Deep-learning Enhanced CT Reconstruction Algorithm for Multiphase-flow Measurement

Multiphase-flow measurement cannot effectively capture mesoscale dynamic structures owing to limitations of spatial and temporal resolutions of current measuring techniques. Dynamic X-ray computed tomography (CT), as a non-invasive multiphase-flow measurement technique, is promising for measuring the dynamic structures of multiphase flow. Focusing on the gas–liquid two-phase flow in multiphase flow, this paper addresses limited angle artifacts and excessive reconstruction time in mesoscale dynamic structures and proposes a U-Net-enhanced simultaneous iterative reconstruction technique (SIRT) reconstruction algorithm for bubble-structure measurements based on gas–liquid two-phase flow. Subsequently, based on the hardware design of a flowfield dynamic measurement system, which is a limited-angle dynamic X-ray CT system, a simulated gas–liquid two-phase flow dataset for training the deep-learning model is constructed from three-dimensional bubble structures obtained from hydrogel phantoms. The proposed method yields good results in the training and testing of the constructed dataset and significantly reduces the reconstruction time, thus providing a new technical approach for the high-spatiotemporal-resolution measurement of multiphase-flow mesoscale structures.