A state-of-the-art review of deep learning algorithms for geotechnical engineering – Part 1: Algorithms

Abstract

Geotechnical engineering is widely recognized as one of the most intricate fields of engineering due to its involvement with earth materials such as soil, rock and intermediate geo-materials, e.g. coal, which are associated with numerous sources of uncertainty. Consequently, physical studies in this field often rely on simplifications and empirical assumptions. However, the increasing availability of data in geotechnical projects and the potential use of deep learning (DL) algorithms to address research challenges in geotechnics have rendered DL a pivotal research subject in geotechnical engineering. Compared to traditional machine learning (ML) approaches, DL algorithms have greater capacity for automated feature extraction and for learning from large and complex data sets. Therefore, DL algorithms have gained broad application across diverse geotechnical domains. This state-of-the-art review is presented in two parts. Part 1 discusses the most commonly used DL algorithms based on the available literature. Part 2 then highlights the specific application of these algorithms in geotechnical projects. In the first part of the paper, the frequency of use for each algorithm is listed based on the number of articles in which they are used. Subsequently, the most important algorithms are examined in detail. This paper also aims to address the computational tools and frameworks used for implementing DL models, as well as the efficiency and scalability of DL algorithms. In addition, a comprehensive summary is provided, which includes published literature, relevant reference materials, adopted DL algorithms and relevant geotechnical topics. The paper concludes by evaluating the challenges and prospects that lie ahead for the future development of DL in geotechnical engineering. This review adopts a systematic literature analysis of DL algorithms relevant to geotechnical engineering. It categorizes DL methods based on usage frequency and evaluates their structures, implementation frameworks (e.g. TensorFlow, Keras, PyTorch) and learning strategies, such as transfer learning. The paper emphasizes the role of data quality and preprocessing and compiles open-access data sets for researchers. It serves as Part 1 of a two-part series, focusing on DL techniques rather than specific applications. Convolutional neural networks, long short-term memory networks and generative adversarial networks are among the most frequently applied DL models in geotechnical research. These models have demonstrated superior feature extraction, accuracy and adaptability to geotechnical data sets. DL frameworks like TensorFlow and PyTorch are widely used due to their scalability and open-source nature. Transfer learning emerges as a powerful technique in data-scarce geotechnical domains. This is one of the first structured reviews exclusively dedicated to DL algorithms in geotechnical engineering. Unlike previous artificial intelligence surveys, it focuses on algorithmic foundations, tools and data requirements. The inclusion of transfer learning and publicly available data sets adds significant practical value. It serves both as an educational entry point and a foundation for further research into real-world DL applications in geotechnics.