Application of machine learning in the prediction of compressive, and shear bond strengths from the experimental data in oil well cement at 80 °C. Ensemble trees boosting approach

Abstract

The current study aimed at predicting shear bond strength (SBS) and compressive strength (CS) using ensemble techniques of gradient boosting regression tree (GBRT) from the experimental data. Experimental data were obtained from CS and SBS studies using class F fly ash as supplementary cementitious materials at different proportions. The experimental results showed that the application of class F fly ash increases both CS and SBS with curing time due to the pozzolanic action of the fly ash. The SBS and CS for 15% replacement after 28 days were 0.353 and 41.9 MPa, respectively compared to 0.324 and 39.5 Mpa for 30% fly ash. This means higher fly ash content decreases both CS and SBS. Cement, OWC, water, fly ash, curing time, and dispersant were set as input data for machine learning (ML) while experimental SBS and CS as output. ML results showed that GBRT overperformed Artificial neural network (ANN), support vector machine (SVM), and Gaussian process regression (GPR)models since it gave the greatest R2 = 0.995 for CS, 0.989 for SBS and the least loss functions (MSE = 0.160 , MAE = 0.174), and (MSE = 0.0005 , MAE = 0.0031) for CS and SBS, respectively.. The comparative findings of both experimental and estimation, therefore affirm that for the long life of oil and gas wells, GBRT can be implemented as an improved approach for cement hydration prediction.