Skirted foundations are widely used in offshore and subsea engineering. Previous studies have shown that soil undrained shear strength variability has a notable impact on probabilistic analyses of skirted foundation bearing capacity. This study proposes an efficient machine-learning method to predict the uniaxial bearing capacity factors of skirted foundations under pure horizontal and moment loads, without relying on traditional time-consuming random finite element methods. A two-dimensional convolutional neural network is adopted to capture the potential correlation between soil random fields and bearing capacity factors. The proposed CNN-based model exhibits satisfactory prediction performance with regard to coefficients of variation and scale of fluctuations in two directions. Specifically, coefficient of determination (R2) values exceed 0.97, while root mean square error (RMSE) values remain below 0.13 for the surrogate model. In addition, more than 96% of the predictions are associated with a relative error of 5% or less, providing evidence of the proposed 2D-CNN model’s satisfactory prediction performance.

This study investigated the seismic response of rectangular tunnels with various embedment depths considering the spatial variability of soil shear modulus. The spectral representation method was adopted to simulate the anisotropic random field of soil. The excess pore water pressure, the liquefied zone, the ground displacement and the uplift displacement of the tunnel were obtained through the random finite difference method to analyze the seismic response. It was observed that the soil excess pore water pressure ratio under the tunnel gradually decreased and the liquefaction degree reduced with depth increase. The peak value of the liquefied zone range increased with the increase in embedment depth. The mean response of stochastic analysis was smaller than the deterministic calculation results when the tunnel embedment depth was less than 10 m. The maximum tunnel floating displacement obtained from random analyses had the probability of 67.3%, exceeding the value calculated by deterministic analyses when H = 12 m.

Seismic signal denoising stands as a vital process that enables precise seismic data analysis because noise interference blocks the detection of weak but valuable seismic signals. The current traditional denoising methods together with deep learning-based data-driven approaches encounter difficulties when they need to remove noise from seismic signals while keeping their fundamental structural elements, especially under conditions of low signal-to-noise ratios. In this study, we propose a novel denoising framework that integrates a physics-guided neural network with adaptive time–frequency decomposition, referred to as TF-PhysNet. The system breaks down broadband seismic data into separate frequency bands. Scientists can use these to study specific noise patterns that appear at various frequency points. The system uses a shared convolutional neural network-long short-term memory architecture to remove noise from each sub-band, which helps it learn both short-term waveform patterns and extended temporal relationships. The system uses physics-guided restrictions to eliminate false signal variations, which appear during the signal recovery process. The experimental findings from synthetic and real seismic data sets show that TF-PhysNet delivers better results than standard denoising techniques and deep learning-based methods for signal-to-noise ratio improvement and correlation coefficient enhancement.