Unearthed Denver 201722 Sep - 24 Sep
From Idea to Prototype on Resource Challenges in just a Weekend
This competition has finished.
Come rock with us, join the community
program you join.
resources industry. All are welcome.
Man vs. Machine: Sort the Rocks
Ore sorting challenge: visual sorting.
The challenge at hand is to develop an image recognition system that could rapidly detect and classify material as waste.
Barrick's Lumwana operation mines and processes copper ore. 25M tonnes of material is processed through the Lumwana plant each year, half of which is waste material barren of copper. Material containing less than 0.05% copper is considered waste, and processing this unnecessarily is a cash flow negative exercise. Ore sorting is a process where the waste material is removed before the ore is processed, and is currently not used at Lumwana. Ore sorting could be a value add at Lumwana if it could avoid the expense of processing waste ($2.70/tonne) without losing any copper (current margin is ~ $1.05/pound). The current margin is the current price of copper less the all-in cost of production. Barrick’s Lumwana operation has recently conducted some trials to test the effectiveness of using X-Ray sorting technology compared to visual sorting. The sample material was sorted into copper bearing ore and barren waste. In none of the X-Ray trials, was the sorting machine able to match the sorting ability of a geologist using just visual recognition. The challenge is to develop a platform that can rapidly detect the waste from the ore using image recognition.
Potential Areas to consider:
•Could you use machine/deep learning methods to recognize ore and waste material in images?
•Could you use a standard camera, a split camera, multiple cameras or an NIR camera to capture distinguishing surface properties?
•Consider how your system might be scaled to apply to the Lumwana plant, processing 25M tonnes per year?
•Conceptually, what would the implementation of your system look like?
• Training dataset composed of 50 pictures of rock chips laid out in a grid pattern and pre-classified as ore and waste by a geologist.
• Use cameras on real samples to increase size of initial dataset:
1kg Sample of Waste (<0.05% Cu)
1kg Sample of Product (>0.3% Cu). The average rock chip size of both samples is +15 -25 mm rocks.
1 large ore and 1 large waste stone
• Blind dataset of 100 rock chips to test your solution.
• A financial formula is provided to determine the financial impact of ore sorting on Lumwana’s operations.