Multi-Stage Corn-to-Syrup Process Monitoring and Yield Prediction Using Machine Learning and Statistical Methods

Area of Science:

• Food Science and Technology
• Chemical Engineering
• Data Science

Background:

• Corn syrup is a vital food industry sweetener, but maintaining consistent quality (dextrose equivalent - DE) is challenging in semi-automated batch production.
• Existing process control methods often focus on continuous systems, leaving a gap for small to medium-sized factories.
• Developing robust, data-driven models is crucial for enhancing efficiency and product consistency.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting key corn syrup quality parameters (feed pH and DE) in semi-automated settings.
• To identify critical process parameters influencing feed pH and DE using correlation analysis.
• To compare the performance of Artificial Neural Network (ANN), Support Vector Machine (SVM), and Linear Regression (LR) models.

Main Methods:

• Correlation coefficients were calculated to identify key process parameters affecting feed pH and DE.
• ANN, SVM, and LR models were constructed to predict feed pH and DE using historical process data.
• Model performance was evaluated based on prediction accuracy, noise tolerance, and robustness with limited data.

Main Results:

• Model accuracy ranged from 91% to 96%, with ANN models outperforming SVM and LR by 1-3%.
• ANN models demonstrated superior noise tolerance compared to SVM, while SVM performance degraded with high-dimensional data.
• Linear Regression models exhibited higher accuracy variation than ANN and SVM.

Conclusions:

• Machine learning, particularly ANN, provides effective process control for semi-automated corn syrup production, enhancing quality consistency.
• The developed models are accurate and robust, even with limited datasets, offering practical solutions for small to medium-sized factories.
• Multi-stage modeling approaches show potential for improved accuracy, though trade-offs exist compared to single-stage methods.