Prediction of Crack Expansion in Ultra-High Temperature Ceramics Based on BP Neural Network

In this pape, a BP neural network-based method is proposed for predicting the crack length of ultra-high-temperature ceramics. It effectively solves the problem faced by crack length detection inside aerospace vehicles.The paper starts with an analysis of the stress intensity factor K and the stress, strain and G release rate as the main factors affecting the Crack extension based on the fracture mechanics theory of brittle solids. Using a pre-set cracked plate as an example, numerical simulations are carried out using abaqus to analyse the cracking of ultra-high temperature ceramics and to collect data about the crack length. The resulting data is pre-processed to obtain the required samples for the next step of deep learning.Based on the above research, the paper uses five main factors affecting crack expansion as input variables crack size as output variables to construct a BP neural network prediction model, 200 samples were selected as the test set and 810 samples as the training set, the prediction index MSE finally converged to 0.0010975 can be seen that the prediction effect is good, but there are still error samples, the paper continues The parameters of the model were improved using a reasonable method. The BP neural network model was optimized using the sparrow search algorithm (SSA), the threshold and initial weights of the BP neural network were adjusted, and a comparative analysis of the prediction calculation of the fracture expansion of ultra-high temperature ceramics was also performed.Based on the above study, the paper constructs a BP network model using five main factors affecting crack expansion as input variables, selects 200 prediction samples and 810 training samples, and improves the parameters of the model using reasonable methods. And the BP neural network model was optimized using the sparrow search algorithm. The threshold and initial weights of the BP neural network were adjusted, and the prediction calculation of the fracture extension of ultra-high-temperature ceramics was also compared and analyzed.The results show that the improved algorithm effectively reduces the prediction error. the MSE finally converges to 0.00046607. the prediction index shows that the optimized BP neural network prediction model using the sparrow search algorithm is feasible, and the method in the paper has a good application in the detection of aerospace vehicles.