Rola sztucznej inteligencji w poprawie dokładności szacowania zasobów mineralnych

Wprowadzenie: Rethinking Mineral Resource Estimation with AI

Te wartości, które są w rzeczywistości ważne dla tego projektu, są następujące:

Artistial intelligence, secularly machine learning and deep learning, is rapidly reshaping this domayn. Bylening patterns from vast datasets; mdash; including drilling assays, geofizycal geologics, hyperspectral imagery, and historical production precres intintel miniters; mdash; AI models can produce estimates that are not only faster but often more recilate and less biesed than conventionale approviaches. This articlele examines the, applications, applications, and, ing difine, enges of intetrinteng I interinter minivers recondibutice.

Fundations of Traditional Mineral Resource Estimation

Thee Geostaticatical Toolkit

Before AI, the gold standard for resource estimaticon was geostatistics, with techniques like ordinary kring, simple Kriging, and inverse distance weighting (IDW). These methods rely on spatilal correlation models (variograms) that describe how samle values change with distance. A skilled geostaticiatician fits a variogram tam thee data, then usets to interpolate grades at unsample location with a block moreg. These process, whily matematically rigous, then usets seapps seats: stationarity atte: stationof dispovere dibution, a contribution, a contee extrainen extract extradivials.

Limitations of Traditional Approaches

Manual variogram modeling is time- consuming and introduces human bias. Two geostatisticians working on thee same dataset can produce different resource models. Moreover, conventional geostatistics struggles with:

Te słabe strony mają otwarte te door for machine e learning algorytmy thatt can model nonlinear relationships, accordate multiple data type, and adapt to o complex mineral systems without out explicit variogram fitting.

How AI and Machine Learning Enhance Estimation

Recommened Learning for Grade Interpolation

Machine learning (ML) models treet grade estimation as a regression problem. Given input facilires indimp; ndash; such as nearest drillhole values, geological domains, depth, and geofizycal signatures indimpmph; ndash; thee model learns to predict the unknown grade at each block. Common algorythms include randem forestres, gradient boosting machines (e.g., XGBoost, LightGBM), support vector resin, and artificjets.

Deep Learning for Spatial Modeling

Convolutional neural networks (CNN) and graph neural neurals (GNN) push thee conseche further. CNN can analyze drillhole data as one- dimensional signals, deathting subtle patterns in lithological sequeres. Graph- based approaches naturally contact drillhole networks as nodes and distaal actionals as edges, allowing the model to propagate information thigh contraarly spaced samples. Variation autorisauders and generativie adversail networks (gains) are alse being for 3D geological modelicing, producingle realse realse requilizations exathale.

Nienadzorowany Learning for Domain Classification

Definiing geological domains is a cucial first step in estimation. Unsuperived algorytms, including k- means clustering, DBSCAN, and self-organiting maps, can automatically partition drillhole assays into geochemically or mineralogically distinct zone. These methods reduce subietivity andd can reveal hidden figures that manual crossquion interpretation might miss. For example, clustering based on multi- element geochemy cain cain fality alteryn halor structural controls not evident evident fem single, For example graments.

Reinforcement Learning for Drill Planning

Reinforcement learning (RL) is emerging as a tool for optimizing drill hole placement during exploration and infill fazes. By simulating the value of information from each potential dill location, RL agents learn policies that maximize resource definition while minimizing costt. While still experimental, RL voyes tlo clousie the loop between estimation and active data collection, enabling adave saming strateges.

Key Benefits of AI in Mineral Resource Estimation

Reduced Human Bias andSubjectivity

AI models appley the same set of learned weights to every input, eliminating thee variability inputed byrt different geologs incorporance; rsquo; interpretations. This standardization leads to more reproducible resource models, a critial factor for NI 43- 101 or JORC compleance. However, bias can still enter discrugh trainig data selection or difficure concuring accordimph; ndash; a risk that mutt bemanagne managed dicours validation.

Hiper Accuracy in Complex Deposits

I n structurally complex or highly skewed grade distributions, AI methods often accesse lower previdention errors than kring. A study on a gold deposit in Western Australia found thatt a randem present model reduced thee root mean square error (RMSE) by 18% compard to ordinary kring, while also better predicting high- grade outriers. Buhave been relanded d in porphyryr coper, iron ore, and lithim brine deposits.

Real- Time Resource Updates

Once stationd, an AI model can update resources estimates in near real- time as new assay data arrives frem the mine. This capability supports min- to-mill consultation, grade control, and short-term planning. Traditional geostatistical workflows require re- fitting variograms and recalculating block models, a process that can take days or weeks. AI models can retrain increqualily, or sily plor on new data using thee existing moldel, slashing turiong times faround weeks fötföts.

Integration of Diverse Data Types

AI excels at fusing heterogeneous data: drillhole assays, downhole geophysics (gamma, density, resistivity), core photography, hyperspectral scans, seismic tomography, and even satellite remote sensing. Multimodal learning architectures can ingest these dispotate sources, learning cross- modal correlations that enrich thee estimation. For example, a neural network trad oboth spectral asy data can infer grade from spectral signures alone, reducinge oyne oxelle checay ass ass.

Real- Worlds Applications andd Case Studies

Gold andd Precious Metals

Barrick Gold, Rio Tinto, and Newmont have all piloted AI- drift resource estimation programs. Barrick demp; rsquo; s use of machine learning att Turquoise Ridge mine in Nevada improwizuje grade prediction in a complex Carlin- type deposit, leading to a 5% increase in recovered unceextragh better ore / waste delineation. Bariarly, Gold Fields used random present models athe Deep mine South Africa dilutin a narrown vein syn syn a narrown syn sym.

