Deep learning-enhanced rapid μ-CT characterization of additively manufactured components with a long-/short-exposure scanning strategy

X-ray micro-computed tomography (μ-CT) enables precise characterization of internal defects in additively manufactured (AM) components but is hindered by long scanning times, limiting its suitability for high-throughput applications. This study proposes a deep learning-enhanced characterization method that integrates a long−/short-exposure μ-CT scanning strategy to achieve both speed and accuracy. High-quality long-exposure images were aligned with rapid short-exposure images to construct training datasets, and a U-Net convolutional neural network was trained to reconstruct high-precision 3D models from short-exposure data. The approach reduces scanning time by approximately 85 %. Using Ti-6Al-4 V body-centered cubic (BCC) lattice specimens as validation cases, three finite element models were established to evaluate geometric fidelity and mechanical accuracy. The reconstructed models achieved thickness errors within 10 μm and produced compression response predictions in close agreement with experimental results. A systematic evaluation of exposure times demonstrated that images with ≥1 s exposure achieved signal-to-noise ratios (SNR) above 15, meeting structural characterization requirements. Overall, this method provides an efficient and scalable solution for μ-CT characterization of complex AM components, enabling reliable high-throughput quality control without compromising accuracy.