Application of Neural Networks to Evaluate Factors Affecting Drilling Performance

Abstract

Achieving the highest Rate of Penetration (ROP) with the least possible Bit Tooth Wear Rate (BTWR) is the aim of every drilling engineer when selecting a drilling bit. Predicting the optimal ROP has become increasingly important given the rise in expenses involved in drilling a well. This has meant that oil companies engage in a perpetual struggle to predict the optimum rock mechanical property parameters. Predicting optimal rock mechanical property parameters, specifically Rate of Penetration (ROP), has become increasingly important given the rise in expenses involved in drilling a well. The prediction of ROP from the current available data is an important criterion for reduction of drilling costs. ROP represents rock bit interaction which relates rock compressive strength and bit aggressivity. ROP prediction is complex because of the numerous variables which lead to difficulties in evaluating drilling parameters. Several models and methods have been published for predicting, and therefore potentially optimizing rate of penetration. However, these models and methods have limitations, too many variables are included, their input parameters are often not readily available, and their relationships are complex and not easily modeled. Therefore, the application of Neural Network is suggested in this study. A new methodology has been developed to predict the rate of penetration using the Artificial Neural Network (ANN). Three case studies representing different formations in Kuwait have been conducted to investigate ROP prediction for various applications. These cases have investigated the prediction of ROP for a specific heterogeneous formation (CASE I); a semihomogenous formation (CASE II); a drilling section composed of a heterogeneous formation and for a drilling section composed of a complex heterogeneous set of formations (CASE III). Predicting ROP parameters is of particular interest, therefore finding a new method to predict ROP for the cases investigated in this study will be a valuable achievement. Application of the new network models would then be used for selecting the best parameters for an optimal drilling strategy based on field data. In addition to the prediction of ROP, several runs were carried out to predict Tooth Wear Rate (TWR) for a drilling section in case III. Rock bit interactions in the field as a function of rock mechanical property parameters was achieved by predicting ROP which relates to rock compressive strength and bit aggressivity; as well as TWR which relates to rock abrasiveness and wear resistance. History of bit runs, mud logging data, geological information, offset well bit records, drill bit characteristics, and wireline data all play an important role in the prediction of rock bit interactions in this study. Based on field data, the prediction of rock mechanical property parameters can be accomplished by the use of a neural network as an alternative prediction and optimization method. Neural network offers a new form of information processing that is fundamentally different from a traditional processing system. The system uses a knowledge base of various drilling parameters, to produce a “correlation” description of the optimal Rate of Penetration.