Well-Shen has been involved in mining and mineral exploration for the past 10 years. He holds a PhD in Mineral Deposits from Laurentian University, Canada, and completed his undergraduate degree at the University of British Columbia, Canada. He has worked in a variety of exploration to feasibility-stage projects in Canada, the USA, Brazil, and Botswana with experience on various base- and precious metal deposit types around the world in his current role at ALS Geoanalytics. He specializes in mineralogy, geochemistry, and machine learning.
Well-Shen is a geometallurgist and explorationist who routinely works at the intersection of various geoscience disciplines. Using the lens of mineralogy and its effect on geochemical, metallurgical, and physical data, Well-Shen has produced multiple predictive models for geologists, resource estimators, engineers, and metallurgists during his career.
Scaling Geometallurgical Insight using Assays and Photos: Machine Learning Applications on Copper Deposits
Risk reduction in exploration and mining projects requires a good understanding of variability in the underlying geology. Mineralogy and texture are the largest drivers of variability in rocks, directly impacting the drivers of project economics. Multi-element assay data is routinely used as a proxy for mineralogy in trying to estimate rock variability; however, textural information is often poorly quantified during exploration and early mine development. Computer vision is a technology that utilizes the information contained within pixels in routinely collected drillcore photographs through a process called image analysis. In this talk, we demonstrate the application of computer vision and routine assays in reducing project risks on North American copper deposits through geometallurgical case studies on acid-base accounting, throughput, and exploration. Through these case studies, we demonstrate that routine assays can be linked with expensive, less abundant test data to infill data gaps and improve decision making and risk management on projects. We also show that combining image data with routine assays, magnetic susceptibility, and short-wave infrared (SWIR) datasets significantly improves the results of predictive lithology and regression models. The image data can also be used alone to build predictive models with rapid turnaround time relative to other routinely collected datasets. Computer vision is shown to be a valuable, complementary tool for geologists to improve data consistency, continuity, and remove subjectivity from the data collection process. This will in turn allow robust geometallurgical models to be built to reduce risk throughout the value chain.
