GEOSYN: SYNTHETIC GEOTECHNICAL CROSS-SECTIONS FOR MACHINE LEARNING APPLICATIONS

The application of machine learning in geotechnical engineering is often hindered by the scarcity of high-quality, labelled datasets. To address this, we introduce GeoSyn, an open source Python-based tool that generates synthetic geotechnical 2D cross-sections, allowing users to define layer size and number, geotechnical properties and anisotropy with random fields, and boundary conditions. The generated data provides an effective solution for the development and training of ML applications in geotechnics. We demonstrate the tool’s utility through two applications. First, we show how a conditional Generative Adversarial Network, trained with synthetic data from GeoSyn, can interpret geotechnical schematisations from Cone Penetration Tests. Second, we explore how Deep Reinforcement Learning can be used to optimise the placement of subsequent in-situ surveys based on prior results. These examples illustrate how GeoSyn enables the development of ML models by leveraging large, flexible datasets to support decision-making in geotechnical engineering.