Mine drilling performance analytics: Transforming raw operational data into engineering insight through advanced visualisation techniques

Mining operations generate massive volumes of drilling data daily, yet most of this data remains underutilised for engineering decision-making. This paper presents a comprehensive visual analytics framework that transforms raw mine drilling operational records from the Kalumbila Mine, operated by First Quantum Minerals Limited in Northwestern Province, Zambia, into actionable engineering insight through the systematic deployment of advanced visualisation techniques. Using a real-world dataset comprising 748 drill holes across 6 operators, 5 drill rigs, and 13 data attributes from blast pattern Main-PitB1136, we address the critical analytics gap between field-logged records and informed engineering decisions. The framework combines star plots, treemaps, node-link diagrams, diverging colourmaps, word trees, spatial problem maps, and an experimental Chernoff-face comparator. Results show that 59.6% of holes were under-drilled (443 of 743), with mean depth deviation significantly below zero (d-bar = −0.110 m, t = −9.66, p = 7.14 × 10 −21 ). All holes in the pattern share a single material class, so the operator comparison is material-controlled by construction. An interquartile-range screen excluded 50 of 699 operator-assigned records with parsed penetration rates, including three physically implausible extreme values; after screening, mean rates differed by 1.30× across operators [ F (5, 643) = 14.74, p = 1.05 × 10 −13 , eta-squared = 0.103]. Screening reduced the apparent contrast while increasing F more than fourfold and decreasing the ANOVA p -value by more than ten orders of magnitude, illustrating that unscreened arithmetic means are unsafe for operator benchmarking on field-logged data. Field-logged drilling issue comments appear to form exploratory clusters when plotted on a hole-identifier-derived ordinal grid. However, because this grid is not based on surveyed spatial coordinates, the apparent clusters must be validated against actual surveyed coordinates before they can be interpreted as physical ground-condition zones. The pipeline was implemented using Python (Matplotlib, Seaborn, Plotly, Squarify and NetworkX), Power BI for dashboard prototyping, and Excel for initial data staging. Our results demonstrate that combining complementary visualisation paradigms – glyph-based, hierarchical, network-based, and spatial – yields insights inaccessible through any single technique or tabular analysis alone.