Capillary flow velocity profile analysis on paper-based microfluidic chips for screening oil types using machine learning

We conceived a novel approach to screen oil types on a wax-printed paper-based microfluidic platform. Various oil samples spontaneously flowed through a micrometer-scale channel via capillary action while their compo-nents were filtered and partitioned. The resulting capillary flow velocity profile fluctuated during the flow, which was used to screen oil types. Raspberry Pi camera captured the video clips, and a custom Python code analyzed them to obtain the capillary flow velocity profiles. 106 velocity profiles (each with 125 frames for 5 s) were recorded from various oil samples to build a training database. Principal component analysis (PCA), support vector machine (SVM), and linear discriminant analysis (LDA) were used to classify the oil types into heavy-to-medium crude, light crude, marine fuel, lubricant, and diesel oils. The second-order polynomial SVM model with PCA as a pre-processing step showed the highest accuracy: 90% in classifying crude oils and 81% in classifying non-crude oils. The assay took less than 30 s from the sample to answer, with 5 s of the capillary action-driven flow. This simple and effective assay will allow rapid preliminary screening of oil types, enable early tracking, and reduce the number of suspect samples to be analyzed by laboratory fingerprinting analysis.

Automated summary

We developed a novel method to screen different types of oil using a wax-printed paper-based microfluidic platform. The oil samples flowed through micrometer-scale channels via capillary action, while their components were filtered and separated. The fluctuation in the capillary flow velocity profiles was utilized to classify oil types. A Raspberry Pi camera captured video clips, which were analyzed using a custom Python code to obtain the capillary flow velocity profiles. 106 velocity profiles were recorded from various oil samples, and principal component analysis, support vector machine, and linear discriminant analysis were used to classify the oil types with high accuracy.