From Non-destructive Testing to Ground Property Inference: Integration of AI and Unsaturated Soil Dynamics

The present paper introduces an innovative Artificial Intelligence (AI) framework specifically designed for the statistical inference of the characteristics of unsaturated soils. This model utilizes non-destructive indentation testing methodologies, such as Light Weight Deflectometer (LWD), providing a more advanced and accurate evaluation of soil properties. Existing methodologies for inferring ground properties present a variety of limitations that challenge their accuracy and applicability. These include an inconsistent representation of the interrelated impact of water content and dry density in the indentation results; a significant loss of precision due to their empirical nature; issues with generalization; and an inability to effectively communicate the inherent uncertainty in the ground assessment and conditions. To address these challenges, our AI-based model integrates a more holistic and comprehensive approach, improving upon these limitations and refining the process of ground inference. The novel framework enhances the overall accuracy of soil property determination and introduces a more dependable method of conveying the uncertainties within the assessments. The model's potential for effective use extends beyond theory, as it is capable of practical implementation in field trials. Our paper will showcase examples of this technology in action, demonstrating how it successfully infers soil properties in actual field scenarios.