A Multi-source Data Fusion Method for Visualizing Image Reconstruction of Hydrogen Leakage Concentration Distribution

Hydrogen safety is intrinsic to the popularization and application of hydrogen energy. Leakage is a major source of hydrogen-related safety accidents, so the research on leakage is crucial. The visual calibration method can quickly visualize the concentration distribution in the area of hydrogen leakage, but the accuracy of the visualized images needs to be improved. To solve the problem, this paper proposes a multi-source data fusion method based on a deep learning framework, which reconstructs the concentration distribution of the hydrogen leakage and obtains a reconstructed concentration distribution image. Firstly, the leakage images are obtained from the schlieren visualization experiment and using the calibration equations for concentration identification. The visualization experiment is simulated by using ANSYS Fluent, and the simulation result was analyzed and studied with the visualization experiment result. Then, the concentration data obtained from the simulation is used for the training, optimization and validation of the multilayer perceptron neural network, and the axial concentration data obtained from the visualization experiments was used as input to the net to obtain the radial concentration data of the reconstructed visualization image, and the attenuation law of radial concentration at three sections were analyzed. From the result, the reconstructed visualization image by this data fusion method can well reflect the concentration distribution of hydrogen leakage.