Quantifying the detachment dynamics of microplastic car tire-wear particles using a deep learning framework in a laboratory wind tunnel.

Traffic-related microscale particles, including passenger car tire-wear particles (PCTWPs), are recognised as one of the primary sources of microplastic pollution. While the chemical composition, shape characterisation, and emission rates of these non-exhaust traffic emission have received some attention, their detachment behaviour from surfaces into the air after deposition remains poorly understood. Their irregular, elongated shapes and relative orientation to the near-surface airflows pose significant challenges which influence aerodynamic performance. Moreover, the effect of their spatial deposition pattern at detachment needs to be studied as it may be relevant for potential particle-particle blockages and impacts. Owing to the multifactorial nature of drivers controlling the detachment process which may not lend itself to simple multi-linear correlation, we use a state-of-the-art deep learning (DL) based instance segmentation model, named You Only Look Once version 8 nano (YoloV8n) to detect, segment, characterise, and resolve particle detachment in high-resolution imagery from wind tunnel experiments. Three different PCTWP seeding approaches, namely tipping, sieving, and fabricated pressurised methods were evaluated and compared to identify the optimal method for uniform particle distribution with minimal agglomerates on substrates. The fabricated pressurised seeding method was selected as the optimal technique adopted for subsequent detachment experiments. PCTWPs were deposited onto glass surfaces and exposed to evolving, turbulent flow conditions. The model performance was evaluated using a variety of statistical quantities from the DL model including precision, recall, Intersection over Union (IoU) and dice coefficient metrics. The resolved detachment was analysed under two different transient conditions, subject to flow evolution, using PCTWPs with various size distributions (50µm - 220µm). For each condition, eight replicates were conducted to ensure statistical reliability. The analysis was performed as a function of time and friction velocity. Our results revealed that PCTWPs exhibit a median threshold fluid friction velocity at 0.46m/s for detachment, which is notably higher than that of polyethylene beads of 0.13 m/s for particles of the same size cohort. This highlights the significant role of local variations in the balance of forces and particle shape plays a crucial role in influencing detachment potential.