To optimize structures and monitor their health, it is essential to build an accurate dynamic analysis model. However, traditional modeling methods based solely on physical information or data-driven techniques may not suffice for many engineering applications. While physical models can accurately simulate complex equipment, they may also incur high computational time. On the other hand, data-driven models may improve computational efficiency but are subject to significant deviations due to the influence of training data. To address these challenges, the Physics-Informed Neural Network (PINN) has gained popularity for imposing physical constraints during the training process, leading to better generalization capabilities with fewer data samples. This paper proposes a physics-informed hybrid modeling (PIHM) approach that combines a reduced-order model, kernel functions, and dynamic equations to predict dynamic output with limited training data and physical information. The method integrates prior physics information into function approximation by incorporating the reduced dynamic equation into a surrogate modeling framework. The loss function considers inertial and damping effects, ensuring physical plausibility. Unlike traditional PINN applications, the proposed modeling method is more explainable, as the trained model can be expressed in function form with engineering interpretation. The approach is verified with a real-world engineering example (telehandler boom) under complex load conditions, demonstrating accuracy, efficiency, and physical plausibility. Overall, the proposed method offers promising capabilities in solving problems where high-fidelity simulation is challenging.