Automated identification and quantification of tire wear particles (TWP) in airborne dust: SEM/EDX single particle analysis coupled to a machine learning classifier
Automated identification and quantification of tire wear particles (TWP) in airborne dust: SEM/EDX single particle analysis coupled to a machine learning classifier
复制标题
DOI：
10.1016/j.scitotenv.2021.149832
复制
发表时间：
2021-09-09
影响因子：
9.8
通讯作者：
Yajan, Phattadon
中科院分区：
文献类型：
--
作者：
Rausch, Juanita;Jaramillo-Vogel, David;Yajan, Phattadon
近年来，由于包括道路交通废气排放在内的其他颗粒物的显著减少，包括轮胎磨损颗粒物（TWP）在内的非排气颗粒物在空气尘埃中，特别是在PM10中的份额有所增加。然而，由于示踪剂的非特异性以及它们通常包含在分析挑战性低浓度（例如Zn、苯乙烯、1，3-丁二烯、乙烯基环己烯）中的事实，TWP的定量是一项苛刻的任务。由于轮胎和路面之间的相互作用，TWP的化学和形态结构不均匀性加剧了这一困难。与批量技术相比，自动单颗粒SEM/EDX分析可以受益于环境TWP的普遍存在的异质性，作为其识别和定量的诊断标准。为此，我们遵循机器学习（ML）方法，该方法利用大量（67）形态，纹理（基于后向散射信号）和化学描述符将环境颗粒区分为以下类别：TWP，金属，矿物和生物/有机。我们提出了一个基于ML的模型，用于对空气样本进行分类（由超过100，000个环境颗粒（包括6841个TWP）进行训练），并在两个瑞士站点的一年监测活动中应用。在这项研究中，TWP的质量浓度在空气中的部分PM 80 -10，PM10-2.5和PM2.5-1进行了测定。此外，还对5621 TWP的粒度分布和形状特征进行了评价。通过FIB-SEM切割TWP证明，通常在TWP中发现的矿物和金属颗粒不仅存在于颗粒表面上，而且存在于整个TWP体积中。在城市背景站点，TWP和微型橡胶在PM10中的年均质量分数分别为1.8%（0.28 μ g/m3）和0.9%。在城市路边站点，相应的值高出6倍，TWP为10.5%（2.24 μ g/m（3）），微橡胶为5.0%。(C)2021作者由爱思唯尔公司出版
The share of non-exhaust particles, including tire wear particles (TWP), within the airborne dust and particularly within PM10 has increased in recent years due to a significant reduction of other particles including exhaust road traffic emissions. However, the quantification of TWP is a demanding task due to the non-specificity of tracers, and the fact that they are commonly contained in analytically challenging low concentrations (e.g. Zn, styrene, 1,3-butadiene, vinylcyclohexene). This difficulty is amplified by the chemical and morpho-textural heterogeneity of TWP resulting from the interaction between the tires and the road surface. In contrast to bulk techniques, automated single particle SEM/EDX analysis can benefit from the ubiquitous heterogeneity of environmental TWP as a diagnostic criterion for their identification and quantification. For this purpose, we follow a machine-learning (ML) approach that makes use of an extensive number (67) of morphological, textural (backscatter-signal based) and chemical descriptors to differentiate environmental particles into the following classes: TWP, metals, minerals and biogenic/organic. We present a ML-based model developed to classify airborne samples (trained by >100,000 environmental particles including 6841 TWP), and its application within a one-year monitoring campaign at two Swiss sites. In this study, the mass concentrations of TWP in the airborne fractions PM80-10, PM10-2.5 and PM2.5-1 were determined. Furthermore, the particle size distribution and shape characteristics of 5621 TWP were evaluated. A cut through a TWP by means of FIB-SEM evidences that the mineral and metal particles typically found in TWP are not only present on the particle surface but also throughout the complete TWP volume. At the urban background site, the annual average mass fraction of TWP and micro-rubber in PM10 was 1.8% (0.28 mu g/m(3)) and 0.9%, respectively. At the urban kerbside site, the corresponding values were 6 times higher amounting to 10.5% (2.24 mu g/m(3)) for TWP, and 5.0% for micro-rubber. (C) 2021 The Authors. Published by Elsevier B.V.