Deep Learning Enabled In-Situ Mechanical Characterization for Rapid 3D X-ray Microscopy

The study explores the use of deep learning to enhance X-ray microscopy (XRM) efficiency for in situ mechanical analysis of additively manufactured Inconel 718. Traditional XRM methods suffer from prolonged data acquisition times due to low X-ray source flux, which is problematic for high-resolution 3D datasets needed during tensile testing of dense alloys like Inconel 718. These extended times make comprehensive testing impractical for routine applications. To overcome this, the study employs a deep learning reconstruction method, DeepRecon Pro (a U-net model), to significantly reduce the number of 2D X-ray projections needed for accurate 3D imaging. This approach allows high-quality 3D datasets to be reconstructed using only 100 projections, compared to the usual 1000 or more, resulting in a tenfold decrease in scan time from over 30 minutes to just 6 minutes per load step. This reduction enhances the efficiency of in situ experiments, enabling higher throughput and better utilization of laboratory-based XRM systems. The technique was tested on in situ tensile experiments with additively manufactured Inconel 718 dogbone samples. The model was trained using a full dataset of 1000 projections, then successfully reconstructed high-resolution 3D images using only 100 projections per load step. DeepRecon Pro consistently surpassed traditional methods in time efficiency. The model accurately monitored pore deformation and sample failure with reduced datasets, proving useful for tracking mechanical failure progression in materials like Inconel 718.This advancement facilitates more widespread and routine use of lab-based XRM in studying complex materials, particularly in additive manufacturing and alloy development. The reduction in scan time from over 30 minutes to 6 minutes marks a substantial improvement, allowing researchers to examine more samples in less time, advancing material science research and development.