基于光谱物理特征感知网络的 DED 稀释率监测

The dilution rate is crucial for the metallurgical bonding strength and forming precision between the substrate and the cladding layer in laser energy deposition. However, existing monitoring methods find it challenging to perform online quality monitoring. Therefore, a real-time dilution rate monitoring system based on a Y-dual-channel fiber in the DED process was developed. This system collects plasma spectral signals and extracts the key representative elemental line ratios of the substrate and powder to characterize the dilution rate variation. The Pi-LGNet, a spectral physical feature perception network, was established, using preprocessed spectral signals and extracted elemental line ratios as dual-channel inputs, achieving classification and identification of the dilution rate during the DED process. The results show that the extracted key representative elemental line ratios have a strong correlation with the dilution rate, and the proposed Pi-LGNet network model achieves an accuracy of 91.8%. Ablation and comparative experiments confirm the superiority of this network in spectral signal recognition.