Deussen, Philipp; Galvanin, Federico; (2023) A new approach to targeting drilling locations: quantifying geological knowledge in drilling campaigns using model-based design of experiments approaches. In: Proceedings of the Geometallurgy Conference 2023. (pp. pp. 121-134). The Southern African Institute of Mining and Metallurgy (SAIMM): Stellenbosch, South Africa.

Preview

Text Duessen_10_550-Deussen.pdf Download (634kB) | Preview

Abstract

The performance of Kriging models depends on their semivariogram and the estimated model parameters. To maximise sampling efficiency, common sampling objectives target high expected grades, minimise Kriging variance or both. An alternative method, model-based design of experiments (MBDoE), selects sampling locations based on the sensitivity of a model to input changes, such as parameter values or locations, at which a new sample has an outsize influence on the model predictions. This maximises the information gain with respect to the model. In this paper, the efficiency of MBDoE is compared to a random design and a pitfall of MBDoE, lack of exploration, is addressed. A 2-D concentration profile was generated in-silico with a spherical correlation structure. With 20 samples, the MBDoE procedure correctly identifies the model from two candidates with 85% confidence and estimates two of three parameters correctly and with statistical significance. The random comparison fails to identify the correct model, successfully estimates only one parameter and underperforms in a variance-based exploration metric by up to 17%.

Download activity - last month

Export as JSONCSVXML

Download activity - last 12 months

Export as XMLCSVJSON

Downloads by country - last 12 months

Export as XMLJSONCSV

Archive Staff Only

View Item