Abstract: Metallic materials are widely used in energy engineering, aerospace, marine engineering, and the manufacturing of critical equipment. During long-term service, metallic components are subjected to coupled effects of mechanical loading, temperature, pressure, corrosive media, and other environmental factors. These coupled service conditions can induce corrosion, fatigue, creep, and coupled damage, which may reduce structural integrity and threaten service safety. Accurate assessment of service-induced damage and reliable life prediction are therefore essential for the safe operation, maintenance planning, and life-cycle management of metallic components. Traditional empirical formulas and physics-based models have been widely used for corrosion rate estimation, fatigue life prediction, crack growth analysis, and creep rupture assessment. However, their applications are often limited when the damage process involves multi-source data, high-dimensional variables, nonlinear relationships, and extrapolation under complex service conditions. In this context, machine learning provides a data-driven approach for extracting hidden correlations from complex service data and establishing relationships among material characteristics, environmental conditions, loading parameters, damage features, and life-related indicators. This review focuses on machine learning methods for service-induced damage evolution and life prediction of metallic materials. First, the general workflow of data-driven modeling is summarized, including data acquisition, data preprocessing, feature engineering, model construction, training and validation, and performance evaluation. Based on the modeling characteristics of metallic service damage, several machine learning strategies are discussed. Pure data-driven models are used to establish nonlinear mappings between input features and damage or life indicators. Physics-informed machine learning incorporates empirical formulas, damage mechanisms, physical constraints, or prior knowledge into the modeling process, thereby improving physical consistency under small-sample or mechanism-constrained conditions. Transfer learning uses knowledge from related materials, similar service environments, or low-cost datasets to support prediction under limited target-domain data. Incremental learning allows models to update continuously with newly collected service data, which is relevant to online monitoring and life-cycle assessment. Ensemble learning combines multiple base models to improve prediction stability and reduce the dependence on a single model structure. This review further summarizes recent advances in machine learning research related to typical metallic damage modes, including corrosion rate prediction, fatigue life assessment, crack propagation analysis, creep life prediction, and multi-damage coupling modeling. For corrosion, machine learning models have been used to analyze the effects of alloy composition, environmental parameters, electrochemical signals, and image features on corrosion evolution. For fatigue, related studies focus on life prediction, fatigue strength evaluation, defect-sensitive modeling, time-series damage analysis, and crack identification. For creep, machine learning is applied to creep rupture life prediction, steady-state creep rate estimation, creep stage identification, and remaining life assessment under high-temperature conditions. For coupled damage, hybrid frameworks that integrate physical damage models with neural networks are discussed as a way to describe the interaction among corrosion, fatigue, creep, and other damage mechanisms. On the basis of existing studies, this review identifies several challenges in applying machine learning to metallic service damage assessment, including incomplete data records, insufficient model generalization, limited interpretability, weak physical consistency, and difficulty in online updating. It also offers insights into the development of multi-source data systems, physical mechanism constraints, coupled damage prediction, and intelligent operation and maintenance. These directions are expected to support the transition of machine learning from general performance prediction to service-damage-oriented life assessment and engineering maintenance decision-making.