Reliability analysis of mechanical structure with surrogate model
Résumé
Structural Reliability analysis (SRA) is a big challenge in civil and mechanical engineering. One of the reasons is that uncertainty quantification of complex physic systems is an essential component to be evaluated. Over the past few decades, a wide variety of SRA methods have been investigated. However, the classical techniques require information about Limit State Function (LSF), which can be computationally expensive. As a solution, machine learning models such as ensemble learning models are recently attracting more and more attention. In this research, the reliability analysis is performed in a surrogate-based ensemble learning model that models structural parameter uncertainties and maps them into failure probabilities. The basic idea is to find a surrogate model equivalent to the performance function. This research uses the Machine Learning-based surrogate model to evaluate TMD structural uncertainties propagation. Seen as a regression procedure, the prediction task is realized by machine learning models such as Random Forest, Gradient Boosting, XGBoost, and Neural network. Different structures with TMD are investigated. The longitudinal vibration and the transversal shear effect in a beam-like structure are also considered. Besides, a beam-like model is created and dynamic response simulations are carried out under seismic excitations. The analytical analysis and numerical examples are carried out to evaluate the benchmarking performance of different ML methods. Simulation results verify that the proposed surrogate mechanics are efficient.