Improvement of Algorithm in the Particle Tracking Velocimetry Using Self-Organizing Maps

The neural network techniques are becoming a useful tool for the particle tracking algorithm of the PIV system software and among others, the self-organizing maps (SOM) model seems to have turned out particularly effective for this purpose. This is mainly because of the performance of the particle tracking itself, capacity of dealing with unpaired particles between two frames and no necessity for a priori knowledge on the flow field (e.g. maximum flow rate) to be measured. Initially, concept of SOM was applied to PIV by Labonte. It was modified by Ohmi and further modified algorithm is developed using the concept of Delta-Bar-Delta rule. It is a heuristic algorithm for modifying the learning rate as training progresses. Earlier, the treatment of unpaired particles, a specific problem to any type of PIV, is not fully considered and thereby, the tracking goes unsuccessfully for some particles. The present research is to bring about further improvement and practicability in this promising particle tracking algorithm. The computational complexity can be reduced employing modified algorithm compared to other algorithms. The modified algorithm is tested in the light of the synthetic PIV standard image as well as in particle images obtained from visualization experiments.Key words: Delta-Bar Delta, Dynamic Threshold Binarization, HVD Algorithm, Labonte's SOM, Modified Algorithm, Ohmi's SOM, Particle Image Velocimetry(PIV), Particle Tracking Velocimetry(PTV), Self-Organizing Map(SOM), Single Threshold Binarization.Journal of the Institute of Engineering, Vol. 7, No. 1,  2009, July, pp. 6-23doi: 10.3126/jie.v7i1.2057