Hydrocyclone Selection And Optimization Based On BP Neural Network

The aim of the present study was to explore the relationship between the separation performance and its structure and operational parameters of hydrocyclone on the basis of construction of BP neural network. This thesis established a BP neural hydrocyclone selection model, which could select a right hydrocyclone according to the concentration, granularity, and production capacity. The result of10test samples was:the insertion depth error:17.86%, overflow diameter error:7.51%, underflow diameter error:10.43%, feeding pressure error:20.24%. The BP neural network therefore could be used for the selection of a hydrocyclone, and for optimization of the hydrocyclone parameters, too.The paper applied the BP neural network model in the hydrocyclone’s selection of underflow diameter, overflow diameter, feeding pressure, insertion depth and other parameters firstly, and developed a novel idea for the hydrocyclone selection. Presently in the designing of the coal preparation plant or mineral processing plant, two ways of hydrocyclone selection are practiced, i.e., table checking method and trial method. The drawback of the table method is that it can’t check the accurate hydrocyclone size, and the problem of the trail method is that it would entail too many experiments to be very costly both in money and time. However, the BP neural network model’s error is only about3%-9%when the test error5%-11%is taken out.This paper examined the influence between training sample number and selection accuracy. When the error was large, the number of training samples was found the most important factors. When the selection error was less than10%, the error won’t decline obviously while increasing the number of samples, and the model’s volatility became the most important factor to restrict the error’s drop. The relationship between selection accuracy of BP neural network model and the training sample number showed a positive correlation.The innovation of this paper is the application of BP neural network model in the bionic test to examine the influence of structure, operating parameters on the separation performance, and the analysis was more accurate so as to avoids making lots of experiments that could help save money and time. It is believed that when the number of training samples is more than20%-30%, the selection precision could meet actual needs.The BP neural network selection model’s accuracy is higher than that of the traditional methods, and much faster in the assistance of hydrocyclone selection. It would be very helpful for those designers who do not need to learn complex production or separation formula but still are able to achieve a comprehensive selection of hydrocyclones. In addition, BP neural network could help to select insertion depth, overflow diameter, underflow diameter, feeding pressure simultaneously.The present study has completed the designing of a hydrocyclone with separation size at0.074mm and production capacity that could meet commercial demand. The overflow and underflow product from the selected hydrocyclone could meet wet and dry coal preparation technology requirements. It is therefore believed that it would be very useful to the development of Chinese coal preparation.