Integrated prediction model of bauxite concentrate grade based on distributed machine vision

• The global and local trends of froth characteristics of different flotation stages are studied.
• A model for prediction of concentrate grade in rougher stage is built.
• The wavelet extreme learning machine algorithm is used for residual compensation model.
• The integrated prediction model based on distributed machine vision has a good generalization capability.
• A good prediction accuracy of concentrate grade is achieved.

Concentrate grade of bauxite flotation is an important technology indicator, which has a direct effect on aluminum quality. Considering the unity, locality and inaccuracy of existing prediction methods of concentrate grade based on machine vision, a distributed machine vision system of bauxite flotation process is built in this paper, from which an integrated prediction model of concentrate grade is presented. At first, we use experimental methods to analyse image data from different flotation stages, as well as comment on the relationship between its global trends and local trends. Then taking advantage of the multiple kernels least squares support vector machine and wavelet extreme learning machine, models for prediction of concentrate grade and its residual compensation are established respectively to predict the concentrate grade through integration. Finally, validation and industrial applications show that the integrated prediction model based on distributed machine vision has a good generalization capability, which can achieve a good prediction accuracy of concentrate grade, with a relative error of less than 6%, thus laying a foundation for optimal control based on mineral grade in flotation process.