<![CDATA[This study presents a reinforcement-learning framework for real-time strain-rate control in Constant Rate of Strain (CRS) consolidation testing to hasten the testing process using the SARSA algorithm. The controller adaptively adjusts deformation rate based on evolving pore-pressure ratio, with a reward strategy designed to maintain an average pore-pressure ratio near 30% to ensure partially drained conditions consistent with CRS theory. Two normally consolidated clays with contrasting compressibility were modeled numerically using a 1-D CRS consolidation model to evaluate learning and testing performance. The results show that the SARSA agent autonomously learns soil-specific strain-rate policies and maintains smooth effective stress trajectories and stable pore-pressure ratio responses. Test duration reductions of 60-75% were achieved depending on soil type. The interpreted compression index (Cc) remains consistent with the baseline CRS values, confirming that reinforcement-learning-based strain-rate control can accelerate testing without compromising data integrity. The study demonstrates the feasibility of reinforcement learning for CRS testing and highlights practical potential for soil-responsive, adaptive strain-rate control. Current limitations include simulation-based evaluation, discretized action selection, and the need for multiple runs to achieve optimal convergence.]]>
Copyrights © 2026
