<p>This paper presents the adaptive multi-domain physics-informed neural networks (PINNs) for simulating the dynamic coupled thermo-mechanical behaviors in two- (2D) and three-dimensional (3D) thermal barrier coatings (TBCs). We decompose the substrate-coating system into multiple layers, each modeled by a dedicated neural network, utilizing domain decomposition technique. We integrate measured data with physical laws to train the network by embedding partial differential equations and incorporating boundary, initial, and continuity conditions into the loss function. An adaptive loss balancing algorithm is introduced to balance the interactions among different terms in the loss function during training process. The performance of the developed framework is evaluated through several numerical examples, including 2D rectangular and arc-shaped single-layer materials, 2D and 3D bilayer materials, as well as 2D and 3D substrate-coating systems composed of GH4169 nickel-based alloy, NiCoCrAlY, and ZrO₂. The numerical results demonstrate that the adaptive multi-domain PINNs effectively and accurately solve dynamic coupled thermo-mechanical problems in TBC systems in a unified and concise manner. Furthermore, compared to the conventional finite element method, the proposed framework achieves higher computational accuracy with less data.</p>