Machine Learning in geotechnics: case study in the tunnelling field

Machine learning (ML) is a branch of artificial intelligence (AI) that enables computer systems to improve automatically through experience. Geotechnical engineering, a discipline at the interface between geology and civil engineering, is facing increasing challenges in terms of complexity and precision in soil analysis and structure design. In this context, the emergence of ML as an analysis and prediction tool opens up promising prospects for meeting the sector's contemporary challenges.

Traditional geotechnical calculation methods, although tried and tested, have certain limitations when faced with the inherent complexity of soils and soil-structure interactions. The natural variability of geotechnical parameters, the non-linearity of mechanical behaviors and the multiplicity of environmental factors sometimes make it difficult to apply conventional analytical approaches.

AI, and ML in particular, brings a new dimension to geotechnical analysis, making it possible to exploit the vast quantities of data accumulated by the profession over decades, or more modestly on a project scale, to recalibrate a model as work progresses. These techniques make it possible to detect complex trends and patterns in the data, to automate certain analysis tasks, and to improve the accuracy of geotechnical predictions, the accepted term in this field, but we could also use the term forecast.

The following two examples illustrate simple applications of AI to geotechnical problems:

• predicting the settlement of an embankment on compressible soil traditionally requires complex calculations incorporating numerous parameters (compression index, effective stresses, etc.); the ML could enrich this approach by exploiting feedback from similar projects to refine predictions;

• the use of neural networks or decision trees for soil classification can speed up or even automate the interpretation of in situ tests, reducing analysis time while maintaining a high level of accuracy.

In this article, we will explore the application of ML to geotechnics along three main axes:

1. the fundamental concepts of ML and their state of the art in geotechnical engineering, illustrated by concrete application cases that demonstrate the relevance of these approaches to our field;

2. the scoping and preparation of ML geotechnical projects, a crucial stage in the success of the approach, from data structuring to the selection of relevant input data;

3. practical implementation, from model training to optimization, evaluation and deployment in real-life conditions.