基于SSDAE深度神经网络的钛板电涡流检测图像分类研究

钛板电涡流成像检测易受工业现场中的噪声影响,包含噪声的检测图像往往难以提取较好的特征,从而影响分类识别精度。针对以上问题,提出了一种基于栈式稀疏降噪自编码(SSDAE)深度神经网络的钛板缺陷电涡流检测图像分类方法。将稀疏性限制引入降噪自编码器并进行逐层无监督自学习,然后将自编码器栈式组合后添加逻辑识别(LR)层,构建出SSDAE深度神经网络,网络在有监督微调后可实现钛板缺陷电涡流图像特征自动提取与分类识别。稀疏性限制的引入提高了特征学习能力,降噪自编码器的栈式组合提高了深度网络的鲁棒性。实验结果表明,相比其他常规方法,所提出方法不仅在理想环境下有更高的分类准确率,且该方法能有效抵抗噪声,在复杂工况下能更有效地对钛板缺陷进行分类识别。

基于涡流传感的金属表面缺陷检测方法研究

针对发动机涡轮叶片在制造及服役的过程中极易产生微裂纹的问题,以航空铝合金6A02为研究对象,对传统的涡流传感探头进行了有限元建模及参数优化。分析了激励线圈与特定被测金属材料耦合作用时的传感特性,基于巨磁阻(GMR)芯片设计了用于检测微细裂纹的电涡流探头并开发了相应的测试系统,研究了探头扫描方向与GMR芯片敏感轴对检测结果的影响,提出了基于小波分析的噪声消除和特征识别的数据处理方法,得到了良好的效果。实验表明探头具有高灵敏度,能够达到200μm宽度的检测水平。

Combining Contactless Inductive Flow Tomography and Mutual Inductance Tomography for Two-Phase Flow Measurements at a Continuous Casting Model

The flow structure in the mould of a continuous caster is of key importance for the quality of the final product. The use of most conventional flow measurement techniques is prevented by the high temperature of the liquid steel.For a downscaled physical model of the continuous casting process,we present combined measurements of the flow in the mould by Contacfless Inductive Flow Tomography(CIFT),and of the conductivity distribution in the submerged entry nozzle by Mutual Inductance Tomograpliy(MIT).In addition,we summarize an experiment with a magnetic stirrer around the submerged entry nozzle and its effects on the flow in the mould.Some new developments towards a robust implementation of CIFT at a real caster,including the use of pickup coils and gradiometric probes,are also discussed.