Reliability assessment of pipelines crossing strike-slip faults considering modeling uncertainties using ANN models

•High accuracy ANN models were developed using a database generated from FE analyses to calculate the tensile and compressive strains of steel pipelines subjected to strike-slip fault movements;.•It is observed that ANN models with very high coefficients of determination can be significantly different from the actual values;.•A reliability-based assessment framework based on model uncertainty is proposed;.•The probabilistic strains provide a better measure of assessment than the deterministic strains, accounting for the modeling uncertainties. Buried pipelines are crucial infrastructure that sometimes suffers from the adverse effects of ground movements resulting from the earthquake. A pipe crossing strike-slip fault can experience high tensile or compressive strains due to fault displacement during the earthquake. Various analytical and numerical methods are available in the literature to calculate the pipe wall strain through modeling the pipe-soil interaction. These methods are often too complex and not suitable for performing extensive parametric studies. In this research, Artificial Neural Network (ANN) based methods were developed for calculating the peak tensile and peak compressive strains. Due to the lack of field data, finite element analysis was performed to investigate the complex behavior of pipe under the fault movements and generate a database to develop the ANN models. The study revealed the weakness of ANN models that although a very high coefficient of determination (i.e., R2 ≈ 1) and low error can be obtained, the strains predicted using these models can significantly be different from the actual strains due to model uncertainty. To this end, the random strains were predicted by adding the calculated strains with the randomized model errors. Using the random strains, calculating the probability of exceedance of strains was proposed. Finally, a framework is developed for failure probability assessment of pipelines subjected to fault movements accounting to the uncertainty of ANN models.