The project focusses on developing preliminary protocols for the processing of wireline and geological big data, then train specialised machine learning lithology classifiers. The classifiers shall automatically interpret borehole wireline logs from a well characterized borehole into the upper oceanic crust with targeted high Correct Classification Rates (CCRs>90%). They should be trained to successfully generalize in their lithology classifications when tested against geological knowledge based on drill core observations. The achievement of best lithology classification can be reached through accurate matching of the big data generated from wireline logging and core recoveries. Such correct data modelling by the classifiers can be achieved only through strong understanding of the domain knowledge for the labelling of the training datasets. The big data concerned in the learning experiments can be semi-structured and multi-spectral in most cases. Hence it exhibits high complexity, unbalancing, asynchronicities and more likely some gaps. Machine learning approaches for lithology classification techniques may need to be supported by additional domain knowledge reasoning models based on fuzzy logic to overcome persistent mis-classifications (low CCRs). In recent years (Sabeur et al, 2013, 2015) have worked on operational drilling data, under the TRIDEC (2010-2013) project, with optimised classifiers that sustain big data complexity, unbalancing and classes inseparability challenges.
G. V. Veres and Z. A. Sabeur (2015). Data Analytics for Operational Drilling States Classifications. 23rd European Symposium on Artificial Neural Networks, In Computational Intelligence and Machine Learning, 22nd -24th April 2015, Bruges, Belgium. ISBN 978-287587014-8.
G. V. Veres and Z. A. Sabeur (2013). Automated operational states detection for drilling systems control in critical conditions. 21st European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning, 24th-26th April 2013, Bruges, Belgium.