Monopiles with large diameter (larger than 6 m) and low aspect ratio (less than 6) are increasingly used in offshore wind farms. These foundations demonstrate a rigid response under lateral loading. The validity of the existing design methods, that are based on small diameter flexible piles, has been questioned by both the industry and researchers. In addition, the monopiles are subjected to both lateral and vertical loads. The influence of vertical load on the lateral design of short rigid monopiles in clay soil is not clear. This study aims to perform a comprehensive study on the influence of vertical load on the lateral response of monopile foundations in clay soil. All analysis in this study was performed using 3D finite element modeling in PLAXIS 3D software. The NGI-ADP constitutive model was adopted to simulate the nonlinear mechanical behaviour of clay. Considered in the analysis is a short rigid pile with a diameter of 10 m (L/D = 3) and a long flexible pile with a diameter of 2 m (L/D = 15). The analyzed clay soil profiles consist of a normally consolidated clay soil and an overconsolidated clay soil with a constant undrained shear strength profile equal to 30 kPa. For each pile in each type of clay soil, a pure lateral loading scenario is performed first to assess the validity of current design methods. Subsequently, a combined loading scenario is performed to assess the influence of vertical loading on the lateral behaviour of rigid monopile in clay soil. Results of the pure lateral loading scenario suggest that current design methods heavily underestimate the lateral capacity of rigid monopile foundations in both clay soil profiles analyzed. According to the findings of this study, it can be concluded that current design methods are not fit to provide an accurate assessment regarding the lateral load response of rigid monopile in clay soil. In order to correctly assess the lateral load response of rigid monopile in clay soil, a method consisting of a 3D finite element model akin to the model used in the research or a PISA design model is advised. A potential third design method, the 1D rotational spring model, is also proposed. Results of the combined loading scenario suggest that the presence of vertical loading causes a decrease in lateral and moment capacity of the rigid pile in both clay soil profiles analyzed. However, the influence is negligible when the vertical load magnitude is smaller than 50% of its bearing capacity. To quantify the influence of vertical load on a monopile foundation, a series of load analysis were performed on a real offshore wind turbine with a 5MW power capacity. It was found that the vertical load on a typical monopile foundation in clay is around 27% of its bearing capacity. According to the findings of this study, it can be concluded that the influence of vertical load on the lateral response of rigid monopiles in clay soil is limited and can be ignored in foundation design.

For preparing standardized sand specimens for physical modelling in geotechnics, especially for the geo-centrifuge tests, a “line-style” sand pluviator has been recently developed by TU Delft. Controlling the falling height of sand hopper, the width of hopper’s bottom gap and the relative moving speed between the hopper and the sample box, specimens with bulk relative density ranging from 50% to 100% can be prepared by this automated machine using the coarse Merwede River sand. Besides, the periodic variation of local relative density along depth was observed using the macro-CT scanner and the features of the fabrics were investigated using the micro-CT scanner. It was also proved by a set of shallow foundation modelling tests that the sand specimens having the same bulk relative density but different heterogeneity and fabric features behaved significantly differently and further research works are recommended to explore the influences of these differences. Additionally, a partially substantiated hypothesis was proposed to conclude the general rules of the sand pluviation process, and the reliability of this hypothesis has been proved by a series of tests on the Geba sand.

In this thesis the performance of five CPT-based p-y models by Novello (1999), Dyson & Randolph (2001), Li, et al. (2014) and Suryasentana & Lehane (2014;2016) is assessed. This is done in part by comparing the measured load-deflection response from a set of lab- and field tests that is representative for offshore monopile structures to the predicted response generated by a MATLAB pile-response model that incorporates the aforementioned p-y models. It is found that with the exception of Suryasentana & Lehane (2014), alle models are more accurate at large- than at small displacements. Furthermore, the models are generally conservative at a groundline displacement of D/100 and unconservative at a groundline displacement of D/10. The latter is likely related to inability of the models to adequately capture the ultimate soil resistance. The strongest found correlation with prediction accuracy is a negative one with pile rigidity at small displacements. The models from Dyson & Randolph (2001) and Li, et al. (2014) generate predictions with the highest mean accuracy, though all models perform relatively poorly when compared to a benchmark from Burd, et al. (2019). The aforementioned CPT-based p-y models are tested further through a number of new instrumented pile load tests in the TU Delft geotechnical centrifuge and the experimental methods for these tests are evaluated. Image analysis of CT scanned sand samples shows that though there is substantial non-uniformity present in samples prepared through hand raining, this method yields acceptable results and is preferable to preparation using a sand raining machine. Secondly, though soil disturbance does occur as a result of pile installation at 1g, this is not likely to greatly impact the results. A comparison of the load-deflection predictions generated with the p-y models (including API) for these load tests supports the notion that the initial response from the API (2014) model is far too stiff and that it is therefore unconservative in the prediction of the limit states. Furthermore, a modified 5th order polynomial fitting method is found to adequately fit discrete bending moment data and in turn p-y curves are produced. These show that the pu from the models Suryasentana & Lehane (2014;2016) is only reached at shallow depths and does not match the measured ultimate resistance. Finally, at all except the shallowest depths the CPT-based p-y models strongly overestimate p though the resulting prediction of the load-deflection response is more reasonable. As mentioned previously, the CPT-based p-y models do not perform optimally for the studied experimental cases, especially at small groundline displacements. This can likely be attributed to the fact that they were generally developed from test data on long and flexible piles. Additionally, rigid piles under lateral load develop moments and shear forces at the base that are not taken into account by methods that rely solely on lateral soil resistances (Peralta 2010). To solve these problems the PISA model with additional soil reaction terms was developed and it has been proven to provide better displacement predictions than traditional methods relying on lateral soil resistance (Byrne, et al., 2015a). However, this model is more complex and may not be the most convenient method to use. A less complex alternative has been proposed which uses a simplified rotational pile-soil interaction model to capture the load-displacement behaviour of perfectly rigid piles of varying diameter, stress level and loading eccentricity. Numerical simulations by Wang, et al. (2022) have shown this to work and in this thesis these results are validated using experimental data. The unification of the pile responses is found to work, but is not as successful for the experimental data. This may be due to several factors, the most likely of which are variations is sand density and imperfect rigidity of the model pile. However, the results support use of methods based on the aforementioned simplified rotational pile-soil interaction model, such as the rotational spring model proposed by Wang, et al. (2022).

