XploreIQ: Machine Learning and Big Data The Successful Use of Algorithms in Exploration Targeting
•
2 likes•28,991 views
Discover the benefits of data science for the mining industry in our presentation from Mines and Technology Helsinki 2019.
Recommended
Recommended
More Related Content
Similar to XploreIQ: Machine Learning and Big Data The Successful Use of Algorithms in Exploration Targeting
Similar to XploreIQ: Machine Learning and Big Data The Successful Use of Algorithms in Exploration Targeting (20)
More from SGS
More from SGS (20)
Recently uploaded
Recently uploaded (20)
XploreIQ: Machine Learning and Big Data The Successful Use of Algorithms in Exploration Targeting
- 1. SGS MINERALS XploreIQ: Machine Learning and Big Data The Successful Use of Algorithms in Exploration Targeting Marc-Antoine Laporte, Global Project Geologist Mine and Technology 2019, Helsinki, Finland
- 2. 2 ABOUT SGS
- 4. 4 SGS MINERALS SERVICES – YOUR GLOBAL PARTNER
- 5. 5 A Brief History of AI 1950 1960 1950: Turing test 1956: Computer game Arthur Samuel checkers program 1956: Dartmouth conference
- 6. 6 A Brief History of AI January 26th, 2018 – Vancouver, B.C.Artificial intelligence in exploration 1950 1960 1950:Turingtest 1951:Computergame 1956:Dartmouthconference 1987: Widespread use of expert systems 1997: Deep blue beats Kasparov at chess 2000
- 7. 7 A Brief History of AI January 26th, 2018 – Vancouver, B.C.Artificial intelligence in exploration 1950 1960 2018 1950:Turingtest 1951:Computergame 1956:Dartmouthconference 1987:Widespreaduseofexpertsystems 1997:DeepbluebeatsKasparovatchess 2000 • Improvement in computation power brought us closer to the dream of an AI. Integration of a wide array of data is now possible, leading the way to new AI enhanced analytical methods. 2005: First DARPA Challenge in Nevada for autonomous vehicle
- 8. 8 A Brief History of AI January 26th, 2018 – Vancouver, B.C.Artificial intelligence in exploration 1950 1960 2018 1950:Turingtest 1951:Computergame 1956:Dartmouthconference 1987:Widespreaduseofexpertsystems 1997:DeepbluebeatsKasparovatchess 2000 2000s: Progress in machine learning and deep learning… IBM Watson System won Jeopardy in 2011
- 9. 9 Exploration in the Era of AI January 26th, 2018 – Vancouver, B.C.Artificial intelligence in exploration
- 10. 10 Exploration in the Era of AI Artificial intelligence in exploration • Mining and exploration companies have now a wide spectrum a data acquisition capabilities ranging from structural, lithological, geochemical and geophysical information. • Mature mining site have hundreds of thousand meters of core logged and legacy information that is no longer realistically be analysed by human alone
- 11. 11 Exploration in the Era of AI January 26th, 2018 – Vancouver, B.C.Artificial intelligence in exploration Machine learningBig data New knowledge
- 12. 12 January 26th, 2018 – Vancouver, B.C.Artificial intelligence in exploration • The machine learning algorithms have been developed to identify correlations between a set of data, the predictors, and a single targeted feature. • This means that an actual set of the targeted feature must exist in the study area so that the algorithm can identify the correlation between this set and the predictors. • This also suppose that the relation between the target feature and the predictors is somewhat constant over the study area. Conceptual Approach
- 13. 13 January 26th, 2018 – Vancouver, B.C.Artificial intelligence in exploration • Many algorithms have been tested but three are currently used in Genesis software and Microsoft Azure: • Specialized phylogenetic algorithm • Boosted decision trees algorithm • Bayesian Gaussian process latent variable XploreIQ System
- 14. 14 January 26th, 2018 – Vancouver, B.C.Artificial intelligence in exploration Specialized Phylogenetic Algorithm • Developed as a tool to map species differences and evolutions based on their genes. • Adapted by SGS Geostat to work on the same principle with rock geochemistry and mineralogical components. • Used during interpretation phase to identify rock families and create lithological groups which serve as a starting point for 3D geological model.
