Engineering construction in loess areas often requires improvement of loess. Basalt fiber-reinforced soil due to complexity of material composition often exhibits poorly predictable mechanical behavior. In this study, digital image technology-based unconsolidated-undrained (UU) triaxial shear tests were carried out on loess samples at three fiber lengths (L) and four fiber contents (η). Results prove the improvement of the shear strength of loess by basalt fiber inclusion, which varies in inverted u-shaped pattern with fiber length or fiber content, with the maximum at η = 0.6%, L = 12 mm. Digital Image Technology was employed to analyze the damage characteristics and strain field of the surface of the sample at different loading time. The volumetric strain of the reinforced sample decreases at higher fiber content or fiber length, from shear contraction to dilatancy. The unreinforced sample exhibits a typical brittle failure mode with visible shear band, while plastic failure for reinforced samples with an overall bulging failure mode. A statistical damage constitutive model of fiber-reinforced loess was established with limited parameters calibrated. The rationality of the model was verified by comparisons of measured and calculated stress-strain data.

A uniaxial strain-softening constitutive model for fiber-reinforced soils is developed based on the disturbed state concept (DSC). The response in the relatively intact state is assumed to satisfy the Duncan–Chang model obtained from the prepeak stress–strain curve, while the fully adjusted state response satisfies the linear model obtained by an extension of the residual strength. The apparent stress–strain curve is a weighted average response derived from the two aforementioned response curves by a disturbance function that acts as the weight. The peak of the stress–strain curve and the postpeak stable point are assumed as the starting and ending points of the disturbance, respectively, which assign a reasonable physical sense to the parameters in the disturbance function. Comparisons of stress–strain curves and peak strength reveal that for a specified fiber, fiber content exhibits a greater influence on the reinforcement effect than fiber length. Five required parameters that vary with fiber content are used in the DSC model. Five sets of uniaxial compression test data of different fiber-reinforced soils are evaluated, and a high consistency between the stress–strain curves predicted by the DSC model and the test curves is noted. Both the consistency index δ and the energy absorption capacity reveal a satisfactory description of the fiber reinforcement effect.