<p>Accurately building state models and enabling interpretability analysis are critical for models to be widely used and trusted in complex electromechanical systems. Despite its importance, this task remains formidable due to the systems' multi-level, multi-energy domain, and nonlinear characteristics. This paper proposed a novel gray-box approach, the State Function Graph (SFGraph), to solve the challenge. Initially, a general framework for modeling that unifies systems with diverse characteristics is constructed based on a mechanism-guided approach. Three modeling levels are proposed: functional structure, state characterization, and mechanism exploration. Subsequently, a standardized process is developed to partition the overall system into multiple structurally and functionally interrelated subsystems across these three levels using the IDEF0. Additionally, interpretable boxes are designed as basic modeling modules from a mechanistic perspective, with the errors generated during state modeling corrected through data. Multiple boxes are integrated to construct a gray box of the overall system. Finally, the effectiveness of SFGraph is demonstrated through the state modeling and interpretability analysis of a wind turbine. SFGraph achieves unified state modeling across multiple domains by conceptualizing the states of complex electromechanical systems as the dynamic transmission of energy and information. It effectively visualizes the system's operational logic while facilitating interpretability analysis using modular interpretable boxes for graph modeling. This method can be flexibly applied to the state representation modeling, state prediction, and reliability assessment of multi-domain systems.</p>