An investigation of the performance of multi layer, neural networks applied to the analysis of PIV images

✍ Scribed by I. Grant; X. Pan

• Publisher: Springer
• Year: 1995
• Tongue: English
• Weight: 753 KB
• Volume: 19
• Category: Article
• ISSN: 0723-4864
• DOI: 10.1007/bf00189704

✦ Synopsis

The performance of a three layer neural net and a four layer neural net in analysing particle image velocimetry images, obtained in the particle tracking mode, was examined. The four layer net is shown able to resolve directional ambiguity. The efficiency of the neural nets were first considered using numerically simulated flows where an exact analysis success ratio (sr) could be obtained. The sr of the multi layer nets is compared with previous studies where a statistical windowing method was used. The present nets are shown to be particularly useful in flows which have systematic directional variation. The neural nets were also applied to PIV images obtained in aerodynamic and hydraulic vortex shedding studies. The nets are shown to provide a reliable interpretation of flows with high levels of rotation present.