Froth Features Based Intelligently Optimal Setting For Level Of Bauxite Flotation Rougher Cells

Abstract:Froth flotation is the most important mineral extraction method in mineral separation process. The level of bauxite flotation cells is usually set by artificial experience in actual flotation process of production, but this method has subjective and arbitrary, so the level has large fluctuations and unqualified grade of concentrate and tailings are also caused, and it is difficult to set the level of flotation cells optimally according to process index. To this end, an intelligently optimal setting method for level of bauxite flotation cells based on multiple froth image features is proposed.On the basis of analysis of working principle of flotation cells and relationship between level and froth image features, the pre-setting model based on CBR, the improved LS-SVM grade prediction model based on multiple froth features, and the self-learning fuzzy reasoning intelligent compensation model based on BP neural network are organically integrated, so the intelligently optimal level’s setting model is built by which the froth images’features are fully used is proposed. The research contents and main innovation points of this paper are as follows.(1) Because that the mechanisms of flotation process are complicated, process index such as concentrate grade is difficult to be measured online, the accurate mathematical model relationship among feeding conditions, process indexes and set point of flotation cells is difficult to built, the pre-setting method of flotation cells’level based on CBR is given. Firstly, the objective grades of flotation cells, the ore’s feeding conditions, and the correspondingly historical level set points are used to make up historical operation cases. Secondly, matching degrees of these historical cases are judged based on similarity threshold that has been set, and then they are reused to act as case solutions of the currently input working conditions’features to go out, so the pre-setting of flotation cells’level is achieved.(2) Considering grades of concentrate and tailings are difficult to be measured in actual flotation process, necessary index data information are unable provided for optimally setting of level online, an improved LS-SVM prediction method using multiple froth images’features that can reflect flotation production status directly. This method solves the problems of larger predictive error and poorer robustness and generalization ability of traditional LS-SVM modeling method, so the accuracy of online prediction of grades of concentrate and tailings improves.(3) Flotation process is very complex, unknown interfering factors that influence its normal operation are numerous, and the feeding ore conditions will also fluctuate. Therefore, the pre-setting level of flotation cells doesn’t meet the requirements of real-time working conditions sometimes, and then it causes grades of concentrate and tailings beyond the scope of process objectives. In order to solve this problem, a self-learning fuzzy reasoning intelligent compensation model based on BP neural network is proposed to compensate the pre-setting level of flotation cells. Compared with the RBR method, this method has higher reasoning efficiency and self-learning ability, it effectively realizes the compensation of pre-setting level of flotation cells.(4) Combining a bauxite flotation actual production process, an optimal setting system of rougher flotation cells’level based on the method proposed in this text is developed. The effectiveness and feasibility of this method is verified by actual production results. Figure...