126: Optimizing Lactose Crystallization Through Machine Learning and Process Analytical Technologies

Introduction

Lactose crystallization is a critical process in dairy industries that extract lactose from whey. Traditional process optimization routine relies heavily on manual trial and error interpretations from experienced crystallization experts. This study explores the integration of machine learning with process analytical technologies (PAT) to refine the crystallization process and achieve superior outcomes.

Methods

The experimental setup involved a controlled crystallization vessel equipped with real-time monitoring tools including a ReactIR system for continuous lactose concentration measurement and an online microscope (Blaze Metrics) to monitor crystal formation. The automated system (integrated in PharmaMV by Applied Materials) is initiated with Direct Chord Length (DCL) feedback control run, where the crystallization process parameters (cooling rate) is adjusted in real-time (according to the PAT data) to promote generation of large crystals. The data from DCL run is then used as the training data for the Adaptive Neuro-Fuzzy Inference System (ANFIS) model, which was then employed in subsequent ML iterative runs to refine the process parameters.

Results

The implementation of the ANFIS model led to enhanced process outcomes compared to published cooling profiles. Learning from the data obtained from DCL and consecutive ML runs, crystals produced from subsequent iterations are larger with shorter processing time and higher yield.

Significance

The integration of machine learning and PAT into the lactose crystallization process represents a significant advancement over traditional methods. By automating the adjustment of process parameters, this method reduces the need for extensive human intervention, paving the way for more sustainable and economically viable dairy processing operations.

Authors: Jong Cha Yong, Akshay Mittal, Felix Lee Jun Jie, Lee May Loo, Goh Yongkai, Qiaolin Yuan, Eunice Yeap Wan Qi, Srinivas Reddy Dubbaka, Harsha Nagesh Rao, Wong Shin Yee