Friction-Induced Heterogeneous Nucleation: Unravelling the Formation Mechanism of Brake Wear Particles via a Hybrid Machine Learning Framework and Multi-Dimensional Characterization

Vehicular non-exhaust emissions have become a dominant source of particulate pollution in urban areas. However, the dynamic physicochemical evolution transforming solid friction materials into aerosols under extreme braking conditions remains elusive, which significantly hinders accurate source apportionment. To bridge this gap, we established a comprehensive macro-to-micro analysis framework, integrating transient emission kinetics, multi-scale chemical fingerprinting, and machine learning techniques to decipher the formation mechanism of brake wear particles (BWPs).

By combining laboratory chassis dynamometer experiments (under WLTC and extreme AMS cycles) with high-resolution online monitoring (EEPS/APS), we captured the real-time formation dynamics. Crucially, at friction interface temperatures exceeding 300°C, we observed distinct "banana-shaped" particle size distribution evolution, directly indicative of rapid particle nucleation and subsequent growth events. However, connecting these macro-kinetics to micro-composition revealed a striking physicochemical paradox. Nanoscale single-particle elemental mapping indicated that ultrafine particles were predominantly composed of metallic elements (Fe/Cu) with negligible carbon signals. In sharp contrast, bulk surface spectroscopy (XPS/FTIR) of collected PM_2.5 samples revealed a composition overwhelmingly dominated by organic carbon functional groups derived from resin binders.

To reconcile this discrepancy, we pioneered a novel hybrid machine learning methodology. This approach uniquely couples Deep Residual Networks (ResNet) for texture extraction with XGBoost for geometric decision-making. This intelligent analysis allowed for high-throughput quantification of single-particle morphology, revealing that these burst-phase nanoparticles exhibit near-perfect sphericity, ruling out mechanical abrasion. Consequently, we propose a mechanism of metal-vaporization induced heterogeneous nucleation, whereby trace metallic components vaporize to form high-density condensation nuclei. These nuclei subsequently trigger the heterogeneous condensation and coating of semi-volatile organic vapors, thereby forming particles with a distinctive metal-core/organic-shell architecture. Our results redefine the braking process as an active high-temperature physicochemical reactor, providing a robust, data-driven foundation for understanding the complex formation mechanisms of non-exhaust emissions.