Feature extraction of time-series process images in an aerated agitation vessel using self organizing map

The batch self organizing map (SOM) is applied to extracting the feature of process images for the dynamic behavior of an aerated agitation vessel. When time-series images preprocessed by particle image velocimetry are computed by the SOM, the generated map provides visible and intelligible information for periodic behavior of patterns for gas dispersion. It is also shown that the sigmoid transformation of data enhances the efficiency of generating a more comprehensible map. Furthermore, the SOM is demonstrated to be effective in extracting the feature of small displacements of the impeller shaft inside the vessel.