Bulk video imaging based multivariate image analysis, process control chart and acoustic signal assisted nucleation detection

This article investigates monitoring technologies, which provide systematic solutions for nucleation detection based on external bulk video imaging (BVI). The methods under investigation rely on multivariate image analysis, image feature descriptors and statistical control charts (SPCs). For the design of SPCs the video information is transformed into time series. The application of SPCs may be hindered by autocorrelated time series, which show oscillatory patterns due to light reflections from the stirrer blades; however, the autocorrelation can be reduced by performing operations with the first principal component (PC1) of the captured color image or by stacking the frames based on the dominating frequency. Another option is to design digital signal filters in the frequency domain to decrease the autocorrelation of the time series.It was found that the fastest methods for nucleation onset detection were the monitoring in the principal score space and control chart based monitoring of the mean gray intensity of the PC1 images sampled at 25 Hz. Furthermore, it was observed that performing principal component analysis (PCA) calculations on multidimensional or multispectral information not only provides the combination of variables that explain most of the variance at a certain time instance but also decreases the autocorrelation of the resulting time series.For acoustic signal based monitoring the gray scale images were converted into a 2 channel stereo sound. It was found that this method has less performant nucleation onset detection capabilities compared to the methods which rely directly on the images.