A Data-Driven Soft Sensor for Mass Flow Estimation

Transporting iron ore by conveyor belts is widely used in the mining industry. The measurement of the mass flow of ore is essential for equipment protection purposes, preventing the conveyor belts from being overloaded, and for operational control purposes, when accounting for the production of equipment during a period of time. The equipment most used in measuring the mass flow of ore are conveyor belt weighers installed on conveyor belts. As they present a high acquisition cost and require a specialized team for maintenance, the installation of these scales is limited to just a few points of a mineral exploration plant. The present work proposes the development of machine-learning techniques to estimate the mass flow of ore in a conveyor belt that is not equipped with a belt weigher. The virtual sensors were designed using current, torque, and motor speed data from a conveyor belt and iron ore flow measurements from a belt weigher installed on a different conveyor. Two of the proposed virtual sensors were implemented in a programmable logic controller of a belt conveyor in a mining area, where it was possible to verify the performance of the virtual sensors in a real situation. As a result, the proposed virtual sensors were able to measure the ore flow with an acceptable error rate compared to a physical belt weigher. The accounting of the production of the proposed sensors also proved to be close to the total production accounted for the physical belt weigher.