(58b) Multivariate Image Analysis, Histogram Matching, Process Control Chart and Acoustic Signal Assisted Nucleation Detection
AIChE Annual Meeting
2009
2009 Annual Meeting
Education
Student Poster Session: Computing and Process Control
Monday, November 9, 2009 - 8:30am to 11:00am
The aim of this work is to investigate several image processing and process monitoring technologies to provide systematic, sensitive, robust and low complexity-solutions for nucleation detection using bulk video imaging (BVI). The image analysis techniques are applied for the crystallization of caffeine as a model for active pharmaceutical ingredients, both externally and in-situ. The image frames are analyzed using feature index monitoring where signals for the mean gray intensity, the maximum intensities of the color channels, and the first principal components (PC1) images are calculated. The resulting time series trends are monitored using Shewhart and exponentially weighted moving average (EWMA) statistical process control charts (SPC) where nucleation is detected when the signals exceed the set control limits. A second approach to nucleation detection is histogram matching and computation of distance measures. Histograms of the images are generated and similarity measures are applied to detect a significant change indicating the beginning of nucleation; the current study evaluates the use of Kolmogorov-Smirnof and Χ2 statistics to determine the statistically significant change between the observed and the reference histograms for a 95% confidence interval. SPC charts were applied to the distance measures. The third approach relies on applying multiway principal component analysis (MPCA) on the color images to obtain the first score, which is also an image. The PCA model residual monitoring performance was also investigated. For acoustic signal based nucleation monitoring, the gray scale images were converted into a 2-channel stereo sound. It was found that this method has less performing nucleation detection capabilities compared to the methods which directly rely on the images. The results demonstrate that MPCA based monitoring is a suited technique for nucleation onset detection. The Statistical control chart assisted monitoring also successfully detect the nucleation onset. Furthermore, the histogram matching methods are not able to provide accurate monitoring due to significant process fluctuations when crystals are not present in solution.