Sample-wised probabilistic stability analysis of circular shafts using the Atom Search Optimization-based Artificial Neural Network

Circular shafts are widely concerned in the construction of urban underground infrastructures and their stability is essential. Uncertainties and randomness in soil properties always exist, therefore probabilistic analyses will allow to obtain more reliable results than deterministic ones. In this study, a sample-wised probabilistic approach SPAA based on the Atom Search Optimization (ASO)-Artificial Neural Network (ANN) model, is proposed to analyze the circular shafts stability considering the soil parameters variabilities. The deterministic analysis is carried out by a Finite Element Limit Analysis. The ASO-ANN surrogate model with self-adaptive convergence has high efficiency. It is able to replace time-consuming numerical simulations. The initial samples are generated by the Latin Hypercube Sampling method and the iterative samples enrichment allowed searching the most representative points for the ASO-ANN model construction. Monte-Carlo Simulations and a Global Sensitivity Analysis are then performed to provide several valuable results, which include the failure probability, probability density function, cumulative distribution function, statistical moments of the system response and sensitivity index of each random variable. Two low-dimensional random variable cases and a high-dimensional random field problem are then considered and discussed based on the proposed hybrid SPAA approach. The results indicate that the proposed SPAA outperforms the existing methods since it requires fewer deterministic simulations with guaranteed results accuracy, particularly for the high-dimensional case.

Automated summary

In this study, a sample-wised probabilistic approach SPAA based on the Atom Search Optimization (ASO)-Artificial Neural Network (ANN) model is proposed to analyze the stability of circular shafts considering soil parameter variabilities. The results indicate that the proposed SPAA outperforms the existing methods, requiring fewer deterministic simulations with guaranteed results accuracy, particularly for high-dimensional cases.