TY - JOUR
T1 - Machine learning approach to predict the dynamic linear swelling of shales treated with different waterbased drilling fluids
AU - Tariq, Zeeshan
AU - Murtaza, Mobeen
AU - Mahmoud, Mohamed
AU - Aljawad, Murtada Saleh
AU - Kamal, Muhammad Shahzad
N1 - Generated from Scopus record by KAUST IRTS on 2023-09-20
PY - 2022/5/1
Y1 - 2022/5/1
N2 - Oil and gas drilling contractors face many challenges during drilling, mainly related to wellbore instability. The water-based drilling fluids (WBDFs) are mixed with various swelling inhibitors; nevertheless, shale swelling could still take place. To quantify the swelling inhibition potential of WBDF, several laboratory experiments are usually carried out. These experiments are costly, time-consuming, and tedious. This study used machine learning techniques to predict the dynamic linear swelling of shale wafers treated with different WBDF containing different inorganic salts and silicates such as KCl, NaCl, CaCl2, sodium silicates, and aqueous alkali aluminosilicates (AAAS). An extensive experimental study was carried out to collect enough datasets to train machine learning models with different WBDF results. The swelling inhibition potentials were measured using a dynamic linear swell meter. All the WBDF solutions were tested on sodium bentonite clay wafers. The linear swell tests were run on each drilling fluid for 24 – 48 h. In addition to the linear swelling measurement, zeta potential and conductivities of all WBDF prepared with different concentrations were measured. ML techniques such as Artificial Neural Network (ANN), Decision Trees (DT), Random Forest (RF), and K Nearest Neighbor Algorithm (KNN) were utilized. ML models were trained on input parameters such as zeta potential, salt conductivity, salt concentrations, and elapsed time. The output of ML models was dynamic linear swelling in percentage. The results showed that the ANN technique could predict the linear swelling percentage as a function of the inputs, as mentioned above. ANN also proved to work better than the RF, DT, and KNN. Average Absolute Percentage Error (AAPE), Root Mean Square Error (RMSE), and Coefficient of Determination (R2) were used to measure the accuracy of the predicted model. In addition to the trained models, an explicit empirical correlation from the ANN model's fine-tuned weights and biases are also presented. The developed ANN model can efficiently measure the swelling potential of different WBDF and work best on the concentration of inorganic salts in WBDF ranging between 0.1 and 7 wt%.
AB - Oil and gas drilling contractors face many challenges during drilling, mainly related to wellbore instability. The water-based drilling fluids (WBDFs) are mixed with various swelling inhibitors; nevertheless, shale swelling could still take place. To quantify the swelling inhibition potential of WBDF, several laboratory experiments are usually carried out. These experiments are costly, time-consuming, and tedious. This study used machine learning techniques to predict the dynamic linear swelling of shale wafers treated with different WBDF containing different inorganic salts and silicates such as KCl, NaCl, CaCl2, sodium silicates, and aqueous alkali aluminosilicates (AAAS). An extensive experimental study was carried out to collect enough datasets to train machine learning models with different WBDF results. The swelling inhibition potentials were measured using a dynamic linear swell meter. All the WBDF solutions were tested on sodium bentonite clay wafers. The linear swell tests were run on each drilling fluid for 24 – 48 h. In addition to the linear swelling measurement, zeta potential and conductivities of all WBDF prepared with different concentrations were measured. ML techniques such as Artificial Neural Network (ANN), Decision Trees (DT), Random Forest (RF), and K Nearest Neighbor Algorithm (KNN) were utilized. ML models were trained on input parameters such as zeta potential, salt conductivity, salt concentrations, and elapsed time. The output of ML models was dynamic linear swelling in percentage. The results showed that the ANN technique could predict the linear swelling percentage as a function of the inputs, as mentioned above. ANN also proved to work better than the RF, DT, and KNN. Average Absolute Percentage Error (AAPE), Root Mean Square Error (RMSE), and Coefficient of Determination (R2) were used to measure the accuracy of the predicted model. In addition to the trained models, an explicit empirical correlation from the ANN model's fine-tuned weights and biases are also presented. The developed ANN model can efficiently measure the swelling potential of different WBDF and work best on the concentration of inorganic salts in WBDF ranging between 0.1 and 7 wt%.
UR - https://linkinghub.elsevier.com/retrieve/pii/S001623612200151X
UR - http://www.scopus.com/inward/record.url?scp=85123008674&partnerID=8YFLogxK
U2 - 10.1016/j.fuel.2022.123282
DO - 10.1016/j.fuel.2022.123282
M3 - Article
SN - 0016-2361
VL - 315
JO - Fuel
JF - Fuel
ER -