Exploring sustainable solutions for soil stabilization through explainable Gaussian process-assisted multi-objective optimization

The adoption of sustainable solutions in soil stabilization has piqued the interest of the scientific community due to the potential reduction in carbon footprint. In this regard, the research community has started looking for the alternate sustainable solutions to limit the quantity of conventionally used ecologically unfriendly soil stabilizers like lime and cement by utilizing the agricultural and industrial by-products. The production of conventional soil stabilizers (lime and cement) is extremely energy-intensive and contributes tons of greenhouse gases to the atmosphere. In general, evaluating suitability of these additives requires in-lab investigation of soil samples with varying additive concentrations and curing periods, making this approach both resource and time-intensive. Hence, this article proposes a computational framework for accelerated characterization of soil-stabilization by using the coupled experimental-Gaussian process (GP) based machine learning (ML) model. The dataset utilized for constructing the GP models consists of input features such as the stabilizer content (lime and rice husk ash (RHA)), coir fiber content, and curing period (measured in days). The target responses are the strength measures of the stabilized soil, such as unconfined compressive strength (UCS), split tensile strength (STS), and California bearing ratio (CBR). The proposed computational framework is deployed to perform multi-objective genetic algorithm (MOGA)-based optimization to achieve maximum engineering performance from stabilized soil. The presented study demonstrated that with the optimal dosage of rice husk ash (agricultural by-product), and cashew nut shell liquid (CNSL) treated coir fiber, the requirement for lime dosage may be significantly reduced while maintaining the engineering performance of the soil. The presented computational framework can be extended to any construction practices for ensuring the strategic selection of the control variables for optimizing the desired performance.