Strength and Compaction Behaviour of Fly Ash–Stabilised Red Soil Reinforced with Biaxial Geogrids: Experimental Investigation and Artificial Neural Network Modelling

Abstract

This study evaluates the combined effect of Class F fly ash stabilisation and biaxial geogrid reinforcement on the compaction and strength behaviour of red soil intended for use as a flexible pavement subgrade, and develops an artificial neural network (ANN) model of the reinforced system. Red soil obtained from the Bangalore University campus, Jnanabharathi, was blended with fly ash at dosages from 2% to 60% by dry weight for compaction testing and from 2% to 20% for strength testing, and assessed through Standard Proctor, California Bearing Ratio (CBR) and Unconfined Compressive Strength (UCS) tests. A fly ash content of 10% gave the highest maximum dry density, UCS and CBR and was adopted as the optimum dosage; it raised the soaked CBR at 2.5 mm penetration from 1.40% to 9.31% and the unsoaked CBR from 4.66% to 14.43%. Three biaxial polypropylene geogrids of differing tensile strength and aperture size — 40 kN/m (38 mm aperture), 20 kN/m (38 mm aperture) and 4.07 kN/m (8 mm aperture) — were introduced into the optimally stabilised soil as single layers at H/2, H/3 and H/4 and as a double layer at H/4 from both faces, where H is the specimen height. Because the aperture of two geogrids exceeded the diameter of the conventional 38 mm UCS specimen, a large-scale mould of 100 mm internal diameter was fabricated so that representative soil–geogrid interlock could develop. Reinforcement at H/4 was the most effective single-layer position for all three geogrids, and double-layer reinforcement produced the largest gain, raising the UCS of the stabilised soil from 300 kPa to 565 kPa. The 4.07 kN/m geogrid outperformed the two higher-strength products at every position. A feed-forward ANN with two hidden layers, trained on the thirteen large-scale UCS configurations and validated by nested leave-one-out cross-validation, predicted UCS with R² = 0.969 and RMSE = 16.4 kPa against R² = 0.671 and RMSE = 53.3 kPa for multiple linear regression, confirming that the reinforced response is strongly non-linear. Sensitivity analysis of the trained network ranked the number of reinforcement layers as the dominant variable and, among the geogrid product properties, placed aperture size above tensile strength. A second network used as a continuous surrogate over the fly ash dosage range located the CBR optimum at approximately 9% fly ash. The results support the combined use of a low fly ash dosage and a small-aperture biaxial geogrid for improving weak red soil subgrades, and indicate that specifications written around geogrid tensile strength alone may select a less effective product.