In-situ optical emission spectroscopy monitoring of laser powder bed fusion based on spectral data prior knowledge and feature-decision level fusion network

Porosity defects are common quality issues in Laser powder bed fusion (LPBF) manufacturing. Optical emission spectroscopy (OES), a key technique that reflects the interaction between the laser and the material, has been widely used in laser processing monitoring. However, the spectral signals in the LPBF process are extremely weak and complex, and there is a lack of reliable OES monitoring approaches. To address these challenges, this study proposed an OES monitoring framework driven by spectral data prior knowledge. First, the spectral signals and plasma features of different porosity defects in the LPBF process were systematically analyzed, and the correlation between the process and spectral signals was explored, revealing the defect-related spectral features of both emission lines and the background continuous spectrum, along with their underlying physical mechanisms. Building on this, a physics-informed spectral signal-to-image mapping method was proposed, which converted high-dimensional spectral data into dual-channel spectral grayscale mapping images, thereby enhancing spectral features strongly correlated with defects. Finally, the MSI-FDFNet model was developed for in situ porosity monitoring during the LPBF process. Results demonstrate that the proposed method effectively extracts distinctive features from different spectral bands, achieving an accuracy of 95.51% in identifying multiple porosity defects, significantly outperforming models using only feature-level or decision-level fusion. Additionally, visualization analysis confirms that the model can adaptively focus on defect-sensitive key spectral band information, enabling reliable monitoring through multi-level fusion. This study highlights the potential of physics knowledge-driven OES monitoring for in-situ monitoring of the LPBF process.