Risk Assessment of Pile Foundations Using an Efficient Hybrid ANN Paradigm Compared with Monte Carlo and Subset Simulations

The presence of uncertainty and variability in soil and sub-soils is a fundamental aspect of pile design. Extensive study has been conducted to accurately measure the reliability or likelihood of structural failure. This work investigates the suitability of Monte Carlo and Subset simulations for the risk assessment of piles. In addition, first-order second-moment method (FOSM)-based hybrid artificial neural network (ANN) frameworks were used to automate the process of risk assessments of piles. Specifically, five hybrid ANNs were constructed using swarm intelligence algorithms. A comparative analysis of Monte Carlo, Subset, FOSM, and FOSM-based hybrid ANN was conducted at different coefficient of variation levels. The effectiveness of the utilized hybrid ANNs was determined using diverse statistical indices. Based on the performance, the best-fitted hybrid ANN was selected and utilized for risk assessment of pile foundations. According to the results, the employed ANN-MPA framework exhibits the best-fitted estimation with 99.2% (R² = 0.9920) accuracy. The probability of failure was subsequently determined using Monte Carlo, Subset simulation, FOSM, and FOSM-based ANN-MPA methods. According to the results, the FOSM-based ANN-MPA approach can be considered as an alternative tool for quick estimation of risk assessment of pile foundations under different coefficient of variations levels.