BPNN Based Multi-factor Synthesis Prediction Method for Coal Rock Fatigue Life

The fatigue characteristics and failure mechanisms of rocks under cyclic loading not only have important theoretical significance, but are also crucial to ensuring the long-term stability of engineering rock masses. Numerical methods are widely used in fatigue life analysis. However, traditional numerical models have shortcomings such as unclear nonlinear characteristic relationships and poor life prediction accuracy in the simulation of rock fatigue damage and destruction. In recent years, neural network methods have been proven to have significant advantages in solving strong nonlinear relationships. This manuscript proposes a backpropagation neural network method to predict the fatigue life of coal and rock based on coal fatigue test data sets. The nonlinear relationship between the number of cyclic loadings and mechanical parameters such as confining pressure, cyclic loading frequency, cyclic load level, upper limit stress level and upper limit stress intensity during coal failure was established. The results show that the R measurement is always higher than 0.9 and the method has high feasibility in predicting rock fatigue cracks. Most importantly, combined with different conditions, the model can comprehensively describe the nonlinear relationship between rock volume strain and rock fatigue life. The model is repeatable and scalable, providing new ideas and methods for rock fatigue life research.