Abstract: Traditional Transfer Path Analysis (TPA) methods require fixed sensor placement to capture vibration characteristics, while physical space constraints or harsh environmental conditions in practical engineering often prevent measurements at critical nodes, significantly compromising analytical accuracy. To address this limitation, this study proposes a physics-monitoring integrated vibration transfer path analysis method. Firstly, a Physics-Informed Neural Network (PINN) model is constructed by embedding elastodynamic equations into neural networks, and a vibration response prediction framework is developed under the constraints of governing differential equations, combining partial surface acceleration measurements. Secondly, a numerical simulation of a three-dimensional cantilever beam system is conducted to evaluate the proposed method's path identification performance with conventional TPA, and experimental results show that the proposed method obtains a peak frequency relative error of 0.15% and absolute amplitude error of 6.19 dB, outperforming traditional TPA and serving as a viable substitute. Thirdly, near-field energy streamline entanglement near excitation sources can be revealed by power flow field analysis, dominant transfer paths can be identified, and vortex-induced energy dissipation at free boundaries can be characterized. Finally, this study not only effectively resolves measurement limitations at critical nodes, but also provides a novel methodology for analyzing multi-source vibration energy transmission mechanisms in complex mechanical systems.