Modeling For The Prediction Of The Pulp Density In Bauxite Froth Rougher Flotation Process Based On Froth Image Feature

Flotation is an important part of alumina production process using ’bayer process with dressing’. The pulp density is an important quality indicator of the froth flotation process. When the pulp density is too high, the pulp inflatable degree reduce and the concentrate grade will be low; while when the pulp density is too low, the slurry glow rate increase and the flotation time shorten, so the recovery rate will be low. At present, most of the flotation plant acquire the value of pulp density through offline laboratory, which leads to a long lag time, and it is difficult to continuously detect. Studies have shown that froth image is the indicator of pulp density, so researching real-time detection of pulp density using froth images is great theoretical significance and practical application value.In this paper, on the basis of analysis of bauxite flotation process mechanism, the structures of the bubble image acquisition system was presented, and the bubble’s image features data was preprocessed through deleting outline data and denoising with wavelet. Considering the how to choose the key froth image feature as the input of pulp density prediction model, a decision table was established, which take the pulp density as decision variable and the froth image feature as attribute set. Then use improved information entropy and an approach for attribute reduction in a real domain decision system select the right key bubble image features which are RGB color component, bubble size, bubble speed and bubble stability. Considering the prediction of the pulp density is a kind of modeling for complex industrial process, the nonlinear PLSS model is proposed, according to the theory of partial least square regression, spline interpolation theory and linearization idea. In order to enhance the precision, a prediction method of Boosting-PLSS which adopt the PLSS as the basic learning algorithm in boosting is proposed for soft sensor of the pulp density. The main idea of Boosting-PLSS is to train a sequence of robust PLSS models on various weighted versions of the original training set and then combine the predictions from the PLSS models to obtain integrative results to establish a regression model between bubble features and the pulp density. Simulation results of actual production data in a flotation plant show the proposed method comparing to conventional multivariate statistical models. Figure22, table9, reference64.