Using artificial intelligence to predict the limit pressures of foundation soils

Résumé

In the infrastructure sector, soil investigation is essential for sizing the foundations of structures. The depth of investigation required by current regulations and depending on the scale of the project is becoming a major problem in Madagascar due to inadequate resources or limited capacity of testing equipment. As a result, engineers are forced to extrapolate unknown values beyond the pressuremeter diagram supplied by the laboratory. But a poor estimate could well lead to colossal errors, with harmful consequences for future structures. With this in mind, an innovative and formidable method such as artificial neural networks is being used to effectively predict the unknown limit pressures at which test equipment reaches its limit at depth. The foundation calculation based on Menard pressuremeter tests was our choice, and the approaches were carried out under the conditions of multilayer formal neurons with a non-looped structure. The simulation results confirm the effectiveness and robustness of the approaches, and the prediction of the pressuremeter diagram ideally reflects both logic and reality. Finally, to ensure the predictive power of the networks, a comparison of the calculation of the piles with the GEOFOND software was carried out with an error rate of 0.49%, followed by a performance analysis with a Nash-Sutcliffe coefficient of 0.9028, justifying the best fit of the model to the observed values. We have concluded that our neural models perform very well and can be used by our engineers with a high level of reliability.