Abstract:

In order to quickly and accurately predict the burst pressure of 20 steel elbow with local wall-thinning defects, the SVM model for predicting burst pressure was established. After verifying the simulation accuracy of the explicit nonlinear finite element method, the 168 sets of data of explicit finite element simulation for burst pressure of 20 steel elbows with different defect sizes were used as learning samples of SVM model. CV, GA and PSO were used to optimize the SVM model. The prediction errors were analyzed by comparing the burst pressure calculated by the three optimized SVM model and 3 types of common criterions (ASME B31G-2009, DNV RP-F101 and SHELL 92). The results show that the prediction errors of the three optimized SVM models are less than that of the current common criterions at home and abroad. When CV-SVM and GA-SVM and PSO-SVM models are used to predict the burst pressure of 20 steel elbows with local thinning defects, and the maximum errors of CV-SVM, GA-SVM and PSO-SVM are-2.33%,-3.4% and 1.94% respectively. SVM model is easy to use, has high prediction accuracy, good engineering practicability and short time consumption.

Key words: support vector machine (SVM), local wall-thinning defects, 20 steel elbow, burst pressures, cross validation (CV), genetic algorithm(GA), particle swarm optimization (PSO)