Experimental investigation and machine learning-based prediction of brittleness index in heavyweight cement slurries

In engineering design, the brittleness index (BI) seems to play a significant role in material selection, failure forecasting, and service performance. Determining BI is traditionally a laborious, expensive, and highly experimental process. This study gives a full framework that combines experimental tests with machine learning methods to model the brittleness behavior of the heavy-weight cement slurries. The study is based on different experimental tests, such as Split Hopkinson pressure bar Test, Uniaxial Compressive Strength Test, Brazilian Test and etc. on mechanical and physical properties of ten different slurry formulations and a dataset of 250 experimental observations from these formulations, each with 14 independent input parameters. Fourteen machine learning models were created, and their accuracy and dependability were statistically compared, and a new Equation has been developed for estimating the BI based on the test results. Gaussian process regression and support vector regression were the two most accurate models based on interaction with each of the models; when using the full set of input variables, their R² values ranged from 0.93 to 0.97. When variable selections were applied to the final models, the number of features taken into consideration was lowered to eight, which resulted in further accuracy improvements and R² values ranging from 0.940 to 0.990. Even though every factor affected the BI, EP had the biggest impact; the main goal of some additives was to lessen brittleness. By using more sophisticated machine learning algorithms, this research provides a new method for predicting business intelligence, which lowers the time and expense required and improves decision-making.