Abstract

Recent advances in data analytics, numerical modeling, and structural health monitoring (SHM) boost the application of machine learning methods in the field of structural engineering. Among them, the multi-layer neural network (MLNN) is one of the most popular ones. However, mathematical details of MLNN have yet to be well understood for structural problems. This study aims to identify optimal MLNN parameters for regression modeling of structural response estimates. SHM data-validated finite element models considering stochastic uncertainties in the natural events and structural properties are used to prepare a large dataset for regression modeling. The efficacy and accuracy of regression modeling are optimized by extensive sensitivity analyses for key MLNN parameters (e.g., numbers of hidden layers and neurons, activation functions and learning rates) via a k-fold cross-validation process. The optimized regression modeling is incorporated into a conceptual smart framework for lifetime structural performance assessment adapting to evolving natural events. The presented optimization process and smart framework is applicable to marine and offshore structures by characterizing the offshore hazards and structural responses.

This content is only available via PDF.
You do not currently have access to this content.