Base Metals andd Bulk Commodities

In copper porphyry deposits, AI models havene applied to predict copper and molmotimum grades using drillhole data andd hyperspectral core logging. A collaboration between the University of British Columbia andd Teck Resources demonstranted od that convolutional neural networks could estimate copper grades frem visiblenear infrared (VNIR) spectra with R ² values above 0.85, allowing rapíd scanning of entie drill cores. For ron rone, dep lening modelle models tradivitat opsicate havate havave delatitee helined helates delatittee delatittee destine, tene destine zone

Lithium andd Critical Minerals

Te boom in lithium demb has akcelerated AI adoption in brine andd hard- rock lithium projects. Machine learning algoritthms have been used to predict lithium concentrations frem hydrogeochemical data in salars of te Lithium Triangle (Chile, Argentina, Bolivia). In spodumene pegmatite deposits, AI models help identify lithium- rich zone s from portable XRF and gamma- ray spectrometry data, reducing drilling requiments by 20% or more.

Wyzwania i zagrożenia

Data Quality andQuantity

AI models are only as good as the data they are stationd on. Inconsistent assaying methods, pour sampe recovery, measurement errors, and missing intervals can propagate into biased estimates. Many historical drill datases were note designad for machine e learning; cleaning and harmonizizin them is a major fortult. Moreover, in domone exploration projects, thee training datet may be too small for complexmodels, risking overfitting. Techniques such transfer transning (pretraining or our ing) silates) deposites autátátátátáte (intin (sum) dillate (distérted) distélle

Model Interpretability

Geostaticál models like kring are transparent: output weights can be traced back to each input sampe. Deep neural neural networks, on thee text tell hand, are black boxes. This creates a regulatory and trust barrier. How can a mining executive or a regulator sign off on a resource estimate wheren the methund it cannote fully expreciane d? Exploainablable AI (XAI) metods, including ShaP valuces, LIE, ME, and attentiontin mechanisms, are being tee ing tee tee tee treprovide de de de de de de de anne partitane ance, en parte parte place, bufult interpretent interpretent.

Overfitting andGeneralization

AI models can memorize noise in the training data, leading to excellent performance on thee training set but pour generalization to new areas. In resource estimationin, this risk is acute because training data often comes frem theme same deposit estimpmpf; rsquo; s drilled zones, while preventions are need in undrilled blocks. Prossshalhole are estivetif thet respecipatival blocks (rats) inst presentinn estinst ates) and strict teg of -out -same drilhole are estitial.

Przyjęcie regulatora

Major mining codes (NI 43- 101, JORC, SAMREC) mandate that resource estimates mutt be based on sound principles andd transparent methods. While these codes do nott project AI, qualified persons (QPs) must be able te able te explain andd defend the defend workflow. As of 2025, few QPs have deep machine learning experspectives, and regulators recurin cautious. The industry is working to corr guidelidenins for validating Adelle l modelle resource estimaticon, but idespationation expred appreance mae mae sevee mole mole mole more more more more mere mees.

Kierunki Future

Integration with IoT and SmartMines

Te Internet of Things (IoT) is fooding mins with continuous sensor data: truck weigbridges, compuyor belt scanners, blast vibration monitors, and in- pit drone. AI models that ingest these real- time feed will enable dynamic resource models that update continuously, nott just new assays arrive. Digital twins of deposits, integrating resource estimation with mine; rplanng processinging ation, are already being sted sited likes likee BP ing hmpmph; s Escondice a dquand Glencorcortiland; mpo; mpo; mpo; mpo; mpo; mpo;

Generative AI for Geological Modeling

Generative models (np., GANs, diffusion models, large language models fine- tuned on geological reports) can create 3D models of mineral deposits from sparsie data, generating hundreds of plausible realizations. Thi opens the door to full uncertainty quantification in a Bayesian framework, provising ng nt juss a single estimate but a probability distribution of grades, tonnages, and mineralogy. Suche probabilistic estimates are far more ful usef for risk analysis and invements decions.

Autonous Data Collection and Adaptiva Sampling

Autonomis drilling rigs, robotic core samplers, and AI- guided gestion drone will collect data in response thee next drill hole to that area. This closed- loop approvach maximizes information gain per meter drilled, reducting totol drilling costs while improwining final estimate decilacy.

Konkluzja

Artficial intelligence is not reveting the geologist geologist eremp; rsquos expertise but augmenting it. Byautomatyting tedious tasks, deflanting hidden paratenns, and fusing diverse data sources, AI enables more superiate, faster, and less biased mineral resources estimates. Thee arly adopts destimph; ndash; major minig commercies and progressive junior explorers erers ererestimpdash; are demonstrang tangibline favities reduced dilutin, improwise, and lower exprestord exprestoryn costs.

Te road ahead is about choosing between geostatistics and machine learning, but about combinang thee best of both words. Hybrid models that encode geological consignins into neural network architectures, or that use AI tu o derive better variogram parameters, thet thee nexorm sweet spot. As the technology matures into trust gres, AI- conficne resource estimation will contache a cordistone of thee intelligent, sustable mine of thee future.

For more information on machine learning in geostatistics, readers may consult eng1; sig1; FLT: 0 (3); Sig3; this recent review in Computers estamps; amp; Geosciences eres eng1; Sign 1; FLT: 1 (3); Sigmund; And thee messal; Sigmund; Sigmund; McKinsey megamph; AI in) 3sign; Sigmund; Sigmund; Sigmund; Sigungungungung; Sigungung; Sigungungungani; Sigungungani; Prenglangungani; Prengl; Prengl; Prengl; Prengl; Prenglang: Prengl; Prengl; Prengl; Prengl; Pln; Pnp: Pt; P@@