Physics-Informed Neural Networks (PINN) stand out, adapting neural networks to predict solutions to Physics phenomena. One of the recent approaches for DNNs is Physics-Informed Neural Networks (PINNs) which are revolutionizing the approaches to problems governed by partial differential equations (PDEs) in Science and Engineering. The Physics is informed during training by adding new components to the loss function, reflecting, for example, the residue of the PDE and its boundary conditions. By incorporating Physics knowledge into the loss function of a neural network, PINNs revolutionize the solutions of PDEs. PINNs, as in any DNN, face the challenge of choosing the best model for the prediction. Model selection refers to the process of choosing the best-performing model from a set of candidate models. This typically involves training and evaluating multiple models with different configurations. According to Brunton and Kutz “model selection is not simply about reducing error; rather, it is about producing a model that has a high degree of interpretability, generalization, and predictive capabilities”. Despite the use of AutoML, the responsibility of deciding whether the best-performing model was found, or another configuration is necessary rests with the data scientists. Model selection relies on the availability and analysis of relevant data to make well-informed decisions. However, existing tools designed to assist data scientists in model selection fall short of providing these relevant data, particularly with PINNs. Considering the lack of support for analytics and reproducibility of the trained models, in this paper we propose the adoption of provenance data. Provenance data relates all these data describing the origins and history of these steps until model selection. These provenance relationships allow for tracing the model back to its data preparation activities. W3C provides a standard schema to represent provenance data in a generic and uniform way. We discuss the role of provenance in steering PINNs using examples provided by DeepXDE, at a high-performance computing environment.