Predictions of failure mode and arresting efficiency of integral buckle arrestors using FEM and machine learning methods.

• FE models were established to examine the behavior of buckling propagation and crossover. • The influences of geometry and material property of pipes and arrestors on the failure mode and crossover pressure were numerically explored. • Four machine learning models were developed to predict the failure mode, crossover pressure, and arresting efficiency, respectively. Integral buckle arrestors are regarded as the most effective arresting devices, which can provide an obstruction to a propagating buckle thereby protecting downstream pipelines. In the present study, numerical frameworks were established to reproduce the phenomenon of buckling propagating and crossing under hydrostatic pressure, and a strong consistency between measurements and predictions was achieved. The stress levels and deformation configurations of the assembly were carefully examined for two different failure modes. Then, broad parametric analyses on the crossover pressure were performed covering key material properties and geometries. After that, machine learning techniques were developed and used for predictions of failure modes, crossover pressure, and arresting efficiency, respectively. Four algorithms, including Random Forest, Multi-layer Perceptron, K-Nearest Neighbors, and Support Vector Machine, were implemented using a dataset with seven variables. The prediction performance was evaluated by standard statistical metrics, and it was found that Multi-layer Perceptron model exhibited the better prediction accuracy for both classification and regression problems. Additionally, the existing experimental results were also used to verify the reliability of the machine learning model. The results demonstrate that the machine learning techniques can offer relatively accurate predictions for both flattening and flipping failure modes. [ABSTRACT FROM AUTHOR]