Research On Strategy For Reagent Addition Control Of Gold-antimony Flotation Process Based On Froth Image Features

The reagent addition is the most important operating indicator in gold-antimony flotation process. Currently reagent addition regulation of the flotation process is mainly based on the observation of image visual characteristics of froth surface by operators. However, such artificial method shows the drawbacks like strong subjectivity, low control accuracy and repeated regulation, which lead to large fluctuations in product quality. This paper researches on reagent addition automation control in gold-antimony flotation process and a control strategy of reagent addition based on multiple image features is proposed.The simple description with the mean and variance of bubble size easily causes missing bubble size distribution characteristics. Thus the kernel function method is used to describe the probability density function (PDF) of bubble size distribution. As the distribution density of samples is not taken into consideration by the general kernel estimation which uses a unified kernel width, a kernel estimation method with variable width is suggested and the kernel function parameters is identified by maximum entropy method to achieve an accurate description of the bubble size distribution characteristics. Take the square of the difference between the detected bubble size PDF and the desired bubble size PDF as the objective function, which is solved by Powell and differential evolution algorithm (DE) respectively to obtain the desired reagent addition. In order to reduce the costs, dosage costs constraints are applied at the stage of selection of optimal individuals of the DE algorithm.A reagent addition compensation control method based on expert rules is introduced to achieve the expected antimony concentrate grade. The froth color and texture features are utilized for the prediction of antimony concentrate grade, which is compared to the desired concentrate grade and a deviation between them is calculated. The deviation and the trend of image color and texture features are considered as input of the expert controller to get reagent addition compensation value, which is for the compensation of the reagent addition based on the bubble size PDF tracking control. Industrial application verifies that the proposed modified DE with expert control compensation method shows good result in solving reagent addition control.