Résumé:
This doctoral thesis addresses a critical issue in geotechnical engineering: the impact of
environmental conditions, particularly freeze–thaw cycles, on the effectiveness and long-term
durability of stabilized clayey soils treated with mineral additives such as lime and fly ash.
Numerous published experimental studies have demonstrated an initial improvement in
unconfined compressive strength (UCS) following stabilization, but also a progressive
deterioration of mechanical performance under repeated freeze–thaw exposure. However,
experimental assessment remains challenging, due to the difficulty in realistically reproducing
climatic conditions, precisely controlling curing parameters, and managing result variability.
These limitations justify the adoption of predictive modeling approaches.
In this context, two modeling approaches: a statistical regression and an ANN, were
developed using literature-based experimental databases for soils stabilized with lime and fly
ash. These models effectively simulate the evolution of UCS in stabilized soils subjected to
environmental cycles. Parametric studies were also conducted to capture the influence of
several key factors, such as the number of freeze–thaw cycles, water content, additive content,
and curing duration. Results show that ANN models offer superior predictive performance
and better capture of nonlinear relationships between input variables and soil strength
response.
By integrating diverse experimental data and leveraging the power of artificial intelligence,
this work contributes to a better understanding of the long-term behavior of stabilized soils
under harsh climatic conditions and provides reliable tools to support resilient geotechnical
design in environmentally sensitive regions.
Keywords: stabilized clayey soils; freeze–thaw cycles; mineral additives; statistical model;
artificial neural network (ANN); unconfined compressive strength (UCS)
