Machine learning stands as a potent solution within the intelligent manufacturing sector. However, the conventional training of deep neural networks typically demands extensive datasets, which can be challenging to compile, particularly in various engineering contexts. Physics-Informed Machine Learning (PIML) offers a solution to this challenge by integrating prior knowledge and physical laws to direct model training, thereby augmenting accuracy, interpretability, robustness, and generalization capabilities. Physics-Informed Neural Networks (PINNs), as a model prominent within the PIML landscape, have gained widespread adoption across intelligent manufacturing applications. This paper provides a comprehensive review of the current research on PIML and PINNs, especially in the intelligent manufacturing sector. The analysis is structured around four key dimensions: (1) The methods of physical constraint implementation in PIML; (2) The modeling techniques employed by PINNs; (3) The training methodologies for PINNs; and (4) The industrial physics and potential embedding methods. The paper also outlines existing challenges and potential future research directions in PIML-driven intelligent manufacturing.