In-situ video frame interpolation and super resolution reconstruction for accurate monitoring of L-PBF process

Abstract

In the laser powder bed fusion (L-PBF) process, the incorporation of in-situ monitoring systems plays a vital role in guaranteeing the quality of the additive manufacturing (AM) process. Nevertheless, the monitoring system based on high-speed cameras is hindered by the high cost of the required high-speed cameras, making it difficult to achieve accurate in-situ monitoring. This paper studies in-situ video frame interpolation and super resolution reconstruction for accurate monitoring of L-PBF process. It introduces a novel in-situ video frame interpolation algorithm, termed CS-EMA-VFI, aiming to improve the temporal resolution of monitoring video. The visual transformer-based video super resolution (ViTSR) algorithm was employed to enhance the spatial resolution of the interpolated video. A U-Net algorithm was utilized for extracting the geometric characteristics of the molten pool during the L-PBF process subsequent to video frame interpolation and super resolution reconstruction. Comparing the CS-EMA-VFI with seven state-of-the-art video frame interpolation methods, the CS-EMA-VFI achieves the highest peak signal-to-noise ratio (PSNR) of 28.16 dB and the highest structural similarity index measure (SSIM) of 0.917 while being lightweight. The ViTSR achieved PSNR of 28.18 dB and 25.31 dB on the original video sequence and interpolated video sequence, respectively. The inference time for the CS-EMA-VFI with fixed timestep, ViTSR, and U-Net were recorded as 18.5 ms, 48.0 ms, and 20.5 ms, respectively. The total inference time of the three-stage strategy varies from 87.0 ms to 142.5 ms, depending on the temporal resolution enhancement multiples. Additionally, the proposed three-stage method achieves a segmentation accuracy of 90.15 % with fixed timestep interpolation, simultaneously enhancing temporal and spatial resolution, thus enabling accurate and real-time monitoring. This paper promotes the wide adoption of in-situ monitoring system in the AM field.