Using artificial neural network and genetics algorithm to estimate the resilient modulus for stabilized subgrade and propose new empirical formula

• Published in: Transportation Geotechnics Vol. 24; p. 100358
• Main Authors: Hanandeh, Shadi, Ardah, Allam, Abu-Farsakh, Murad
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.09.2020
• Subjects: Genetic Algorithms; Neural Network; Resilient modulus; Soil properties; Soil stabilization; Soil stabilization; Neural Network; Soil properties; Genetic Algorithms; Resilient modulus
• ISSN: 2214-3912; 2214-3912
• DOI: 10.1016/j.trgeo.2020.100358

This paper presents the results of using rigorous modeling artificial neural network and genetic algorithm to examine the proper stabilization of very weak subgrade soils at high moisture contents. The experimental database was performed in Louisiana transportation research center for four types of soft soil, 125 samples data were prepared and used in development ANN and genetic algorithm models. For two models, the input variables include eight parameters, namely cement percentage, lime percentage, PI, silt percentage, fly ash, optimum moisture content OMC, moisture content M.C, and clay percentage, the output variable includes resilient modulus for different types of stabilized subgrade. Furthermore, mathematical models were proposed to predict the resilient modulus for stabilized weak subgrade with different types of stabilizer agent such as cement, lime, and fly ash with four different subgrade soil types of different plasticity indices. Besides, the proposed models for estimating resilient modulus for stabilized subgrade were derived by an artificial neural network model and genetic algorithm. The scheme method displayed is a particular process of which resilient modulus for stabilized subgrade can be determined directly. The results show impressive due to obtain a high value for regression for sets of models; we obtained another accurate result for Mr by using Gene expression programming. Following the model design is stablished; the powers and deficiencies of the proposed models are tested by matching the resilient modulus proposed from two models with the resilient modulus extracted from experimental test concerning the R2 values. Further, in the neural network model, an exact assessment was achieved using r2 = 0.97. Genetic algorithm with a coefficient of determination (R2) of 0.95 to determine the resilient modulus of stabilized subgrade. Achievement estimation of the ANN and genetic algorithm pointed out that the theses methods were capable to predict resilient modulus of stabilized with powerful and higher efficiency and outcomes of these models was more conventional to the experimental results. Finally, sensitivity analysis of the achieved models has been performed to examine the impact of input variables on output (Mr) and determines that the cement percentage, lime percentage, fly ash percentage, PI, clay percentage, MC, OMC, and silt percentage are the powerful variables on the resilient modulus of stabilized subgrade.