Enabling Resolvent Analysis Through Assimilation of Experimental Mean Flows with Physics-Informed Neural Networks: A Case Study on the Boeing Gaussian Bump

We present a data-analysis workflow designed to study flow dynamics based on Particle Image Velocimetry mean fields. The approach utilizes mean flow data assimilation with Physics-informed Neural Networks (PINNs) to enable linearized mean field methods, specifically Resolvent Analysis (RA). As a case study, we apply the method to flows over the Boeing Gaussian Bump at Reynolds numbers of 1 and 2 million. For the PINN-based mean field assimilation, we propose a novel three-step validation strategy: comparison with the training data, validation against unseen data, and verification of the governing equations through analysis of their contributing terms. For the cases considered, the assimilated mean fields closely match the measurements and are found to be physically consistent. Based on these fields, we perform RA and compare the results with Spectral Proper Orthogonal Decomposition of surface pressure signals and instantaneous velocity measurements. This comparison confirms the model’s capability to capture the dominant dynamics of the separated flow. The overall approach presented here thus provides a modeling framework that enables the investigation of flow dynamics based on sparse mean field data.