The utilization of Deep Neural Networks (DNNs) in physical sciences and engineering applications
has gained traction due to their capacity to learn intricate functions. These models require large
training datasets, which can be prohibitively expensive for engineering applications. The
development of a new field of scientific machine learning, Physics Informed Machine Learning,
addresses this issue of reliance on large datasets for training the models. This is achieved by
modifying the loss function by incorporating the governing partial differential equation (PDE) into
the training process as a soft constraint. The PDE acts as a regularizing agent, ensuring that the
predictions of the models adhere to the physics of the phenomenon. This work envisages building
physics-informed surrogate models to reconstruct flow field data from sparse datasets. The
proposed surrogate models are built for two test problems: i) Wake flow reconstruction of laminar
flow past a circular cylinder and 2) Wake flow reconstruction of turbulent flow past an Ultra Large
Container Ship (ULCS). The training dataset is produced utilizing commercial CFD software. Spatial
locations within the wake field are chosen randomly to form a sparsely sampled dataset, which
serves as the training data. CFD data also serves as the ground truth for validating the predictions of
the Physics Informed Neural Network (PINN) model. The success of this study hinges on the adept
development of models that serve as surrogates for existing CFD models. These models are designed
to facilitate rapid inference across the spatiotemporal domain on which they have been trained.