This paper investigates the use of physics-informed neural networks (PINNs) for anomaly detection in IIoT-enabled mechatronic systems. Classical data-driven diagnostic approaches based on artificial neural networks (ANNs) often require large amounts of labelled data and may not explicitly enforce the underlying physical laws, which limits their interpretability and robustness in complex industrial environments. In contrast, PINNs embed prior knowledge in the form of governing equations and boundary conditions directly into the learning process. As a proof-of-concept study, we consider a physical pendulum with multiple damage scenarios modelled as changes in geometry and moment of inertia. Synthetic time series of angular position are generated from the analytical model and used to train a PINN that enforces the nonlinear equation of motion and energy conservation. The trained network acts as a physics-based filter and supports the construction of a dictionary of damage states, enabling the assignment of observed trajectories to candidate models. We discuss methodological aspects of the approach, including the definition of the loss function, the role of normalisation and the sensitivity to hyperparameters. Finally, we outline how analogous PINN-based diagnostic methods can be developed for selected subsystems of the CTTP4.0 industrial testbed, which integrates mechatronic transport and manipulation units with an Industrial Internet of Things platform.