基于多尺度融合的蜂窝复合材料缺陷检测网络

Defect Detection Network of Honeycomb Composites based on Multi-scale Fusion

摘要:

针对航空蜂窝板复合材料外部蒙皮破损导致内部产生泥沙、积水以及裂纹等影响飞行安全的问题，提出采用电容层析成像技术(Electrical Capacitance Tomography，ECT)进行蜂窝板复合材料缺陷检测。针对平面ECT成像精度低的问题，通过构建多尺度融合策略、残差编码解码融合模块，引入一种新的池化模块(Soft-pool)等形成多尺度残差编码解码路径的深层神经网络(Multi-scale Residual Encoding and Decoding paths，Ms RED)，使最终的结果完全融合解码阶段学到的特性，对使用共轭梯度成像算法的重建图像进一步改善。结果表明，应用平面ECT技术可以实现蜂窝材料的缺陷检测，通过Ms RED网络可以提升图像重建效果，更清晰重建出蜂窝结构缺陷图像。

Abstract:

In order to solve the problems that the flight safety is affected by internal sediment, stagnant water and cracks caused by the damage of the outer skin of aeronautical honeycomb panels, electrical capacitance tomography (ECT) is proposed to detect the defects of honeycomb composite materials. In order to solve the problem of low accuracy of planar ECT imaging, by constructing multi-scale fusion strategy, residual coding and decoding fusion module, and introducing a new pooling module to form a deep neural network of multi-scale residual coding and decoding path, the final result is fully integrated with the characteristics learned in the decoding stage, and the reconstructed image using conjugate gradient imaging algorithm is further improved. The results show that the defect detection of honeycomb materials can be realized by using planar ECT technology, and the image reconstruction effect can be improved by multi-scale fusion honeycomb composite defect detection network, and the defect image of honeycomb structure can be reconstructed more clearly.