SEGMENTATION OF DRILL HOLE ZONES USING AUTOMATED CHANGE POINT DETECTION

The segmentation of drill holes into distinct zones to form an understanding of the intersected geology is critical in the understanding the relevant geological architectures and processes. An algorithm has been developed that applies a neural-network-based change point detection (CPD) model to continuous X-ray fluorescence (XRF) data produced during scanning of drill hole core, to automatically identify changes in geochemistry along the drill hole that indicate changes in lithology, alteration and mineralization.

The algorithm identifies change points in the XRF data by considering each depth as a change point candidate and calculating a dissimilarity score of the data immediately preceding and following the depth in question. Depths having large dissimilarity scores calculated for the data surrounding them are identified as being likely geological boundaries, thereby partitioning the drill hole and defining a zonation. A CPD model using parameters calibrated to logged holes can be deployed on data from unlogged holes to instantly provide a consistent data-driven zonation framework.

The algorithm can be run multi-dimensionally, producing a boundary set from as many variables as is desired, at once. It can also be run multiple time uni-dimensionally, with the resulting boundary sets from separate runs able to be combined during post-processing into a single boundary set. During post-processing, input variables can be assigned weights such that their influence on the final, combined boundary set can be tuned to the user’s desire. After tuning of weights has provided the desired result, the analysis pipeline can be quickly deployed for any number of drillholes, quickly providing a geological framework for each drill hole that can inform and enhance the geologists’ subsequent analysis efforts.