Machine learning-assisted optical thermometry with carbon dots: photophysical foundations and intelligent temperature sensing
Abstract
Carbon dots (CDs) have emerged as promising luminescent nanothermometers for optical thermometry owing to their tunable optical properties, low toxicity, environmental compatibility, and structural versatility. However, the intrinsic complexity of their emission mechanisms and the multidimensional nature of synthesis–structure–property relationships continue to hinder the rational development of high-performance temperature-sensing platforms. This review examines the rapidly evolving intersection of CD thermometry and machine learning (ML), with emphasis on how data-driven methodologies are transforming both materials design and thermal-signal interpretation. The fundamental photophysical origins of temperature-dependent luminescence are first analyzed, focusing on the interplay between structural heterogeneity, excited-state dynamics, and thermometric information encoding. Subsequently, machine-learning frameworks relevant to optical sensor development are discussed from the perspectives of descriptor engineering, predictive modeling, uncertainty assessment, and adaptive design. By synthesizing recent advances in ML-assisted CD research, this review identifies emerging design principles governing optical-property prediction, luminescence optimization, and multidimensional temperature sensing. Particular attention is given to the transition from empirical optimization toward predictive and increasingly autonomous thermometric systems. The analysis further highlights unresolved challenges related to mechanistic ambiguity, model transferability, reproducibility, standardization, and real-world deployment. Finally, a future roadmap is proposed in which physically informed ML, integrated data infrastructures, and closed-loop discovery frameworks converge to enable the next generation of intelligent CD thermometric technologies.