Machine Learning-Based Classification of Oils Using Dielectric Properties and Microwave Resonant Sensing

TL;DR

This study introduces a machine learning approach utilizing microwave resonant sensors to accurately classify different oils based on their dielectric properties, enabling real-time, non-destructive industrial applications.

Contribution

It presents a novel combination of dielectric property measurement with machine learning classifiers for high-accuracy oil classification.

Findings

Achieved 99.41% accuracy with random forest classifier.

Demonstrated the method's suitability for real-time industrial use.

Validated the approach's effectiveness across various oil samples.

Abstract

This paper proposes a machine learning-based methodology for the classification of various oil samples based on their dielectric properties, utilizing a microwave resonant sensor. The dielectric behaviour of oils, governed by their molecular composition, induces distinct shifts in the sensor's resonant frequency and amplitude response. These variations are systematically captured and processed to extract salient features, which serve as inputs for multiple machine learning classifiers. The microwave resonant sensor operates in a non-destructive, low-power manner, making it particularly well-suited for real-time industrial applications. A comprehensive dataset is developed by varying the permittivity of oil samples and acquiring the corresponding sensor responses. Several classifiers are trained and evaluated using the extracted resonant features to assess their capability in distinguishing between oil types. Experimental results demonstrate that the proposed approach achieves a high classification accuracy of 99.41% with the random forest classifier, highlighting its strong potential for automated oil identification. The system's compact form factor, efficiency, and high performance underscore its viability for fast and reliable oil characterization in industrial environments.