Combining materials design and deep learning: AI-enhanced luminescence thermometry with a novel Eu3+/Tb3+ polymeric coordination compound

Abstract

Conventional thermal sensors often face limitations due to their reliance on direct contact and restricted measurement ranges, leading to the emergence of novel techniques like luminescence thermometry. However, sensitivity of luminescent thermometers is limited by the only used Boltzmann-based Mott–Seitz model, which is imperfect. To overcome this, we complemented Mott–Seitz model applying machine learning algorithms, achieving supreme accuracy improvement. Thus, here we report a combined approach to luminescence thermometry, utilizing novel mixed-metal polymer Eu³⁺/Tb³⁺ tris-complex and a deep learning algorithm. The complex, synthesized using 4,4,4-trifluoro-1-(5,5-dimethyl-1H-pyrazol-4-yl)butane-1,3-dione, exhibits maximum relative thermal sensitivity of 5.5% K⁻¹ and a temperature uncertainty ranging from 0.1 to 1.8 K across a wide temperature range (190 to 300 K). We enhanced accuracy seven-fold from RMSE 2.54 K for the conventional intensity ratio method to RMSE 0.36 K for combined method using convolutional neural network. These results highlight the potential of combined approach to achieve record-high precision thermometers even for common compounds.