Machine learning prediction of the unconfined compressive strength of controlled low strength material using fly ash and pond ash

Abstract

Abstract The sustainable use of industrial byproducts in civil engineering is a global priority, especially in reducing the environmental impact of waste materials. Among these, coal ash from thermal power plants poses a significant challenge due to its high production volume and potential for environmental pollution. This study explores the use of controlled low-strength material (CLSM), a flowable fill made from coal ash, cement, aggregates, water, and admixtures, as a solution for large-scale coal ash utilization. CLSM is suitable for both structural and geotechnical applications, balancing waste management with resource conservation. This research focuses on two key CLSM properties: flowability and unconfined compressive strength (UCS) at 28 days. Traditional testing methods are resource-intensive, and empirical models often fail to accurately predict UCS due to complex nonlinear relationships among variables. To address these limitations, four machine learning models—minimax probability machine regression (MPMR), multivariate adaptive regression splines (MARS), the group method of data handling (GMDH), and functional networks (FN) were employed to predict UCS. The MARS model performed best, achieving R2 values of 0.9642 in training and 0.9439 in testing, with the lowest comprehensive measure (COM) value of 1.296. Sensitivity analysis revealed that cement content was the most significant factor with obtaining R = 0.88, followed by water (R = 0.82), flowability (R = 0.79), pond ash (R = 0.78), curing period (R = 0.73), and fine content (R = 0.68), with fly ash (R = 0.55) having the least impact. These machine learning models provide superior accuracy compared to traditional methods, particularly in handling complex interactions between mix components. The proposed models offer a practical approach for predicting CLSM performance, supporting sustainable construction practices and the efficient use of industrial byproducts. The novelty of this study lies in the development of precise design equations for evaluating UCS, promoting both practical applicability and environmental sustainability.

Keywords

• Unconfined compressive strength
• Controlled low-strength material
• Pond ash
• Fly ash
• Machine learning