- 15. 15 January 26th, 2018 – Vancouver, B.C.Artificial intelligence in exploration Specialized Phylogenetic Algorithm Data Qemscan XRF ICP Phylogenetic Classification Rock Grouping Modelization
- 16. 16 January 26th, 2018 – Vancouver, B.C.Artificial intelligence in exploration Boosted Classification Trees Algorithm • Boosting build an ensemble of classifier from the different variables and combined simple classifier into a robust classifier • The algorithm is adaptive and aim at lowering the prediction error to the lowest • Decision trees are good for non linear correlation
- 17. 17 January 26th, 2018 – Vancouver, B.C.Artificial intelligence in exploration Bayesian Gaussian Algorithm • Brings the variable in a Bayesian distribution • Predict the variable in a latent space and vet the variable with the best impact on the predictor • The algorithm was design for non linear correlation of variable to predictors • Good to handle missing data
- 18. 18 January 26th, 2018 – Vancouver, B.C.Artificial intelligence in exploration SGS Machine Learning Workflow Boosted Classification Trees Algorithm
- 19. 19 January 26th, 2018 – Vancouver, B.C.Artificial intelligence in exploration Integra Gold Corp. Gold Rush 2016 Targeting challenge – First Prize 500,000$, 1300 teams from 80 countries
- 20. 20 January 26th, 2018 – Vancouver, B.C.Artificial intelligence in exploration Integra Gold Targeting Results
- 21. 21 AFRICAN CASE STUDY ALG (TSX-V) listed Canadian junior mining company. Founded in 2015 Focus on gold exploration
- 22. 22 January 26th, 2018 – Vancouver, B.C.Artificial intelligence in exploration African Gold Project: Tijirit • Current project involve the integration of a vast database on a well developed mining camp. • Identification of targets for open pit as well as underground mining for Au- PGE-Cu-Zn and Ag. • Use of conditional simulation to quantify the risk in order to be more efficient on drilling. Prospectivity 0-10 10-25 25-35 35-45 45-60 ≥60 Drillhole/Trench 2 km Eleonore N
- 23. 23 Artificial intelligence in exploration West African Gold Project: Tijirit Prospectivity 0-15 15-30 30-40 40-50 50-60 ≥60 Eleonore Lily Sophie3 Sophie2 Sophie N Drillhole/Trench NewTargets 1,000 m Drillhole/Trench New Targets Prospectivity 0-10 10-25 25-35 35-45 45-60 ≥60 100 m Eleonore N
- 24. 24 Proposed Workflow Phylogenetic QEMSCAN + GEOCHEM PhyloGeology (using AI) Groups Samples into Classes Interpretation to create stratigraphic model Client’s Input: Interpretation Production data Seismic Etc. QEMSCAN or Geochem Stratigraphic Prediction Model (using AI) New Data Limited set for best predictions REALTIME GEOLOGICAL MODELING PhyloGeology (using AI) Groups Samples into Classes Units + Sub Units Rock changes + Sub Units Machine Learning Model
- 25. 25 Phylogenetic: Mineralogy Data The mineralogy data with x, y and z position of each sample was submitted to the classification algorithm with the following classification tree. Unit 1 Unit 2 Unit 5 Unit 3 Unit 4
- 26. 26 Phylogenetic: Mineralogy Data The mineralogy data with x, y and z position of each sample was submitted to the classification algorithm with the following classification tree. Unit 3
- 30. 30 January 26th, 2018 – Vancouver, B.C.Artificial intelligence in exploration Phylogenetic in the Oil Industry • SGS Geostat used its machine learning capabilities on many targeting project with various commodities including precious and base metals (7 projects completed and 18 months) . • A specialized lithological classification service has also been developed for the oil and gas industry with 95% success to predict lithologies.
- 31. 31 January 26th, 2018 – Vancouver, B.C.Artificial intelligence in exploration • The geology world is al about domains… Can we learn from a domain an applied it the another? • At the moment, the statistical probability distributions between the learning, test and application domains are too different and optimisation of the error is difficult • Not all variable can be threshold for interpretation purposes • The mineralisation model may not be as simple as we think • We might not be looking for the right vector • Target are size and block model volume must be realistic or multiple targeting exercise will be needed Limitation of the Method
- 32. 32 January 26th, 2018 – Vancouver, B.C.Artificial intelligence in exploration • The versatile capabilities of machine learning assisted classification makes it a powerful tools for other analytical work. • Test are on the way to validate the use of machine learning for other activities such as ore reconciliation, geometallurgical modeling, image recognition. • At the moment, all option are possible and the success will drive the ML forward for the mining industry Conclusion
- 33. 33 Questions