Modeling of Seismic Energy Dissipation of Rocking Foundations Using Nonparametric Machine Learning Algorithms

Sivapalan Gajan

doi:10.3390/geotechnics1020024

Abstract

The objective of this study is to develop data-driven predictive models for seismic energy dissipation of rocking shallow foundations during earthquake loading using multiple machine learning (ML) algorithms and experimental data from a rocking foundations database. Three nonlinear, nonparametric ML algorithms are considered: k-nearest neighbors regression (KNN), support vector regression (SVR) and decision tree regression (DTR). The input features to ML algorithms include critical contact area ratio, slenderness ratio and rocking coefficient of rocking system, and peak ground acceleration and Arias intensity of earthquake motion. A randomly split pair of training and testing datasets is used for initial evaluation of the models and hyperparameter tuning. Repeated k-fold cross validation technique is used to further evaluate the performance of ML models in terms of bias and variance using mean absolute percentage error. It is found that all three ML models perform better than multivariate linear regression model, and that both KNN and SVR models consistently outperform DTR model. On average, the accuracy of KNN model is about 16% higher than that of SVR model, while the variance of SVR model is about 27% smaller than that of KNN model, making them both excellent candidates for modeling the problem considered.

Highlights

Shallow foundations, with controlled rocking during earthquake loading, have been shown to have many beneficial effects on the seismic performance of structures by effectively acting as geotechnical seismic isolation mechanisms
It should be noted that no arbitrary threshold limit is set for mean absolute percentage error (MAPE) value to classify whether the model predictions in terms of MAPE are acceptable or not for the chosen performance parameter (NED)
All three nonparametric machine learning models developed in this study (KNN, support vector regression (SVR) and decision tree regression (DTR)) perform better than the parametric multivariate linear regression (MLR) model in capturing the complex relationship between Normalized energy dissipation (NED) and input features of rocking foundations

Summary

Introduction

With controlled rocking during earthquake loading, have been shown to have many beneficial effects on the seismic performance of structures by effectively acting as geotechnical seismic isolation mechanisms. The cyclic moment-rotation relationships show seismic energy dissipation in soil due to rocking through mobilization of bearing capacity and shearing of soil (total area enclosed by moment-rotation hysteretic loops) This beneficial seismic energy dissipation in soil, in turn, reduces the acceleration (ax), lateral force, and lateral drift demands transmitted to the structure (i.e., rocking foundations effectively act as geotechnical seismic isolation systems)

Objectives

Results

Conclusion