A hybrid mechanics-guided machine learning-based predictive framework for the performance of rocking foundations during earthquake loading

Recent research findings reveal that rocking foundations during earthquake loading perform as efficient geotechnical seismic isolation mechanisms and have the potential to eliminate or reduce the damage to the building and bridge structures they support. The objective of this project is to develop a hybrid predictive framework for the performance of rocking foundations by combining the mechanics that governs the physics of the problem with the knowledge discovered from the use of big data and machine learning (ML) techniques. This NSF-funded Engineering Research Initiation (ERI) project is the first attempt to combine mechanics with data science to model the seismic performance of rocking foundations (CMMI-2138631).

As part of this project, the following models were developed: (i) purely data-driven ML models for performance prediction of rocking foundations using centrifuge and shaking table experimental results (dataset available in Design-Safe), (ii) numerical models of rocking foundations using mechanics-based models available in OpenSees finite element framework, and (iii) mechanics-guided machine learning (MGML) models for rocking foundations by combining the OpenSees numerical model simulation results with ML algorithms. Several ML algorithms have been utilized in this project to develop multiple MGML models, including gradient boosting regression, extreme gradient boosting regression, random forest regression, support vector regression, and k-nearest neighbors regression. The performance parameters of rocking foundations considered include maximum moment and peak rotation of rocking foundation, seismic energy dissipation in soil during rocking, permanent settlement of foundation, and maximum acceleration transmitted to structure during earthquake loading.

All the MGML models developed in this project, along with their source codes (in Python) and documentations, are available in this database. The experimental dataset used, input data files, and the output results (model predictions) of all MGML models are also available in this database (in Excel sheets). Potential users of this database/dataset include researchers, students, and engineers in the fields of geotechnical engineering, earthquake engineering, and soil-structure interaction. The models and results/data available in this database can be used to investigate the behavior of rocking founders further and/or for the development of future numerical/ML models for rocking foundations.