Physics-informed neural networks (PINN) are a popular paradigm of scientific machine learning which incorporate the domain knowledge in the form of differential equations into the loss function of the neural network representing the solution. Though this approach has been extensively studied for solving a variety of differential equations including heat equation, the use of PINN for system level heat transfer modelling is not explored much. In the present work, taking spacecraft thermal modelling as an example, use of PINN for modelling coupled, multi-mode heat transfer phenomenon at the system level is studied. A simplified compact thermal model for a spacecraft sub-system consisting of fewer nodes (e.g. heat source, conductive-convective heat flow path, radiative sink) is developed. In present study, Tensorflow, the popular open-source machine learning framework is utilized. Results are validated using an equivalent thermal mathematical model developed on commercial software NX/TMG. Validation is done for both steady and transient load profiles. A good agreement of ±2 °C is obtained between PINN and NX/TMG.