A comprehensive health diagnosis method for expansive soil slope protection engineering based on supervised and unsupervised learning

There is a complex multifactorial coupling effect among the damages of various protection structures on slopes. Existing research focused on the health assessment of individual structures is often insufficient in representing the overall health status of the protection engineering system. Considering the characteristics of expansive soil slope protection engineering, this study proposes a health diagnosis method using combined weights and binary K-means clustering algorithm. The method quantifies the damage data of protection structures based on subjective and objective weights, and clusters the data by combining the binary K-means method and target vector layer to obtain the diagnosis results. Furthermore, an XGBoost-based surrogate diagnosis model is constructed to omit the repetitive modelling process in practical applications to achieve dynamic diagnosis. The proposed method is validated to an expansive soil slope in Gaochun district, Nanjing. The results show that the proposed method can accurately evaluate protection engineering with different degrees of damage; the surrogate model follows the same weight assignment process as the diagnostic method to establish reliable prediction. Based on the proposed method, damage coupling effects between individual protection structures are captured, and targeted maintenance and repair can be implemented. The proposed method can be further extended to other slope engineering.