Computational prediction of workability and mechanical properties of bentonite plastic concrete using...

Area of Science:

• Civil Engineering
• Materials Science
• Machine Learning Applications

Background:

• Bentonite plastic concrete (BPC) is crucial for dam seepage control, requiring predictable workability and strength.
• Accurate prediction of BPC properties, such as slump, compressive strength (fc), and elastic modulus (Ec), is vital for efficient construction and cost savings.
• Traditional regression models may not fully capture the complex relationships influencing BPC characteristics.

Purpose of the Study:

• To investigate the efficacy of multi-expression programming (MEP) for predicting key BPC properties: slump, compressive strength (fc), and elastic modulus (Ec).
• To compare the predictive accuracy of MEP models against conventional linear and non-linear regression techniques.
• To identify the key factors influencing BPC properties using SHapley Additive exPlanation (SHAP) analysis.

Main Methods:

• Collected experimental data for slump (158 points), compressive strength (169 points), and elastic modulus (111 points).
• Developed MEP models, partitioning the dataset into 70% training, 15% testing, and 15% validation sets.
• Employed SHAP analysis to determine the influence of input variables (water, cement, bentonite, curing time) on predicted BPC properties.

Main Results:

• MEP models achieved high accuracy, with correlation coefficients (R) of 0.9999 for slump, 0.9831 for fc, and 0.9300 for Ec.
• MEP models significantly outperformed traditional regression models in predicting BPC slump, fc, and Ec.
• SHAP analysis revealed water, cement, and bentonite as key drivers for slump; water for compressive strength; and curing time and cement for elastic modulus.

Conclusions:

• MEP is a highly accurate and reliable technique for predicting the slump, compressive strength, and elastic modulus of bentonite plastic concrete.
• The developed MEP models offer a valuable tool for optimizing BPC mix designs and construction processes, leading to enhanced efficiency and cost-effectiveness.
• Machine learning, specifically MEP, provides a pathway for rapid and precise early estimations of BPC properties, supporting improved construction and design workflows.