基于数据增广灰色神经网络的ACFM裂纹角度预测

Prediction of ACFM Crack Angle Based on Data Augmented Grey Neural Network

交流电磁场检测(alternating current field measurement, ACFM)技术广泛应用于制造业等工业领域中金属结构物的缺陷检测。针对单传感器在非预知缺陷检测过程中存在的角度偏转及裂纹定位等问题展开了研究,首先通过COMSOL Multiphysics仿真结果可知:场强的X和Y方向分量在角度偏转的过程中存在信号互补的规律,然后通过建立比例因子进而实现了数据增广型灰色神经网络模型(data augmented grey neural network model, DA-GNNM)的预测,同时模拟预测对比回归预测可知DA-GNNM模型的预测效果较优。此外通过多梯度偏转仿真实现了偏转裂纹的重构,其次通过搭建实验平台以及信号特征提取等工作验证了DA-GNNM预测模型的合理性,平均预测误差2.56%;最后通过预测角度进一步改善了非平行检测过程中裂纹重构图像的偏转问题。

Alternating current field measurement(ACFM)technology is widely used in defect detection of metal structures in manufacturing and other industrial fields.The angle deflection and crack location problems of single sensor in the process of unpredictable defect detection were studied.Firstly,through the COMSOL Multiphysics simulation results,it can be seen that the X and Y direction components of the field strength have a law of signal complementarity during the angular deflection process,and then the data augmented grey neural network model(DA-GNNM)prediction was realized by establishing a scale factor.At the same time,simulation prediction and regression prediction showed that DA-GNNM model has better prediction effect.In addition,the deflection crack was reconstructed by multigradient deflection simulation.Secondly,the DA-GNNM prediction model was verified to be reasonable by building an experimental platform and signal feature extraction,with an average prediction error of 2.56%.Finally,the deflection of the reconstructed image of the crack in the non-parallel detection process is further improved through the prediction angle.