【摘要】 针对激光定向能量沉积（LDED）熔覆层强度实时可靠评估难题，提出了基于双路光电倍增管（PMT）谱线比时间尺度纹理图的激光能量沉积稀释率实时评估方法。利用所开发的双路窄带滤光片-PMT采集系统，实现了对基板合金元素和熔覆粉末合金元素等离子体光谱信号的同步采集，进一步通过多尺度滑动窗口将所选合金元素光谱信号比值转换为二维纹理图，有效保留了光谱信号动态结构与时间相关性；最后提出了融合可变形卷积与SKAttention机制的改进深度学习网络SK-D ResNet稀释率分类模型。验证结果表明，该网络在测试集上实现98.03%的分类准确率，较传统卷积网络提升4.77%以上。所提方法为核能等领域重大装备关键构件LDED修复与制造过程熔覆层质量的可靠高精度监测提供了有效解决思路。更多还原

【Abstract】 To address the challenge of real-time and reliable evaluation of the strength of laser-directed energy deposition(LDED) clad layers, this paper proposes a real-time evaluation method for laser energy deposition dilution rate based on the time-scale texture map of the spectral line ratio using a dual-channel photomultiplier tube(PMT). A developed dual-channel narrow-band filter-PMT acquisition system is utilized to achieve synchronous collection of plasma spectral signals from substrate alloy elements and cladding powder alloy elements. Furthermore, by employing multi-scale sliding windows, the onedimensional spectral line ratio signal is converted into a two-dimensional texture map, effectively preserving the dynamic structure and temporal correlation of the spectral signals. Finally, an improved deep learning network for dilution rate classification, named SK-D ResNet, is proposed, which integrates deformable convolution with the SKAttention mechanism. Verification results show that the network achieves a classification accuracy of 98.03% on the test set, representing an improvement of more than 4.77% compared to traditional convolutional networks. The proposed method provides an effective solution for reliable and high-precision monitoring of LDED repair and manufacturing processes for critical components of major equipment in fields such as nuclear energy.更多还原