(174ay) Ensemble-Based Machine Learning for Industrial Fermenter Classification and Foaming Control | AIChE

(174ay) Ensemble-Based Machine Learning for Industrial Fermenter Classification and Foaming Control

Authors 

Agarwal, A. - Presenter, VIRGINIA POLYTECHNIC INSTITUTE
Liu, Y. A., Virginia Polytechnic Institute and State University
McDowell, C., Novozymes
Abstract

In industrial fermentation, foaming remains an inevitable side effect of mixing, shearing, powder incorporation, and metabolic activities of microorganisms. Excessive foaming can interfere with mixing of reactants and lead to problems, such as decreasing effective reactor volume, microbial contamination, product loss, and increased reaction time. Physical modeling of foaming is an arduous process as it requires estimation of foam height, which is dynamic in nature and varies for different processes. This work demonstrates a novel application of ensemble-based machine learning methods for prediction of different fermenter types in a fermentation process (to allow for successful data integration) and of the onset of foaming. Ensemble-based methods are robust nonlinear modeling techniques which aggregate a set of learners to obtain better predictive performance than a single learner.

We apply two ensemble frameworks, Extreme Gradient Boosting (XGBoost) and Random Forest (RF), to build classification and regression models. Using real industrial fermentation data from a leading bioinnovation company, we first develop a classification model that predicts the fermenter type with 99.49% accuracy, thus enabling us to integrate all plant data from different fermenters. We then build classification and regression models, which predict the exhaust differential pressure (foaming indicator) and achieve 82.39% accuracy with an RMSE value of ±12mbarg, that is well within the tolerance for foaming prediction in industrial practice. These results demonstrate the usefulness of the ensemble-based machine-learning models in data integration and foaming prediction. Using these tools, we can orchestrate the addition of antifoam agents (AFA) or defoamers in an ad hoc manner to mitigate the adverse effects of excessive AFA addition.

This work differentiates itself from previous studies of big data analytics in industrial fermentation through the following contributions: (1) robust modeling of different fermenter types based on operating parameters, allowing for data integration with no prior information about the fermenter design or the type of microorganism used in the process; (2) accurate prediction of foaming based on exhaust differential pressure using both classification and regression models; and (3) usage of industrial multivariate fermenter data sets (64 batches with 4 different fermenter designs and over 183,000 instances).