Abstract: In transmission systems employed within industrial equipment, gears are of critical importance as core components. The operational reliability of these gears directly determines the safety of the equipment in use and its economic efficiency. In order to achieve precise intervention before the occurrence of gear wear failure, condition monitoring technology based on oil analysis has become a core method for predictive maintenance. Conventional methods are inadequate in terms of their ability to correlate multi-dimensional oil data.　　In recent years, machine learning and multi-parameter fusion methods driven by mathematical models have significantly improved the accuracy of wear state diagnosis and remaining useful life (RUL) prediction, especially addressing the engineering limitations of traditional offline detection (requiring equipment shutdown and disassembly) and single-dimensional monitoring (such as the high cost of vibration analysis). Based on multiple industrial cases and academic literature, this paper comprehensively reviews the technological evolution of gear wear monitoring, with a focus on data fusion, statistical model optimization, and machine learning-driven multi-parameter analysis strategies for oil. Ultimately, it analyzes the key challenges (sample dependence/sensor calibration/model generalization) and breakthrough paths (micro-MEMS sensors/digital twins/small sample learning) in this field, providing a theoretical paradigm for intelligent equipment health management.