Dissertation
To receive a Master's degree
Mining Engineering - Mineral Processing Orientation
Abstract
Galghar Complex Magnetite Processing Factory Sirjan includes three circuits of dry crushing and classification, dry magnetic separation and more magnetic crushing and separation, the tonnage of feed entering the factory is 450 tons per hour and the final concentrate tonnage (more concentrated and dry) is 200 tons per hour. The number of nodes and currents for the purpose of mass balance in the circuit of dry crushing and classification and dry magnetic separation are 4 and 8, respectively, and in the circuit of crushing and magnetic separation, 9 and 16, respectively. In this research, first, the rate of unknown flows was calculated using the data obtained through manual sampling of accessible flows in steady state and in different working shifts of the factory, and then using the technique of data adaptation, the obtained results were corrected, and finally, the weight of the samples and parameters of the optimal sampling system were calculated. The problem of adapting the data in this research is non-linear. After balancing the mass, adapting and correcting the measured grade and tonnage values, using different approaches, including solving matrix equations in MATLAB software using classical and meta-heuristic linear and numerical nonlinear analytical methods (such as genetic algorithm and hybrid method), bilinear Crow, matrix projection and Simpson methods (using unmeasured currents 1, 3, 11 and 14) and correction method The data has been analyzed in the balance software to detect outliers and remove them. In the following, for the important and necessary points in the magnetite processing circuit, the weight, number and time interval of taking the samples have been calculated using the variogram method. In this research, it was shown that the combined method for all three circuits of the factory has the best minimization results. For example, the maximum value of the error of setting the conditions was reduced by the combined method for the magnetic separation circuit from 1310.730 to 1.364 x 10-12 units. In addition, the total estimated recovery value in the magnetic separation circuit increased from 86.93% to 96.22%.
Key words: mass balance, data integration, hybrid method, sample component weight, sampling system, Golgarh.
- Introduction
In ore processing plants, the validity of the measured data and their comprehensiveness plays a fundamental role in the correct evaluation of the system; So that invalid data may cause the officials to make a mistake in making decisions. On the other hand, in a processing plant, measurements always have errors; Therefore, it is necessary to correct the measured data before use. In addition, many times, it is not possible to measure some data from factory flows, due to some technical or economic limitations. For example, in most processing plants, most flow rates[1] are not measured; Therefore, the values ??of such data must be estimated somehow [1]. In general, measured and unmeasured data are divided into two categories [2]:
1- Measured data
Adjustable: A measured variable can be adjusted when its values ??can be optimally modified under the mass balance model (Appendix [2]), which is called data redundancy.
Determined data (not adjustable, for example, by weighing systems, etc.): when the flow rate has already been measured and cannot be corrected by the data integration method.
2- Unmeasured data
Observable data [3]: A data is visible when it can be estimated using the mass balance model and measured values ??in steady state.
Unobservable data [4]: ??An unobservable data is when it cannot be estimated using available measured data and steady-state mass balance equations..
Adjusting the data [5] as part of a mass balance problem may only include the correction of known data (tonnage and grade) or before correcting the data, it may also calculate the unknown tonnage and grades [2]. Mass balance in processing circuits and the expression of different methods of adapting data in a stable manner [6]. An operational unit is stable with respect to operational variables; If its variables do not change with time. For example, in a crushing unit in steady state, the dimensional distribution of the feed and its flow should not fluctuate with respect to time. If it is said that the Asia is working in steady state, no operational variables of the Asia should change with respect to time and therefore, the Asia product should also be obtained with a constant flow rate and dimensional distribution. In the robust methods that are the subject of this research, depending on the form of the condition equation, the adaptation problem may be linear [7], bilinear [8], or nonlinear [9]. Considering that the conditions necessary to solve the problem of matching the data are written in the form of equations or inequalities, these conditions can be separated as follows]2[:
1- Equation (conditions of equality)
Solid survival: considering that this condition, in the form of the product of the connection matrix [10] of nodes and currents, in the matrix of solid tonnages, which is considered a variable, It is written, it is linear.
Metal survival: it is written as the product of the matrix of coefficients in the matrix of tonnages and grades (both variables) and therefore it is considered bilinear.
2- Inequality: considering the conditions as an inequality.; For example, the rate of the feed flow is smaller or equal to the rate of the concentrate flow. In general, the inequality conditions make solving the data matching problem non-linear. Therefore, by involving inequalities as inequality conditions, a nonlinear state occurs. According to the above concepts, the problem of fitting the data is linear when the models are linear and all the variables are measured, or there is an unmeasured variable. In the case that the variables are not measured, the so-called linearization [11] is done to make it easier to solve the problem of adapting the data, which is examined in detail below. Therefore, the bilinear method is considered a type of non-linear method and it is non-linear when there are unequal conditions or inequalities in the problem [2]. In this research, the methods used to adjust the measured data in a steady state and in a non-linear way are: analytical, classical, meta-heuristic methods (genetic algorithm and hybrid method). Each of these methods consider conditions for solving the problem numerically and try to minimize the sum of squares function or the objective function [12] and minimize the error of establishing the conditions [13] in order to adapt the data or the error for which the conditions are established in solving the minimization problem. In the next step, taking into account the unmeasured data, data adaptation is done using Crow's bilinear methods, matrix projection and Simpson[14]. It should be noted that Crowe and Simpson's method solves the problem by using appropriate linearization.
In order to check the sensitivity of the mass balance to the data according to their information content, some types of data sensitivity analysis adapted using the standardized correction value of the data and the degree of skewness and deviation of the data from the standard mode [1] have also been introduced in this chapter. Also, in this chapter, the variogram method is introduced in order to determine the number of samples required in different parts of the processing circuits, taking into account the level of engineering confidence and determining the error of different sampling methods, including systematic, random and random[3], and examples are provided at the end.
In order to control the operation and adjust it in an ore processing plant and in order to achieve the right conditions, it is necessary to examine the load in different paths of the plant, qualitatively and quantitatively, according to the plan. This requires having samples that represent the load in those routes [1].