Pattern recognition insight into drilling optimization of shaly formations

Slow drilling in deep shale formations leads to a considerable expenditure to the petroleum industry. An important factor that contributes to slow penetration rate is bit balling in water-reactive shale formations. Bit balling is recognized as the major reason for inefficient performance of the bit while drilling shaly formations. The corresponding research centers and industry are always interested in finding practical solutions to mitigate bit balling associated with slow shale drilling. In this study, three well-performing and robust pattern recognition techniques including linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), and neural network pattern recognition (NNPR) are presented for problem identification in drilling engineering. Each method constructs three structures considering important inputs: normalized rate of penetration, depth of cut, specific energy, and cation exchange capacity to diagnose effective bit cleaning in shale formations of the Southern Oil Field in Iran. The models correlate operational parameters to cation exchange capacity to determine whether effective bit cleaning (reversal of balling) or ineffective cleaning (irreversible balling) is taking place. The common evaluation approaches include cross penalty error, confusion matrix output, area under the curve (AUC), and receiver operating characteristic (ROC) to evaluate efficiency of the multi-strategy classifier. These indicators provide useful information on the number of classified and misclassified cases, global accuracy, and discriminatory ability of diagnostic tests. We show that pattern recognition methods can assure both stability and high accuracy in classification situations.