Histogram Matching, Hypothesis Testing, and Statistical Control-Chart-Assisted Nucleation Detection Using Bulk Video Imaging for Optimal Switching between Nucleation and Seed Conditioning Steps

Abstract

This article investigates the systematic nucleation detection performance of hypothesis testing, histogram matching, and control charts based on bulk video imaging. The control charts are built on the time series of the bin-by-bin distances between the reference and online acquired image histograms. In addition to the bin-by-bin histogram measures, the cross-bin earth mover’s distance measure has been evaluated. Hypothesis testing using the Kolmogorov−Smirnof (KS) statistics was also investigated as an additional method to detect nucleation. This work is relevant to the crystallization process control with in situ seed generation. Following the nucleation step in unseeded crystallization processes, it is common to keep the temperature constant or to increase it in order to allow the stabilization of the suspension, also called the digestion step. The nucleation detection methods discussed in this work allow the automated and robust switching between the nucleation and digestion steps. It is concluded that histogram matching can be successfully used for systematic nucleation detection, while the KS hypothesis-testing-based detection is strongly dependent on the histogram resolution. Furthermore, it is suggested that histogram matching is performed on the first principal components of color images in order to decrease the autocorrelation of histogram distance time series.