This paper comprehensively reviews Physics-Informed Neural Networks (PINNs). The PINNs combines deep learning and physics knowledge to solve partial differential equations. They can utilize physical laws to guide model learning and show advantages when data is scarce or noisy. It is demonstrated that the PINNs has application potential in fluid dynamics, materials science and other fields. The paper elaborates on the challenges faced by the PINNs, such as overfitting, gradient vanishing, gradient explosion, local optimization and pseudo-optimality, computational challenges, etc. It also introduces the research methods and solutions for the above problems. This paper systematically summarizes the research status of PINNs. It provides important references and directions for future research. It also facilitates its further development and application in science and engineering.