基于BP神经网络的超声无损测定表面开口裂纹高度

Ultrasonic testing of height of surface breaking cracks base on BP neural networks

为了准确测量服役材料中的裂纹高度,研究了基于端点回波反射法的BP神经网络模型,进行了K值及对应高度裂纹的端点反射波声程差共28组数据作为输入向量对网络训练。结果表明:网络预测值与实际裂纹高度符合得很好,绝对误差均保持在±1mm之间;可以准确快速地预测表面开口裂纹高度,实现测量的智能化。

In order to test the height of cracks in the in-service components, a BP neural network based on the tip echo method was introduced and established, the 28-group data containing different K values and different heights of cracks were taken as the inputting vector and trained, the heights of cracks were simulated by the established model, the simulated results were in agreement with the actual crack heights, absolute error was within -1 to ＋lmm. It was shown that the tip echo method combined with BP neural network could forecast the height of surface breaking crack precisely,and make the testing more intellectual.