Experimental and machine learning approaches to investigate the application of sugarcane bagasse ash as a partial replacement of fine aggregate for concrete production

Abstract

Sugarcane bagasse ash (SCBA) is produced during the cogeneration of dry sugarcane bagasse in the sugar and ethanol industries. Managing the huge amount of SCBA is a difficult task as it requires a lot of time and effort, resulting in its direct disposal in open landfills and polluting the environment. To address these problems, the present study is aimed at utilizing this waste as a sustainable material. The objective of the current study is to analyze the impact of using SCBA as a partial replacement material for natural fine aggregate (NFA) in concrete production using two methodologies, namely experimental and artificial intelligence approaches. In the experimental approach, the mechanical properties of SCBA-based concrete made by replacing NFA at various proportions (10%, 20%, 30%, 40%, and 50%) are assessed. These properties include mechanical (compressive strength, split tensile strength), and durability (initial surface absorption, ultrasonic pulse velocity, and sulphate attack). Additionally, to analyze the practical usability of the SCBA-based concrete, analytical evaluations such as cost estimation, global warming potential, statistical analysis, and the linear relationship between the parameters have been carried out. In comparison to the control mix, the experimental results showed that a 10% substitution of SCBA for NFA significantly improved compressive strength (14.1%), split tensile strength (24.05%), ultrasonic pulse velocity (0.52%), initial surface absorption (19.2%), and resistance to sulphate attack (11.5%). In addition, the analytical evaluation showed a significant reduction in the concrete production cost and a negligible increase in global warming potential. The statistical analysis and mathematical relationship showed a statistically significant difference in the compressive strength at different curing ages and a strong relationship between the compressive strength and other parameters of SCBA-based concrete. On the other hand, seven machine learning (ML) models (linear regression, artificial neural network, K-nearest neighbor, K-star, decision tree, support vector machine, and additive regressor) were considered to predict the compressive strength of SCBA-based concrete produced by replacing NFA content. The performance of these models has been evaluated using errors, correlation coefficients, and sensitivity analysis. In addition, scatter plots, box and whisker plots, histograms, and correlation matrices are used to visualize the available datasets. The results of the artificial intelligence approach revealed that the additive regressor model performed well compared to other models, with the highest coefficient (0.95) and minimal errors in predicting compressive strength.