Akademik Veri Yönetim Sistemi
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE, cilt.167, sa.Part I, ss.1-16, 2026 (SCI-Expanded, Scopus)
The simultaneous reconstruction of temperature fields and radiative properties in industrial furnaces is a significant ill-posed inverse problem, characterized by strong parameter coupling and high computational cost. Although traditional decoupled reconstruction methods are accurate, their intensive iterative processes make them impractical for online monitoring. This paper proposes a deep learning-based method for the simultaneous reconstruction of two-dimensional temperature fields and radiative properties. A radiative imaging model was first developed to relate the internal temperature field, radiative properties, and boundary radiative temperatures. A large dataset was generated based on radiative transfer forward problem. A multi-task deep neural network was then developed and trained on the dataset. In the simulation study, the reconstructed temperature field and radiative properties showed average relative errors of less than 1 % and 6 %, respectively. Comparative analysis with the decoupled reconstruction method demonstrates that the model achieves similar accuracy while reducing computation time from 2 s to 8 μs, enabling real-time capability. Furthermore, the model maintains accurate reconstruction under various noise conditions, demonstrating robustness. Experimental validation on a 350 MW (MW) furnace, using transfer learning with limited annotated data, confirmed effectiveness of the method. As furnace load increased from 260 to 320 MW, reconstructed temperature fields increased from 1424 to 1627 K (K), while absorption and scattering coefficients gradually increased. The average relative error of inversely calculated boundary radiative temperature under different loads was less than 1.5 %. These indicates that the proposed method has promising potential for the study of flames.