Probabilistic assessment of geosynthetic reinforced soil walls using ANN-based response surface method

An artificial neural network-based response surface method is proposed to demonstrate the probabilistic performance of the geosynthetic reinforced soil (GRS) walls backfilled with cohesionless soil. Response surfaces are formed either in terms of performance functions or design outputs of the GRS walls using the uniform design method to achieve better accuracy of the response surface in predicting the reliability of walls. The probabilistic assessment of two GRS walls is performed using the proposed approach. In the first problem, the feasibility and efficacy of the present method on the probabilistic performance evaluation are examined. Also, the effect of variability of different input variables on the reliability index of the wall is analysed. Results show that the soil friction angle is the most sensitive parameter affecting the overall stability of the wall. A well constructed GRS test wall is further assessed using the present approach under the finite difference numerical scheme. The proposed method is proved to be an efficient technique in evaluating the reliability of more complex reinforced soil walls despite having any explicit closed-form solution of the limit state functions. Further, Sobol sensitivity indices of the input variables on the outputs are evaluated for both problems.