A Machine Learning-Based Approach for Predicting Installation Torque of Helical Piles from SPT Data

Abstract

Helical piles are advantageous alternatives in constructions subjected to high tractions in their foundations, like transmission towers. Installation torque is a key parameter to define installation equipment and the final depth of the helical pile. This work applies machine learning (ML) techniques to predict helical pile installation torque based on information from 707 installation reports, including Standard Penetration Test (SPT) data. It uses this information to build three datasets to train and test eight machine-learning techniques. Decision tree (DT) was the worst technique for comparing performances, and cubist (CUB) was the best. Pile length was the most important variable, while soil type had little relevance for predictions. Predictions become more accurate for torque values greater than 8 kNm. Results show that CUB predictions are within 0.71,1.59 times the real value with a 95% confidence. Thus, CUB successfully predicted the pile length using SPT data in a case study. One can conclude that the proposed methodology has the potential to aid in the helical pile design and the equipment specification for installation.