Prediction of free fatty acid in crude palm oil using near infrared spectroscopy

• Main Author: Abdull Rani, Siti Nurhidayah Naqiah
• Format: Thesis
• Language: English
• Published: 2015
• Subjects: TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://eprints.utm.my/id/eprint/53875/1/SitiNurhidayahNaqiahMFKE2015.pdf; http://eprints.utm.my/id/eprint/53875/; http://dms.library.utm.my:8080/vital/access/manager/Repository/vital:86547

Free Fatty Acid (FFA) value is widely used as an indicator for crude palm oil (CPO) quality. However, current methods used to measure FFA value are quite time consuming and complex. The application of near infrared (NIR) spectroscopy has drawn the interest to replace the conventional methods to measure FFA value as NIR has been shown to be effective in other food and agriculture industries. At the same time, improved predictive models have facilitated FFA estimation process in recent years. In this research, 176 CPO samples acquired from Felda Johor Bulker Sdn Bhd were investigated. A FOSS NIRSystem was used to take absorbance measurements from these samples. The wavelength range for the spectral measurement is taken at 1600nm to 1900nm. FFA content of each sample was determined by chemical titration method and three prediction models were developed relating FFA value to spectral measurement. The first prediction model based on Partial Least Square Regression (PLSR) yielded a regression coefficient (R) of 0.9808 and 0.9684 for the calibration and validation set respectively. The second prediction model built from Principal Component Regression yielded an R of 0.8454 and 0.8039 for the calibration and validation set respectively. The third prediction model built from Artificial Neural Network (ANN) yielded R of 0.9999 and 0.9888 for the calibration and validation set respectively. Results show that the NIR spectroscopy in a spectral region of 1600nm to 1900nm is suitable and adequate for FFA measurement of CPO and that the accuracy of prediction is high. Results shows that the prediction model using ANN gives the best prediction model of all three models tested.