Development of an Artificial Neural Network Utilizing Particle Swarm Optimization for Modeling the Spray Drying of...

Area of Science:

• Food Science and Technology
• Computational Intelligence
• Chemical Engineering

Background:

• Spray drying is crucial for preserving coconut milk, extending its shelf-life.
• Developing accurate predictive models for spray drying processes is essential for quality control.
• Artificial Neural Networks (ANN) offer potential for modeling complex drying phenomena.

Purpose of the Study:

• To develop a Particle Swarm Optimization-enhanced Artificial Neural Network (PSO-ANN) model.
• To predict the coconut milk spray drying process using the developed PSO-ANN.
• To compare the performance of PSO-ANN with other models like GA-ANN and standard ANN.

Main Methods:

• Utilized a 2^k factorial design to select parameters for PSO tuning (number of particles, acceleration constants C1 and C2).
• Determined optimal PSO settings: global best C1 = 4.0, personal best C2 = 0, number of particles = 100.
• Evaluated model performance using Mean Squared Error (MSE) and conducted sensitivity analysis on input parameters.

Main Results:

• PSO-ANN achieved an MSE of 0.077, outperforming standard ANN (MSE = 0.082) but underperforming GA-ANN (MSE = 0.033).
• Sensitivity analysis revealed that inlet temperature significantly influences model performance across all tested models.
• Optimal PSO parameters were identified for enhancing ANN performance in spray drying prediction.

Conclusions:

• PSO-ANN is a viable tool for predicting coconut milk spray drying, showing improvement over standard ANN.
• GA-ANN demonstrated superior predictive accuracy compared to PSO-ANN for coconut milk powder quality.
• Inlet temperature is a critical factor affecting the accuracy of spray drying models for coconut milk.