Since the introduction of Paris’ law, crack growth modeling has been a cornerstone of fatigue damage tolerance design. Mechanism-based models often become complex with multiple influencing factors, while data-driven models, though accurate for specific datasets, lack generalizability. This paper introduces a parametric symbolic regression (PSR) framework that discovers unified crack growth models from heterogeneous datasets. By integrating a multi-objective genetic algorithm and multi-criteria evaluation, PSR optimizes parameters adaptively while maintaining a shared mathematical structure. Using FAA metallic fatigue datasets, the discovered models achieve comparable accuracy to classical NASGRO formulations with fewer parameters and improved interpretability. Compared with existing data-driven approaches, the proposed PSR framework enables the discovery of accurate and stable unified equations across different experimental conditions, providing a powerful tool for cross-domain knowledge discovery and physics-consistent modeling.