Multivariate Analysis for Quality Improvement of an Industrial Fermentation Process

The results of a successful implementation of multivariate analysis on multiple fermentations are shown in the paper. To minimize batch-to-batch variability of an industrial fermentation process, multi-way partial least squares (PLS) was used. Eighteen baseline batches from different fermentors were analyzed. After applying diagnostics tools and contribution charts, three batches were clearly identified to be abnormal, and even among the remaining batches , inconsistencies were found. The root causes of the variability were determined by a combination of variable importance in the projection plot and fermentation knowledge. A new fermentation procedure was applied and the quality improvement was demonstrated on 20 new batches. These batches were more consistent as evidenced by the improvement in the model fit (RX = 0.829 for the new batches versus RX = 0.681 for the baseline batches) and in the percent of out-ofspecification (3.3% for the new batches versus 20.4% for the baseline batches). The online multi-way PLS model was shown to detect bad batches promptly and to determine abnormal variables accurately. The success of this implementation demonstrates the value of applying multivariate analysis to large-scale industrial batch bioprocesses . Copyright © 2004 IFAC