Multi-model soft measurement method of the froth layer thickness based on visual features

• The multi-model soft measurement approach for froth layer thickness is proposed.
• Kernel extreme learning machine is used to construct the soft measurement local models.
• The membership degree of different working conditions with visual features is calculated.
• The online global model according to the membership degree of different working conditions is given based on KELM methods.
• The industry data validation results demonstrate that the model has high prediction accuracy.

The flotation froth layer thickness is an important factor for production performance. Influenced by the harsh environment of the production site, the flotation froth layer thickness is difficult to measure accurately. This paper proposes a multi-model soft measurement method of the froth layer thickness based on the visual features. In this method, the froth layer thickness is established by the kernel extreme learning machine (KELM) according to different working conditions. The membership of the current froth image for each of the working conditions is obtained based on the combined similarity coefficient and is used as the combination of weights of multiple sub-models to realize the global soft measurement model. The performance of the soft measurement model is verified by the industry production data. Based on the KELM global model, the root mean square error and the average relative error of predicting the thickness of the froth layer in the experiment are 3.01 and 3.98%, respectively, which show that the model has high prediction accuracy and strong generalization performance. This method provides a new way for froth layer thickness measurement and the basis for controlling flotation process.