Load-Deflection Prediction of Helical Piles Using Hyperbolic & Machine Learning Methods-PPT Presentation

PPT Presentation

Pdf Summary

This presentation addresses the lack of reliable methods for predicting helical pile deflection under load, despite the importance of meeting structural engineers’ deflection limits. Building on earlier research by Cherry and Perko, the study analyzes 179 compression and 174 tension load tests.<br /><br />Two prediction approaches are compared. The hyperbolic method linearizes nonlinear load–deflection data, producing parameters that describe the curve. The machine-learning methods are a deep neural network (DNN), which learns relationships by optimizing model weights, and a random forest, which combines predictions from multiple decision trees.<br /><br />The models’ mean squared errors (MSEs) are compared across load tests. The results generally show substantially lower errors for the random forest than for the DNN and the Perko hyperbolic model, although performance varies by test, and the hyperbolic model is better in some cases. The findings suggest machine learning—particularly random forests—may be useful for predicting helical deflection, but the results are preliminary and depend on the available data and model inputs.<br /><br />Planned next steps include publishing an online prediction tool, collecting more test data, and developing a larger model that accounts for factors beyond load, such as test type, soil conditions, and pile plate number and size.

Keywords

helical pile deflection

load testing

machine learning

random forest

deep neural network

hyperbolic method

mean squared error

structural engineering

prediction models

soil conditions