Department of Biomedical Engineering, The University of Arizona, Tucson, AZ, 85721, United States.
Korea Institute of Ocean Science and Technology, Geoje-si, Gyeongsangnam-do, 53201, Republic of Korea.
Biosens Bioelectron. 2020 Jul 1;159:112193. doi: 10.1016/j.bios.2020.112193. Epub 2020 Apr 10.
Oil spills can be environmentally devastating and result in unintended economic and social consequences. An important element of the concerted effort to respond to spills includes the ability to rapidly classify and characterize oil spill samples, preferably on-site. An easy-to-use, handheld sensor is developed and demonstrated in this work, capable of classifying oil spills rapidly on-site. Our device uses the computational power and affordability of a Raspberry Pi microcontroller and a Pi camera, coupled with three ultraviolet light emitting diodes (UV-LEDs), a diffraction grating, and collimation slit, in order to collect a large data set of UV fluorescence fingerprints from various oil samples. Based on a 160-sample (in 5x replicates each with slightly varied dilutions) database this platform is able to classify oil samples into four broad categories: crude oil, heavy fuel oil, light fuel oil, and lubricating oil. The device uses principal component analysis (PCA) to reduce spectral dimensionality (1203 features) and support vector machine (SVM) for classification with 95% accuracy. The device is also able to predict some physiochemical properties, specifically saturate, aromatic, resin, and asphaltene percentages (SARA) based off linear relationships between different principal components (PCs) and the percentages of these residues. Sample preparation for our device is also straightforward and appropriate for field deployment, requiring little more than a Pasteur pipette and not being affected by dilution factors. These properties make our device a valuable field-deployable tool for oil sample analysis.
溢油事故可能对环境造成毁灭性影响，并导致意外的经济和社会后果。应对溢油事故的协同努力的一个重要组成部分包括能够快速对溢油样本进行分类和特征描述，最好是在现场进行。本工作中开发并展示了一种易于使用的手持式传感器，能够在现场快速对溢油进行分类。我们的设备使用 Raspberry Pi 微控制器和 Pi 相机的计算能力和可承受性，结合三个紫外线发光二极管 (UV-LED)、衍射光栅和准直狭缝，从各种油样中收集大量的紫外荧光指纹数据集。基于包含 160 个样本（每个样本重复 5 次，略有不同的稀释度）的数据库，该平台能够将油样分为四大类：原油、重燃料油、轻燃料油和润滑油。该设备使用主成分分析 (PCA) 来降低光谱维度（1203 个特征），并使用支持向量机 (SVM) 进行分类，准确率为 95%。该设备还能够预测一些理化性质，特别是基于不同主成分 (PC) 与这些残基百分比之间的线性关系，预测饱和物、芳烃、树脂和沥青质的百分比 (SARA)。我们的设备的样品制备也很简单，适合现场部署，只需要一个巴斯德吸管，并且不受稀释因素的影响。这些特性使我们的设备成为一种有价值的现场可部署的油样分析工具。
