Estimation of friction capacity of driven piles in clay using artifical Neural Network

Abstract

The load capacity of driven piles is a crucial mechanical property, and correctly determine the corresponding value is important in geotechnical engineering. Concerning piles driven in clay, the load capacity is mainly associated with the side resistance of the pile. The soil load capacity of conventional piles is determined by different methods and then reassessed by the static load test. Nonetheless, this method is time-consuming and costly. Therefore, the development of an alternative approach using machine learning techniques to solve this problem has been investigated recently. In this work, the backpropagation network model (ANN) with a 4-layer structure [4-8-6-1] was introduced to predict the frictional resistance of pile driven in clay. The dataset for the development of the ANN model consisted of 65 instances, extracted from the available literature. The performance of the proposed ANN algorithm was assessed by two statistical measurements, such as the Pearson correlation coefficient (denoted as R), and Root Mean Square Error (RMSE). In addition to the original contribution, the present work conducted a step further toward a better knowledge of the role of inputs used in the prediction phase. Using partial independence plots (PDP), the results of this study showed that the effective vertical stress and the undrained shear strength were the prediction variables that had a significant influence on the friction capacity of driven piles.