This work proposed a new model based on transformers for multimodal image fusion, with explicit attention paid to fusing infrared and visible images toward enhanced detail and information content. This method, which incorporates Nystrom attention and skip connections, optimizes the fusion process so that important features of both modalities are preserved. Traditional models find it hard to keep details in the content and introduce artifacts. On the contrary, this model can efficiently cope with these challenges, as it makes use of an attention mechanism that is computationally highly effective. This will efficiently fuse large-scale image data and may be useful in those fields where very detailed and comprehensive visual information is demanded. The detailed evaluations prove that our model has superior performance over the current state-of-the-art methods in texture detail preservation, thermal differentiation, and overall image quality, which possesses great potential for advanced imaging applications.

ABSTRACT Real‐time surface quality monitoring of selective laser melting (SLM) is critical for defect identification and process stability but is still plagued by the trade‐off between high precision and efficiency. We propose DyLoFViT, a light‐weight vision transformer specifically optimized for industrial SLM defect classification. DyLoFViT makes four significant architectural contributions: (1) dynamic positional encoding that retains spatial coherence and facilitates translation invariance, (2) dynamic sparse gating that adaptively filters the most informative tokens, (3) Nyström‐based sparse self‐attention that estimates long‐range dependencies with lower computation, and (4) frequency‐domain compression with discrete cosine transform that filters high‐frequency noise with preservation of low‐frequency structure vital for defect classification. Experiments on an industrial test set of 1320 high‐resolution images of four melt quality classes show that DyLoFViT achieves a Top‐1 accuracy of 96.1% and an F1‐score of 95.6% with merely 13.2M parameters and 2.9 GFLOPs. Compared with the prior CNN and transformer architectures, DyLoFViT achieves superior precision with an order of magnitude lower inference latency (0.3 ms/image), making it extremely well‐suited for real‐time, edge‐device deployment in additive manufacturing. DyLoFViT hence sets a new standard for efficient, accurate in situ metal 3D printing defect classification.

Defects in the metal 3-D printing process exhibit randomness and low frequency, making them difficult to predict and control. This severely hinders the application of this technology in critical industrial fields. Extracting useful features from massive process monitoring data to ensure forming quality has become a popular research direction for intelligent additive manufacturing practitioners. In this study, a coaxial machine vision monitoring system is utilized to monitor the entire forming process of the melt track. More importantly, this study constructed a high-speed video dataset of typical metal 3-D printing working conditions by changing the powder layer thickness, which can serve actual industrial production. To promptly recognize unhealthy melt tracks from massive process monitoring data, a 2-D transformer-based framework named super frame feature pyramid transformer (SFFPT) is designed for video classification. This framework transforms the video understanding task into a 2-D feature map processing task, allowing the video classification task to be completed directly using only an image classifier. In comparison to state-of-the-art methods, SFFPT achieves the best classification accuracy in this study.

Porous structures had traditionally been generated using triply periodic minimal surfaces (TPMS), which however often encounter problems with the continuity and integrality of the generated curved surfaces when applying to irregular models. The present work describes a new porous-structure designing method using a combination of parameterized hexahedral mesh and traditional TPMS surfaces. The primary steps of the proposed method include: 1) re-meshing an input model into a hexahedral mesh; 2) solving parametric expressions of 12 geodesic lines using heat conduction models; 3) constructing quadrilateral surface meshes, and hexahedral volume meshes with harmonic-map; 4) generating standard TPMS surfaces within a regular hexahedron using the Marching Cube (MC) algorithm and then mapping the generated surface onto the parameterized hexahedral mesh using tri-linear interpolation. The experimental results have shown that the present method can solve the problem of discontinuity when the constraint hexahedron is deformed.