This study explores the potential of data-driven physics-informed neural networks (DD-PINNs) for the realization of digital twins (DT). The research investigates various adaptive sampling techniques for collocation points in PINNs, which enable mesh-free construction of virtual representations without manual mesh generation. The performance of DD-PINNs is evaluated in DT scenarios, showing improved scalability and prediction accuracy compared to traditional PINNs. Multi-fidelity DD-PINNs are proposed to handle datasets with varying fidelity, achieving significant improvements in extrapolation tasks. The study also examines uncertainty quantification in DT scenarios, demonstrating the potential of DD-PINNs for accurate predictive uncertainty measurement. The research highlights the advantages of DD-PINNs in DT applications, including automated virtual space construction, data-driven model updating, and scalability to general physics. The study proposes a novel vorticity-aware adaptive sampling technique, which enhances the performance of PINNs in fluid dynamics problems. The results show that DD-PINNs outperform data-free PINNs in predicting flow fields, especially at higher Reynolds numbers. The study also investigates the effects of hyperparameters such as stochastic intensity and collocation sampling ratio on the performance of DD-PINNs. The findings indicate that careful selection of these parameters is crucial for achieving accurate predictions. The research concludes that DD-PINNs are more suitable for DT scenarios than traditional PINNs, offering improved performance in real-time prediction and uncertainty quantification.This study explores the potential of data-driven physics-informed neural networks (DD-PINNs) for the realization of digital twins (DT). The research investigates various adaptive sampling techniques for collocation points in PINNs, which enable mesh-free construction of virtual representations without manual mesh generation. The performance of DD-PINNs is evaluated in DT scenarios, showing improved scalability and prediction accuracy compared to traditional PINNs. Multi-fidelity DD-PINNs are proposed to handle datasets with varying fidelity, achieving significant improvements in extrapolation tasks. The study also examines uncertainty quantification in DT scenarios, demonstrating the potential of DD-PINNs for accurate predictive uncertainty measurement. The research highlights the advantages of DD-PINNs in DT applications, including automated virtual space construction, data-driven model updating, and scalability to general physics. The study proposes a novel vorticity-aware adaptive sampling technique, which enhances the performance of PINNs in fluid dynamics problems. The results show that DD-PINNs outperform data-free PINNs in predicting flow fields, especially at higher Reynolds numbers. The study also investigates the effects of hyperparameters such as stochastic intensity and collocation sampling ratio on the performance of DD-PINNs. The findings indicate that careful selection of these parameters is crucial for achieving accurate predictions. The research concludes that DD-PINNs are more suitable for DT scenarios than traditional PINNs, offering improved performance in real-time prediction and uncertainty quantification.