Estimation of California Bearing Ratio of Improved Peat Soils by Artificial Neural Networks

Nowadays, the use of deep soil mixing method has expanded to improve road pavement. One of the most important goals of this approach is to increase the California Bearing Factor and reduce the pavement settlement. In recent years, modeling by computational intelligence has found a special place in civil engineering, and partly due to the use of these methods, the behavior and process of stabilization, which has been encountered with many problems, is possible. The main objective of this research is to develop a computational model for estimating the California Bearing Factor for peat soils. For this purpose, firstly, this soil was mixed with various percentages of cement and well-graded sand based on validated standards, and experimental tests such as single-axial compressive strength and California bearing were performed on the stabilized specimens. After laboratory testing, a set of information was compiled to construct the computing intelligence model. In this research, the multi-layer (MLP) with different architectures including one and two hidden layers with different number of neurons were used for estimation. For this purpose, input parameters of uniaxial compressive strength, curing time and the amount of sand were considered and the sensitivity analysis was carried out using the Garson algorithm. The experimental results of this study showed that by increasing the amount of sand as a natural filler a significant effect on California's bearing ratio was observed. For example, in a constant amount of cement of 400 kg/m3, adding 200 kg/m3 sand, increased the CBR from 31% to 59%. Also, the results showed that the best model with an average mean square error of 0.41 and an average regression coefficient of 0.99 showed the best performance in estimation of the California Bearing Factor.