Abstract

This abstract introduces an AI-driven solution designed to improve ore sorting and material handling efficiency through the integration of camera data with a multi-sensor ore sorting systems. Advanced sensor-based technologies equip shovels, loaders, and conveyor belts with X-ray fluorescence capabilities, enabling precise, high-frequency measurements of material grade from excavation through to downstream processing. The primary objective of such system is to classify material at the highest possible resolution, particularly at the bucket level at the mine face, using in-pit sensor-based ore sorting technologies. These bucket-level classifications are then aggregated to assign trucks as either ore-bearing or waste, enabling optimal routing decisions that are integrated with the mine’s Fleet Management System (FMS). Accurately batching and aligning bucket-level classifications with the correct trucks is critical to ensure that material is sent to the appropriate destination. The system relies on FMS data to send truck batching information, which is sometimes incorrect. In this abstract, an alternative approach and solution is described where a camera is used to create a more resilient multi-sensor system that identifies faulty truck batching information using computer vision. The approach of using a multiple integrated and synchronized sensor system to improve resiliency has the potential for broad application for sensing systems deployed in mining industry.

The system works by using an AI pipeline which integrates computer vision within a multi-sensor system. This approach identifies trucks arriving at the shovel, achieving better truck batching results than individual data streams could provide. The identified truck information is then combined with measurements from the other sensors in the system to create high accuracy truck batching of bucket-level measurements. This information can then be applied to identify and resolve issues – increasing the resilience of the overall system. A comprehensive 30-day case study using field data collected on a mining shovel identified that the number of trucks without a valid ore/waste classification could be reduced by 10% using the improved computer vision truck batching solution. Moreover, for trucks with a previously valid aggregated ore grade, this system successfully assigned additional classified buckets to existing trucks, improving the accuracy of 4.7% of truck ore grade measurements. The study demonstrates potential significant improvements in mining operational efficiency and profitability by reducing mis-aggregations in bucket-to-truck batching through AI pipeline integration within the current operation.

This presentation will provide an overview of the methodologies used to develop an AI-driven computer vision model for real time truck ID detection. These techniques are integrated with a robust multi-sensor ore sorting system to review, validate, and correct how sorted material is assigned to the right trucks. The presentation includes a detailed case study and data-driven analytical section to demonstrate how the integration of computer vision with an ore sorting technology can result in a more accurate and resilient system. These advancements demonstrate how integrating AI-driven methods, such as computer vision, with multi-sensor systems like ore sorting technology can significantly enhance material handling, reduce errors, and optimize resource management throughout the mine-to-mill process in the mining industry.