Predicting Moisture Content in Microcrystalline Cellulose During Fluidized Bed Drying Using Machine Learning Techniques

Abstract

This research aims to develop a nonintrusive method for predicting moisture content in a fluidized bed dryer using machine learning techniques. Data were collected from experiments using microcrystalline cellulose, with sensors measuring temperature and air relative humidity at various points in the drying process. The data were preprocessed, normalized, and used to train several machine learning models, including ridge regression, support vector machines (SVR), and random forest regressors. The ridge regression model emerged as the most effective, achieving a prediction accuracy of 96.5%. The study employed k-fold cross-validation to ensure model robustness and avoid overfitting. The results demonstrate the feasibility of using machine learning for real-time moisture prediction, significantly enhancing the efficiency and accuracy of the drying process. This approach eliminates the need for process interruption for moisture content measurement, thereby improving operational efficiency and product quality.