinadoi.cn/10.16265%2fj.cnki.issn1003-3033.2016.01.002