Load-settlement behavior estimation of piles using driving log

Recently, there has been a trend towards increasing the amount of data collected during construction and engineering surveys. A large amount of data allows the production of data-dependent models for assessing the behavior of the soil and its interaction with structures. For this, it is proposed to use soft computational methods, in particular neural networks. Neural networks models allow the approximation of complex dependencies and give more accurate results than regression methods. Such models allow predicting any property of interest as long as the input data is suitable. This study proposes to use pile-driving log data as an input to estimate precast concrete piles behavior under static loading. Data from static pile tests were used as training output data. A database of static load tests and pile driving logs was collected. Several neural network models were tested with different sets of input data. A trained neural network made it possible to estimate pile settlement under varying static load using driving data. Neural network models can be infinitely improved by adding more varied and complete data. Despite the medium accuracy, the proposed model does not use data from other types of geotechnical tests and can be used in case of lack of data or as an additional check.