Abstract
Machine learning algorithms, when leveraged to use data that mine operators regularly collect but typically do not fully use (e.g., multi-element assay), allow mines to operate more efficiently. Multi-element proxies, for instance, can be developed from assay to help operators segregate mine materials with significantly greater efficiency. Single variable assay data has historically been used in mine planning as a simple proxy during mine planning and in operations to generate volume estimates of mine material metallurgical and environmental properties (e.g., acid-generating potential, grade control). The multivariate proxy developed herein using machine learning workflow resulted in considerable project efficiencies when compared to the single variable approach. For example, a 10% increase in material classification accuracy for a project with 500MT of mine waste, means 50MT more going in the right direction resulting in considerable cost savings for mine operators and increasing efficiencies in permitting, mine planning and operational segregation. The python-based workflow includes exploratory data analysis, supervised classification and interpretation, and supervised proxy analysis for extension of exploration, geo-metallurgical and environmental characterization datasets to the site assay database. The workflow consists of two general steps:1) identify material types (classify or domaining) and 2) extend the classification to the site-wide assay database (proxy or surrogate analysis) for mine planning.
These methods are particularly effective for mine sites where identification of material types is challenging, where material types (e.g., borrow, construction, acid-neutralizing) are in short supply, where operators are running out of room for disposal, where complex ore assemblages exist, or where an overly conservative permitting approach has limited operatory flexibility.