Schlieren imaging visualizes the density gradients of a flow and is sensitive enough to capture the weak thermal gradients without disturbing the flow. However, the schlieren image is qualitative showing only the flow path not the temperature or velocity fields. Converting the schlieren image into a reliable and accurate quantitative field is the next step. Novel algorithms that use physics-informed neural network (PINN) are being developed to produce the fields using schlieren data while recovering thermodynamic quantities inaccessible to optical flow-based velocimetry alone. The goal of this project is to build a calibrated acquisition and pre-processing pipeline that turn raw high speed camera recordings into quantitatively trustworthy, low-noise inputs for a PINN to reconstruct temperature and velocity fields, with a stated uncertainty on every input. This is done by upgrading the pre-processing claim and adding a full bit depth, background and vignetting removal, and dual knife registration. Finally, the created pipeline will be validated against a reference flow of known temperature and speed. Tasks: - Literature research - Radiometric correction by implementing a no-flow background and a flat-field/vignetting correction - Spatial calibration to correct the magnification error relative to the mirror plane - Condition the image by denoising the faint schlieren signals and removing the silhouettes - Measurements for reference validation |
