Modell based, computer-aided Analysis of Data provided by the DPIV-Method
by Mario Gleirscher, September 2004
Part of this thesis deals with the investigation of the quality and efficiency of currently used particle image velocimetry (PIV) methods and how to enhance their input by means of digital image processing. That is, in order to appropriately use the corresponding approaches mainly in the growing field of Micro-PIV to gain better results. Another major part was to implement a flexible stand-alone software to carry out experiments for the evaluation of these methods.
Very early in 1904, one of the most popular fluid-mechanicians, Ludwig PRANDTL, had the idea of a tunnel containing circulating water, which he inspected by simply watching seeded particles. Today physical science and state-of-the-art techniques have lead this principle to a very sophisticated set of methods. With the help of high-resolution digital imaging and processing systems and via statistical methods, it became possible to gain vector field data out of exposures. In many use cases the current techniques make it feasible to get very accurate results at rather low cost. Increasingly higher usage of digital PIV (DPIV) methods with microscopic, turbulent, compressible flows like in Micro-PIV showed some problems when processing the low-quality images with non-standard particle image characteristics.
To achieve acceptable results for vorticity and deformation, furthermore, to analyse the topology of flows, various approaches, such as some methods for edge detection, out of the digital image processing discipline have been revisited. The features have to remain but the unwanted information have to be removed from the exposures. Observations of the signals in the spatial and frequency domains lead to further thoughts, which have to be investigated in detail during further derivative work. In addition, one of the tasks was to show the negative influence of noised images on various PIV methods.
2. Software, experiments and results
By now, the results of the test experiments pointed out, that it is important to optimize the analysis of the frequency information of the image by using appropriate wavelets for the respective image class. This must be done to obtain much better results, i. e. to reduce noise and to strengthen the power of the features wanted in the PIV analysis procedure.
One final conclusion was, that the previous knowledge of some global parameters such as the overall velocity distribution is fundamental to get initial estimators in order to start an iterative process leading into valid and highly accurate results. A number of methods will be discussed in the thesis. The first part of the overall results of this work is an evaluative overview of variously motivated approaches out of the digital image processing discipline. As a second part a validation of the new approaches has been made via the implementation of a specific testing software (called PIVA). The results of the tests, which were carried out on artifically generated images and exposures of biological structures, show the significance and the effect of the various system parameters on the analysis (see figures 1 and 2 below).
- M. Gleirscher, Modellbasierte, computergestützte Analyse von Daten aus dem DPIV-Verfahren, Dimplomarbeit 2004 [PDF]