Abstract
Large manned or autonomous haul trucks are used to take mined material from the surface of an open pit mine to primary crushers.
In this project, a novel approach is presented that employs the combination of rugged 3D machine vision and supervised deep learning to automatically analyze the load of each haul truck on the haul road. A rugged stereo camera with onboard processing capability and automatic lens cleaning system is installed together with several powerful LED lights on a special purpose gantry looking down at the haul road. Supervised deep learning is applied to the acquired 3D images to automatically determine the presence of haul tucks and to analyze their load with no interruption to their journey, in order to: (i) Measure the fragmentation of the loaded rocks, (ii) Accurately detect presence of the large rocks (boulders) (iii) Sense the 3D profile of the haul truck load and indicate any mis-aligned loading, and (iv) Measure the bulk volume of the material on the bed of the haul truck. The required computing is distributed and performed using a small (credit-card sized) processing unit on board the 3D smart camera as well as on the cloud. The results are then communicated with the secure cloud platform for visualization, boulder notification, and report generation.
Experimental results over a period of 130 days where the system automatically analyzed 120,000 haul truck loads are presented from trials at a major copper mine in South America.
Authors: Enoch Chow, Kevin Joseph, Erica Jung, and Shahram Tafazoli
Presented by: Dr. Shahram Tafazoli