基于GA-BP神经网络的微裂纹漏磁定量识别技术

针对漏磁检测定量识别技术中识别的缺陷尺寸大多为1~10 mm的较大裂纹，与实际自然裂纹相差太大的问题，将基于遗传算法优化的BP神经网络（GA-BP）算法应用到微裂纹缺陷的漏磁定量识别中，使得漏磁检测定量识别缺陷的宽度、深度达到小于0.50 mm的微细裂纹，并通过基于磁偶极子模型的理论计算与漏磁检测实验两种方法构建了微裂纹（0.10~0.30 mm）缺陷样本库.由于在实际检测过程中存在干扰噪声的原因，实验数据的预测结果误差比理论计算数据预测结果明显偏大，最大为16.73%，但预测结果能够基本反映微裂纹缺陷的尺寸大小. 

Magnetic Flux Leakage Quantitative Identification of Micro Crack Based on GA-BP Neural Network

As the fact that the crack sizes identified based on magnetic flux leakage are larger than 1 mm generally, which are far different from the natural cracks in macro-crack check area. An algorithm with GA-BP neural network was investigated to detect quantificationally the rectangular micro-cracks with less than 0.50 mm width and depth. And a database was developed for micro crack defects among 0.10~0.30 mm based on theoretic calculation of the magnetic dipole model and experiment of magnetic flux leakage. Results show that, due to the noises interference existing in the actual detection process, the prediction error of the experimental data is larger than that of the theoretical data, and the maximum can reach 16.73%, but the prediction results can basically reflect the size of the micro cracks.