Quantitative thermal flow visualization using image processing is very useful for obtaining velocity vector or temperature profiles in a thermal flow field. The velocity vector and temperature profiles are able to be simultaneously measured from color images visualized by a thermo-sensitive liquid crystal suspension method. The temperature profile, however, partly lacks information on temperatures in the field because of the narrow temperature range in which the liquid crystals present color. Temperature profiles over the entire flow field, therefore, are unable to be obtained.
This paper proposes a new algorithm for estimating unmeasurable temperatures from measurable temperatures. As the first step, an unsteady heat conduction field is modeled by using composite dynamic/static recurrent neural networks from the known information on temperatures. Consequently, the unknown information on temperatures is estimated from the neural network model.
