Multi-objective optimisation design for GFRP tendon reinforced cemented soil: Fid-Bau Portal

Multi-objective optimisation design for GFRP tendon reinforced cemented soil

Zhang, Genbao / Chen, Changfu / Li, Kefei / Xiao, Fan / Sun, Junbo / Wang, Yufei / Wang, Xiangyu

Highlights • The cement content and curing time positively affect the mechanical properties of GRFP tendon reinforced cemented soil comprising unconfined compressive strength (UCS) and ultimate pullout strengths ( $T_{p}$ ), whereas the water content generally has a negative effect. • The 10-fold cross-validation and beetle antennae search (BAS) effectively tune the hyperparameters of the support vector regression (SVR) models for UCS and $T_{p}$ . • The outstanding performance of the BAS-SVR model is shown in high correlation coefficients (UCS: 0.988, $T_{p}$ : 0.972) and low RMSE values (UCS: 313 kPa, $T_{p}$ : 167 kPa). • The multi-objective optimisation design solusions for UCS-cost and $T_{p}$ -cost are both successfully obtained through generating Pareto fronts based on MOBAS-SVR. • Sensitivity analysis determines the most significant varible among all the input features is cement content.

Abstract Rebar reinforced cemented soil is employed widely to solve the weak foundation problem led by sludge particularly. Nowadays, the glass fiber-reinforced polymer (GFRP) becomes a new tendon material instead of steel to avoid the performance degradation resulting from steel corrosion. The interface bond strength of GFRP tendon-reinforced cemented soils (GTRCS) displays its excellent mechanical capacity. Nevertheless, its application is obstructed by the deficient studies between the bond strength and influence factors. Therefore, this study investigates the effects of varying water contents ( $C_{w}$ : 50%-90%), cement proportions ( $C_{c}$ : 6%-30%), and curing periods ( $T_{c}$ : 28 days, 90 days) on both pullout strength ( $T_{p}$ ) and unconfined compression strength (UCS) of GTRCS. The results showed that the pullout strength and compressive strength were positively related to $T_{c}$ and $C_{c}$ and negatively related to $C_{w}$ . Besides, these experimental results were also utilised to develop support vector regression (SVR) models. The beetle antennae search (BAS) algorithm was used to adjust the SVR’s hyperparameters. The high correlation coefficients (0.988 for UCS and 0.972 for $T_{p}$ ) proved the reliability of the established BAS-SVR models. In addition, the multi-objective beetle antennae search algorithm (MOBAS-SVR) was developed for bi-objective optimisation designs (UCS-cost and $T_{p}$ -cost). Finally, sensitivity analysis was conducted to range the significance of variables for $T_{p}$ and UCS.