Statistical Optimization and Machine-Learning-Based Analysis of Palm Oil Pretreatment (3-90% FFA) for Enhanced Free...

Area of Science:

• Chemical Engineering
• Renewable Energy
• Process Optimization

Background:

• Palm oil is an efficient feedstock for biodiesel, but high free fatty acid (FFA) content hinders production.
• Esterification pretreatment converts FFA to fatty acid methyl ester (FAME), preventing byproducts.
• Optimizing pretreatment conditions is crucial for efficient biodiesel yield from palm oil.

Purpose of the Study:

• To optimize esterification pretreatment conditions for raw palm oil with high FFA content (3-90%).
• To develop mathematical models using Response Surface Method (RSM) and machine learning for predicting optimal conditions.
• To achieve a final FFA content below 1% for enhanced biodiesel production.

Main Methods:

• A Box-Behnken experimental design with four factors (reaction time, methanol:FFA ratio, catalyst, initial FFA) was employed.
• Three mathematical models were developed and validated across different FFA ranges.
• Machine learning algorithms (decision tree, random forest, gradient boosting) were used to refine optimal conditions.

Main Results:

• Optimal pretreatment conditions were determined for different initial FFA ranges (3-30%, 30-60%, 60-90%).
• The decision tree model achieved a high R-squared of 0.9762, indicating strong predictive accuracy.
• A gradient boosting model identified new optimal conditions yielding <1% final FFA across the 3-90% initial FFA range.

Conclusions:

• Optimized esterification pretreatment is effective in reducing high FFA content in palm oil for biodiesel.
• Machine learning approaches significantly enhance the prediction accuracy of optimal pretreatment conditions.
• The study provides a robust method for maximizing biodiesel production efficiency from challenging palm oil feedstocks.