| Popis: |
In 2023, pivotal advancements in artificial intelligence (AI) have significantly experienced. With that in mind, traditional methodologies, notably the p-y approach, have struggled to accurately model the complex, nonlinear soil-structure interactions of laterally loaded large-diameter drilled shafts. This study undertakes a rigorous evaluation of machine learning (ML) and deep learning (DL) techniques, offering a comprehensive review of their application in addressing this geotechnical challenge. A thorough review and comparative analysis have been carried out to investigate various AI models such as artificial neural networks (ANNs), relevance vector machines (RVMs), and least squares support vector machines (LSSVMs). It was found that despite ML approaches outperforming classic methods in predicting the lateral behavior of piles, their ‘black box' nature and reliance only on a data-driven approach made their results showcase statistical robustness rather than clear geotechnical insights, a fact underscored by the mathematical equations derived from these studies. Furthermore, the research identified a gap in the availability of drilled shaft datasets, limiting the extendibility of current findings to large-diameter piles. An extensive dataset, compiled from a series of lateral loading tests on free-head drilled shaft with varying properties and geometries, was introduced to bridge this gap. The paper concluded with a direction for future research, proposes the integration of physics-informed neural networks (PINNs), combining data-driven models with fundamental geotechnical principles to improve both the interpretability and predictive accuracy of AI applications in geotechnical engineering, marking a novel contribution to the field. |
