Optimisation of deep mixing technique by artificial neural network based on laboratory and field experiments

Ground improvement techniques are inevitable for weak soils that cannot endure the design load imposed by superstructures. Deep mixing technique (DMT) as one of these methods is promising and effective when a deep soil layer with low bearing capacity is encountered. Such deposits are quite common in the South-west of Iran where the studied site is located. In order to validate the influence of DMT on the enhancement of strength, both in-situ and laboratory tests were conducted. Afterwards, a parametric study was carried out to investigate the influence of key factors including cement content, water-cement ratio, curing time and plasticity index (PI) on the performance of DMT. In summary, a total of 192 different conditions were examined in this study by using two methods of 3D plotting and artificial neural networks (ANNs) as the optimisation tool. Results proved the importance of water-cement ratio as a key parameter in DMT. Based on the trained networks, ANN was revealed to give satisfactory predictions on the strength of an improved soil with different admixture conditions. More important, the optimisation made by ANN could determine the specific values for selected key admixture factors to reach a desired strength level with the coefficient of determination higher than 0.85.