This study explores the potential of physics-informed neural networks (PINNs) for realizing digital twins (DT) from various perspectives. The research investigates adaptive sampling techniques for collocation points to verify their effectiveness in mesh-free frameworks, allowing automated construction of virtual representations without manual mesh generation. The performance of the data-driven PINNs (DD-PINNs) framework is examined, which can utilize acquired datasets in DT scenarios. The scalability of DD-PINNs to more general physics is validated using parametric Navier-Stokes equations, where PINNs do not need to be retrained as the Reynolds number varies. Additionally, multi-fidelity DD-PINNs are proposed and evaluated, showing significant improvement in extrapolation tasks with 42-62% better performance compared to single-fidelity approaches. The uncertainty quantification performance of multi-fidelity DD-PINNs is also investigated using an ensemble method to verify their potential in DT. The study concludes that DD-PINN frameworks are more suitable for DT scenarios than traditional PINNs, bringing engineers closer to seamless DT realization. Digital twins, Physics-informed neural networks, Adaptive sampling, Data-driven approach, Multi-fidelity data, Parametric Navier-Stokes equations, Uncertainty quantificationThis study explores the potential of physics-informed neural networks (PINNs) for realizing digital twins (DT) from various perspectives. The research investigates adaptive sampling techniques for collocation points to verify their effectiveness in mesh-free frameworks, allowing automated construction of virtual representations without manual mesh generation. The performance of the data-driven PINNs (DD-PINNs) framework is examined, which can utilize acquired datasets in DT scenarios. The scalability of DD-PINNs to more general physics is validated using parametric Navier-Stokes equations, where PINNs do not need to be retrained as the Reynolds number varies. Additionally, multi-fidelity DD-PINNs are proposed and evaluated, showing significant improvement in extrapolation tasks with 42-62% better performance compared to single-fidelity approaches. The uncertainty quantification performance of multi-fidelity DD-PINNs is also investigated using an ensemble method to verify their potential in DT. The study concludes that DD-PINN frameworks are more suitable for DT scenarios than traditional PINNs, bringing engineers closer to seamless DT realization. Digital twins, Physics-informed neural networks, Adaptive sampling, Data-driven approach, Multi-fidelity data, Parametric Navier-Stokes equations, Uncertainty quantification