Economic impact analysis for steel piles driven in intermediate geomaterials using machine learning algorithms

Abstract

To mitigate the existing challenges with piles driven in intermediate geomaterial (IGM), this study presents an economic impact assessment for steel piles in IGMs based on the newly developed and existing static analysis (SA) methods using 149 test pile data from seven US states. The assessment determines the differences in the number of piles and the equivalent steel pile weight. The proposed SA methods yield, on average, a smaller difference in steel weight based on states, pile types, and bearing IGM layers. Three machine learning (ML) algorithms: random forest, support vector machine (SVM) and neural network are applied to predict the difference in steel weight. Three percentage-based variables are employed in the ML algorithms as inputs: total pile penetration, total shaft resistance and end bearing in IGM. Based upon 31 testing data, SVM with the lowest RMSE, MAD and highest pseudo-R² is identified as the best algorithm. The predicted difference in steel weight from SVM is optimized to zero using a novel application of the genetic algorithm, and various contour maps are generated. These contour maps can be used to predict the difference in steel weight graphically based on the three percentage-based variables for future driven steel piles in IGMs.