Pre-Bored PC Pile Capacity Prediction in Drained Soils Using Artificial Neural Network

This study evaluated the performance of artificial neural network (ANN) in determining pre-bored precast concrete (PC) pile capacities, utilizing a database of 39 drained pile load tests under axial compression. The ANN model developed in this study is limited in applicability to pre-bored PC piles installed under drained conditions in Southern Taiwan. Measured capacities, obtained through various interpretation methods, were compared with the predicted capacities. Predicted capacities were determined in two approaches: (1) via conventional methods in determining pile capacity, namely β method for side resistance and the general bearing capacity equation for tip resistance, and (2), using ANN to predict the load-displacement curves. Based on the analysis, the L1 method yielded the lowest interpreted capacities and displacements, while the Fuller and Hoy method and Chin method resulted in significantly higher capacities and displacements. The measured versus ANN-predicted loads resulted in r2 values ranging from 0.93 to 0.97, indicating a good fit. The ULS model factors, which are the ratio of the measured to predicted values, exhibited low dispersion, with most values being unconservative. The SLS model factors a and b for serviceability limit states are also provided for reliability-based design using the hyperbolic curve fitting method. The ANN-predicted capacities closely matched the L2 measured capacities and predicted capacities computed via conventional methods. Additionally, the distribution range of the data points of the hyperbolic model factors a and b does not differ significantly between the conventional method and ANN, when the load is normalized by the predicted capacity. This demonstrates the potential of using ANN in combination with a hyperbolic approach in predicting the load-displacement curve. This study emphasizes the importance of continuous improvement in predictive modeling techniques for pile foundation designs.