Suction caissons have been used extensively for anchoring and supporting the offshore installations like oil platforms and wind turbines. These foundations are normally subjected to complex combinations of the vertical, horizontal and moment loads (i.e. V, H, M) from the self-weight, wind, wave and currents. In the past decades, extensive studies have been conducted to investigate the combined V-H-M loading behaviour of suction caissons in clay. However, most existing studies are focused on the ultimate bearing capacity, while the deflection response is more critical in foundation design for recent infrastructures like offshore wind turbines. Due to the complex load conditions, predicting the three-dimensional (3D) deflection response of the foundation is still challenging. Machine learning (ML) appears on the research horizon due to its excellent capacity of solving nonlinear problems with desired speed and accuracy. However, conventional machine learning approaches were limited in their capacity to analyze raw natural data without artificial interventions. Meanwhile, the deep learning technique (DL), as a branch of machine learning, allows a machine to be fed with raw data, automatically extract the features, and discover intricate structures in high-dimensional data. The deep learning technique has been used in many fields like language translation, auto-pilot and image recognition. And Deep neural networks, including deep learning algorithms and architectures, are gradually being developed. In light of these backgrounds, this study proposed to develop a deep learning based surrogate model to predict the 3D deflection response of suction caissons under combined V-H-M loading. The advanced three-dimensional nonlinear finite element (FE) simulations under complex V-H-M loading paths were performed on suction caissons of different geometric configurations and in clay soils with different stiffness and strength properties. The 3D FE simulation data was then used to train the deep learning based design model. Three popular neural network structures, i.e., Feed forward Neural Network (FNN), Convolution Neural Network (CNN), Recurrent Neural Network (RNN) have been employed to develop the hybrid surrogate design model. In this study, two different training strategies were proposed for this geotechnical problem. In the first category, the 3D load-deflection behaviour of suction caisson is idealized as a point-to-point mapping problem, i.e. mapping between the deflections (i.e. displacement and rotation) with loads (i..e force and moment). This task was achieved by Fully-Connected Neural Network model (FC-NN) based on FNN, One Dimension Convolution Neural Network model (1D-CNN) based on CNN and Long Short Term Memory model (LSTM) based on RNN. In the second training strategy, the load-deflection response was idealized as a time series process, a line-to-line mapping problem, mapping between the past loading paths (i.e. 10 groups of forces and moments) with future loading paths (i.e. 90 groups of forces and moments). Besides the three neural network models mentioned before, another two complex and advanced models, LSTM Model combined with convolution neural network (1D-CNN+LSTM) and Temporal Convolutional Network model (TCN), are also applied for temporal prediction. The performance and training efficiency of these models were also systematically evaluated by interpolation and extrapolation experiments. Basically, all the models can well capture the 3D deflection response of the foundation with significantly high accuracy (i.e., root mean squared error is smaller than 0.05 and coefficient of determination is near 1.000) than the traditional design approach (such as macro-elements model), and with greater efficiency than the 3D FE simulations. Among all the models, the TCN model has the highest prediction accuracy and robustness. However, the FC-NN model has the simplest model structure and highest computational efficient in learning the non-linear relationship between deflection response and V-H-M load. Besides capturing the relationship between input and output, the deep learning model can also assist to identify the intrinsic failure mechanism. By observing the fluctuation of generalisation ability, the evolution of the failure mechanism of suction caisson with embedment depth was revealed.