Fragility analysis of continuous pipelines subjected to transverse permanent ground deformation

Abstract The structural integrity of buried continuous pipelines can be jeopardized by transverse permanent ground deformation (PGD) induced by landslides. A probabilistic analysis can facilitate understanding the likelihood of damage to pipelines in landslide regions, further minimizing the risk. However, empirical fragility curves for the landslide-pipeline interaction problem are not available due to the lack of field data. The problem can be addressed by numerical approaches. In this study, a simplified two-dimensional numerical model is developed. It characterizes the pipes as beam-type structures and the surrounding soil as Winkler springs. It is compared here against three-dimensional continuum-based analyses, which could save extensively on computational efforts. All input parameters were sampled randomly and paired with the displacement demands to form a limited set of statistically significant, yet nominally identical, pipeline samples, and the demand models for the maximum tensile strain were evaluated. A supervised machine learning technique, called Lasso regression, was then used to establish a predictive relation between the input and the output using the limited dataset, based on which a large dataset (one million) was calculated for the fragility analysis. The use of a Winkler-based analysis and the supervised machine learning technique makes it possible to generate fragility curves for pipelines subjected to transverse PGD for the first time.