Two-Branch Convolutional Neural Network Based on Multi-Source Information for Flotation Overflow Identification

The overflow identification of flotation process is important to ensure the flotation quality and improve the comprehensive economic efficiency of the enterprise. However, direct monitoring of flotation froth overflow is difficult due to the strong nonlinearity, information redundancy and uncertainty of industrial processes. To solve the above problems, a two-branch convolutional neural network model (MSI-two-branch CNN) based on multi-source information is proposed. By merging multiple network branches, the features of industrial processes can be extracted effectively. In addition, the one-dimensional time series are transformed using Gram angular domain (GAF) to generate two-dimensional images, which not only can introduce process variables into the overflow recognition model, but also can improve the recognition efficiency by using the correlation between variables. Finally, the method was applied to the gold flotation process. The experimental results show that MSI-two-branch CNN can perform overflow recognition with high accuracy and efficiency in a practical industrial context.