Multi-condition flotation process condition recognition method based on deep learning and multivariate data pre-processing

The state and trend of flotation surface foam play a crucial role in flotation operations as they provide important indications of working conditions, index predictions, and operational volume adjustments. While researchers have focused on manual feature design of froth images or automatic extraction of depth features and their relationship to flotation conditions, the information contained in the flotation process variables and the importance of data pre-processing are usually overlooked. To address these shortcomings, this study proposes an integrated framework for flotation process data preprocessing and condition recognition. It includes a fast imputation algorithm based on multi-condition data center of mass (FCKNN) for overcoming the challenges of imputing missing values in on-line industrial processes, as well as a high-precision work condition recognition method that fuses deep visual features with key process variables. Finally, the framework was applied to the gold flotation process. The experimental results show that the proposed framework can perform work condition recognition with high accuracy and solve the challenge of online missing value imputation.