Rock Mechanics
Farhad Mollaei; Ali Moradzadeh; Reza Mohebian
Abstract
The important aspects of this study are to estimate the mechanical parameters of reservoir rock including Uniaxial Compressive Strength (UCS) and friction (FR) angle using well log data. The aim of this research is to estimate the UCS and FR angle (φ) using new deep learning (DL) methods including ...
Read More
The important aspects of this study are to estimate the mechanical parameters of reservoir rock including Uniaxial Compressive Strength (UCS) and friction (FR) angle using well log data. The aim of this research is to estimate the UCS and FR angle (φ) using new deep learning (DL) methods including Multi-Layer Perceptron (MLP), Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), and CNN + LSTM (CL) by well log and core test data of one Iranian hydrocarbon field. As only 12 UCS and 6 FR core tests of single well in this field were available, they were firstly calculated, and then generalized to other depths using two newly derived equations and relevant logs. Next, the effective input logs' data for predicting these parameters have been selected by an auto-encoder DL method, and finally, the values of UCS and φ angle were predicted by the MLP, LSTM, CNN, and CL networks. The efficiency of these four prediction models was then evaluated using a blind dataset, and a range of statistical measures applied to training, testing, and blind datasets. Results show that all four models achieve satisfactory prediction accuracy. However, the CL model outperformed the others, yielding the lowest RMSE of 1.0052 and the highest R² of 0.9983 for UCS prediction, along with an RMSE of 0.0201 and R² of 0.9917 for φ angle prediction on the blind dataset. These findings highlight the high accuracy of deep learning algorithms, particularly the CL algorithm, which demonstrates superior precision compared to the MLP method.
Rock Mechanics
Mahdi Bajolvand; Ahmad Ramezanzadeh; Amin Hekmatnejad; Mohammad Mehrad; Shadfar Davoodi; Mohammad Teimuri; Mohammad Reza Hajsaeedi; Mahya Safari
Abstract
Bit wear is one of the fundamental challenges affecting the performance and cost of drilling operations in oil, gas, and geothermal wells. Since identifying the factors influencing bit wear rate (BWR) is essential, and the ability to predict its variations during drilling operations is influenced by ...
Read More
Bit wear is one of the fundamental challenges affecting the performance and cost of drilling operations in oil, gas, and geothermal wells. Since identifying the factors influencing bit wear rate (BWR) is essential, and the ability to predict its variations during drilling operations is influenced by environmental and operational factors, this study aims to develop an Adaptive Bit Wear Rate Predictor (ABWRP) algorithm for estimating the BWR during drilling operations for new wells. The structure of this algorithm consists of a data transmitter, data processor, deep learning-based bit wear rate estimator, and a bit wear updating module. To develop a model for the BWR estimation module, data from two wells in an oil field in southwest Iran were collected and analyzed, including petrophysical data, drilling data, and bit wear and run records. Both studied wells were drilled using PDC bits with a diameter of 8.5 inches. After preprocessing the data, the key factors affecting the bit wear rate were identified using the Wrapper method, including depth, confined compressive strength, maximum horizontal stress, bit wear percentage, weight on bit, bit rotational speed, and pump flow rate. Subsequently, seven machine learning (ML) and deep learning (DL) algorithms were used to develop the bit wear rate estimation module within the ABWRP algorithm. Among them, the convolutional neural network (CNN) model demonstrated the best performance, with Root Mean Square Error (RMSE) values of 0.0011 and 0.0017 and R-square (R²) values of 0.96 and 0.92 for the training and testing datasets, respectively. Therefore, the CNN model was selected as the most efficient model among the evaluated models. Finally, a simulation-based experiment was designed to evaluate the performance of the ABWRP algorithm. In this experiment, unseen data from one of the studied wells were used as data from a newly drilled well. The results demonstrated that the ABWRP algorithm could estimate final bit wear with a 14% error. Thus, the algorithm developed in this study can play a significant role in the design and planning of new wells, particularly in optimizing drilling parameters while considering bit wear effects.
