Biodiesel classification by base stock type (vegetable oil) using near infrared spectroscopy data

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Chemical Engineering

Background:

• Biofuels, including bioethanol and biodiesel, are increasingly used.
• Near-infrared (NIR) spectroscopy offers a cost-effective and rapid method for quality control compared to traditional techniques.
• Real-time, on-line analysis is feasible with NIR spectroscopy.

Purpose of the Study:

• To establish a correlation between NIR spectra and biodiesel base stock (originating vegetable oil).
• To classify biodiesel samples into 10 distinct groups based on their feedstock.
• To evaluate the effectiveness of various machine learning techniques for this classification task.

Main Methods:

• Near-infrared (NIR) spectroscopy was employed to obtain spectral data from biodiesel samples.
• Principal Component Analysis (PCA) was used for outlier detection and dimensionality reduction.
• Four multivariate data analysis techniques were applied: Regularized Discriminant Analysis (RDA), Partial Least Squares-Discriminant Analysis (PLS-DA), K-nearest Neighbors (KNN), and Support Vector Machines (SVMs).

Main Results:

• Biodiesel classification based on feedstock type was successfully achieved using NIR spectroscopy and machine learning.
• K-nearest Neighbors (KNN) and Support Vector Machines (SVMs) demonstrated high effectiveness in classifying biodiesel by feedstock oil type.
• An SVM-based approach achieved a classification error (E) below 5%, while KNN offered a practical alternative with E=6.2%.

Conclusions:

• Modern machine learning techniques combined with NIR spectroscopy provide a robust solution for classifying biodiesel by its feedstock.
• SVMs offer superior accuracy, whereas KNN presents a computationally efficient option for industrial implementation.
• The methodology proves relatively simple and effective for biodiesel classification, distinguishing it from other fuels like gasoline and motor oil.