Machine learning-assisted rapid screening of oil types on microfluidic thread-based analytical devices (μTADs)

Rapid and low-cost classification of oil and its derivatives is critically important for on-site inspection, quality monitoring, and field analysis within the petroleum industry. Conventional analytical techniques, such as gas chromatography or mass spectrometry, typically depend on expensive, bulky instrumentation and involve complex, time-consuming laboratory procedures. Consequently, these methods are inherently ill-suited for real-time, on-the-spot analysis, creating a pressing need for portable, user-friendly, and cost-effective alternative technologies to enable immediate decision-making in the field.

To address this challenge, this work presents a novel oil classification system that combines capacitance data generated from the capillary flow on microfluidic thread-based analytical devices (μTADs) with machine learning techniques. To achieve efficient classification, linear discriminant analysis (LDA) was employed for dimensionality reduction and discriminative modeling of the original features. In practice, the system firstly classified eight typical petroleum oil samples, achieving an average accuracy of 82.00 % under six-fold cross-validation. Subsequently, seven non-petroleum oil samples were introduced, forming a fifteen-class classification task. The system maintained robust performance, achieving a six-fold average accuracy of 72.22 %, demonstrating strong capability in category assignment and classification robustness for oil samples with similar properties.

The feasibility of this low-cost, portable system paves the way for practical field applications, enabling rapid on-site oil analysis. This advancement significantly supports real-time quality monitoring, inspection, and immediate decision-making within the petroleum industry and regulatory processes.