Fracture properties of dolomite and prediction of fracture toughness based on BP-ANN

In this study, based on the digital image correlation technique, three-point bending tests were conducted on dolomite semi-circular bending (SCB) specimens, and the fracture properties of the SCB specimens, including crack opening displacement, apparent fracture toughness (K_If) calculated using the initial crack length and true fracture toughness (K_Ic) calculated using the effective crack length, were investigated first. Then, two datasets were constructed, incorporating specimen size parameters (such as radius, span, and initial crack length), tensile strength and fracture toughness. Considering the impact of specimen size, three backpropagation artificial neural network (BP-ANN) models were developed, two for predicting K_If and one for K_Ic respectively. Finally, the size effect of fracture toughness was studied. The results show that nonlinearity is exhibited in the pre-peak load-crack opening displacement curve. The analysis about the evolution of crack opening displacement and strain concentration zone suggests that the nonlinearity observed before the peak could be attributed to the emergence and progression of cracks. The K_Ic of D-1 specimen (1.86 MPa·m^1/2) is significantly larger than the K_If (1.08 MPa·m^1/2), indicating that the fracture toughness estimated using the LEFM underestimates the inherent toughness of the rock. The trained BP-ANN models have good predictive and generalization performance. As the span and initial crack length of the specimen increase, a gradual decrease is observed in both K_Ic and K_If. Simultaneously, a progressive downward shift is noted in both the K_If-f_t and K_Ic-f_t curves.