Integrated data-driven and mechanics-based modeling framework for pile design

Abstract

To conquer the limitation of using empirical load-transfer curves for axially loaded pile design under various soil and site conditions, an innovative modeling framework is proposed by integrating data-driven load-transfer curves and mechanics-based pile solutions. A pile database consisting of twenty-four field load tests is compiled, and used to train and test the multi-layer feed-forward artificial neural networks (ANNs). The optimization of ANN model architecture is realized through genetic algorithm. The performance of the ANN-based load-transfer curves is evaluated by comparing with other empirical curves. Finally, the ANN-based load-transfer curves are implemented into the matrix displacement method. The effectiveness of the integrated data-driven and mechanics-based modeling is demonstrated through a case study.