Deep Learning Application of Geotechnical Property Characterization for US Atlantic Outer Continental Shelf Soils

Abstract Deep learning (DL) is capable of learning complex representations of data by discovering hierarchical patterns and features in the data without the need for manual feature engineering. Due to recent advances in processing power and the availability of large datasets, DL has become increasingly popular in geotechnical site investigation to characterize soil properties. This paper presents examples of practical application of DL to characterize the geotechnical properties of US Atlantic Outer Continental Shelf (OCS) soils, including piezocone penetration test (PCPT)-driven soil behavioral type (SBT) classification and prediction of in situ shear wave velocity (Vs). A comprehensive offshore geotechnical site investigation campaign has been undertaken for a large wind farm development site along the US Atlantic OCS offshore New Jersey, a frontier location with few published data available on soil characterization. Significant data volumes have been acquired through offshore in situ tests, including seabed and downhole PCPTs, seismic PCPT and P-S suspension logging tests, and onshore soil element laboratory tests. In this paper, deep learning algorithms (DLAs) have been generated based on feedforward neural network algorithms and engineering judgment to predict the soil classification by the soil behavior type (SBT) chart using Robertson's (2009) Ic method and to estimate Vs based on the measured PCPT values. Accuracy matrices of the DL results (i.e., how close the DL SBT zones are to the engineering Ic computed values; how close DL Vs predictions are to the in situ seismic PCPT measurements) are presented alongside those derived from empirical geotechnical correlations. Additionally, discussions of analysis results and recommendations of future study are provided herein. This paper illustrates the great potential of utilizing a DL-enhanced process for site characterization and estimation of soil properties at offshore wind frontier areas.