Intelligent Moiré deflection tomography using physically processed deep learning

Abstract

Moiré tomography is a vital technique for diagnosing high-temperature and complex flow fields. However, traditional information extraction and reconstruction methods lead to significant time and resource expenditures. In this paper, we propose a Deep Learning-based Moiré Deflection Tomography (DLMDT) method that achieves high-precision predictions of the refractive index distribution in measured flow fields while ensuring the results align closely with physical principles. A carefully designed two-step deep learning model serves as the foundation of this approach. The model's accuracy and computational efficiency are first validated through numerical simulations. Subsequently, the trained model is applied to the diagnos of real flow fields. The findings reveal that the proposed model accurately predicts the refractive index distribution of measured flow fields, exhibiting high precision and structural similarity. Compared to traditional moiré tomography, our proposed method significantly enhances the speed of high-precision diagnostics for high-temperature and complex flow fields, which provides a robust foundation for advancing the intelligent capabilities of moiré tomography.