Petroleum products rely on additives to enhance performance across transportation, construction, and energy sectors. These additives regulate viscosity, friction, wear resistance, and thermal stability. Traditional additive research depends on empirical trial and error, which is slow, costly, and limited in scope. Recently, machine learning (ML) methods have been applied to identify complex structure–property relationships, predict performance before experimentation, and guide formulation strategies. This review presents a unified framework for ML integration into petroleum additive research, organizing existing studies into a systematic taxonomy that links data modalities, algorithmic choices, and application domains. It overviews taxonomy of ML tasks and a unified framework that maps data types, methods, and application domains relevant to lubrication science. Through representative examples, the review illustrates the breadth of current applications, ranging from gallate additive design using deep neural networks informed by quantum‐scale data, to asphalt binder optimization through predictive modeling, and the screening of lubricants using statistical and tree‐based methods, and highlights emerging efforts that apply deep learning to real‐time lubricant degradation and condition monitoring through multi‐sensor data fusion. The review provides a practical roadmap for addressing the key limitations and accelerating additive discovery, enhancing predictive maintenance, and supporting the development of more sustainable lubricant formulations.
Shah et al. (Sun,) studied this question.