Abstract
The construction of the open channel projects in the northern Xinjiang region of China often involves traveling through vast areas of loess. The apparent collapsibility of loess is a major concern for engineers as it can lead to uneven deformation and failure of channel slopes. Collapsibility tests and scanning electron microscopy (SEM) analysis were conducted on remolded loess to comprehensively investigate the settlement and deformation mechanisms of collapsible loess from both macro- and micro-perspectives. Furthermore, a prediction model was developed and its applicability was verified. The test results indicated that with the increase of the vertical load, the collapsibility coefficient exhibited a trend of rapid increase followed by slow increase, and eventually stabilized. This trend satisfied a hyperbolic function relationship, which was negatively correlated with the changes of the water content and dry density. SEM analysis on the loess specimens confirmed that collapsible deformation involved a gradual transition from a shelf structure to a mosaic-colloid structure. Factors such as pore size and particle morphology were found to have significant influences on the collapsibility. For prediction purposes, statistical theory and machine learning algorithms were utilized to select variables such as dry density, moisture content, initial porosity ratio, and pressure test parameters. The GA-SVM model had higher accuracy and better applicability. The findings of the current study can provide valuable guide for the construction and management of water-conveyance projects in loess regions.