Design of Soft Sensor for Industrial Antimony Flotation Based on Deep CNN

Froth flotation is the most widely used method of mineral separation. The froth state reflects the information of mineral content in the froth. Therefore, the most common method is to evaluate the froth surface state to control the operation to improve the mineral recovery. However, it is still a challenge to evaluate the froth state in real time, efficiently, quickly and accurately. This paper uses a deep convolutional neural network (CNN) to design an online antimony flotation process detection system, instead of traditional artificial observation or on-stream analyzers (OSA), is used for real-time detection of froth grade and abnormal condition. And through transfer learning, the detection system is trained and tested using samples from the antimony flotation site. Experimental results show that the accuracy rate of froth state detection reaches 99.17%, which is higher than the accuracy rate of human classification. Therefore, the deep CNN model can identify the froth state more reliably.