Data-driven modelling and optimization of palm oil refining process

• Main Author: Sulaiman, Nurul Sulaiha
• Format: Thesis
• Language: English
• Published: 2021
• Subjects: TP Chemical technology
• Online Access: http://eprints.utm.my/id/eprint/102146/1/NurulSulaihaPSChE2021.pdf.pdf; http://eprints.utm.my/id/eprint/102146/; http://dms.library.utm.my:8080/vital/access/manager/Repository/vital:145703

Bleaching earth and citric acid are expensive chemicals utilized in the palm oil refining process. Both chemicals are important in removing impurities such as colour pigments and other breakdown products in crude palm oil. The purpose of this study was to successfully develop refined oil quality prediction model and to optimize the dosage of bleaching earth and citric acid utilized in a palm oil refining process. First, response surface methodology (RSM) was used to develop an empirical model to relate the removal rate of free fatty acid (FFA) and red colour pigments in the refined oil to the dosage of both bleaching earth and citric acid. Next, artificial neural networks (ANNs) models were developed based on the pilot-scale experimental data and actual palm oil refining plant data as input and output data for the models. The quality parameters of crude oil (FFA and colour pigment content) as well as process parameters (bleaching earth and citric acid dosage, pressure and temperature of the deodorizer) were the input data while the quality of refined oil (FFA and red colour pigments content) were the output data. For comparison purpose, three types of ANN models were developed, which were multi-layer perceptron (MLP), stacked neural network (SNN) models and radial basis function neural network (RBFNN) models. All the ANN models developed were able to predict the quality of the refined palm oil. The developed models were compared and the best model, the SNN model, was chosen as the model for refined palm oil quality prediction. Finally, optimization frameworks to optimize the palm oil refining process were developed based on the multi-objective genetic algorithms approach using RSM and ANN models. The RSM models were validated by comparing the predicted value with the actual pilot-scale data, while all ANN models were validated by a new set of actual palm oil refining plant data. From the findings, bleaching earth dosage was found to be the major contributor to the impurities removal rate affecting the palm oil refining performance. Optimizing the dosage of bleaching earth and citric acid enabled significant savings in raw material usage, thus creating a more sustainable and cost-effective palm oil refining process. The models developed from this study were able to predict the quality of the refined oil accurately and quickly, gave the optimum dosage of the bleaching earth and citric acid, as well as provided optimum point of pressure and temperature of the deodorization process.