| 03 Feb 2026
From seeding to detachment: leveraging deep learning to quantify the transport of tire wear microplastics in a wind tunnel
Abstract. The transport dynamics of tire wear particles (TWPs) remain poorly understood despite their growing contribution to airborne microplastic (MP) pollution. This study addresses this gap by experimentally quantifying the TWP detachment rate and threshold friction velocities (u*th ) from an idealised reference surface. Detachment experiments were conducted in a boundary layer wind tunnel over glass substrates seeded with a near-monolayer of particles. Time resolved imaging at 0.1 Hz was combined with automatic particle detachment and segmentation using an open source You Only Look Once version 8 nano (YoloV8n) model, which allowed individual detachment events and particle size and shape to be tracked with a mean average precision at an intersection-over-union threshold of 0.5 (mAP@50) above 85 % for both the bounding box and mask outputs. For the detachment experiments, pristine tire wear particles generated on a laboratory test stand with passenger car (PC) test tire were supplied by Continental GmbH, providing a well characterised and idealised TWP source. Among the three deposition method tested, the low-cost pressurised seeding approach produced the most uniform and reproducible particle distribution for detachment analysis. Across the analysed size range (80 to 300 μm), larger and more irregularly shaped particles exhibited significantly higher detachment (u*th) than smaller and more rounded fragments. Ensemble fits yield a bulk u*th of approximately 0.36 m s−1, with size and shape resolved u*th values varying by roughly a factor of 1.5 between the most easily detached and most resistant classes. The application of the Shao and Lu semi-empirical fluid threshold model reproduced the size-dependent u*th of smooth PE microsphere, but underestimates the TWP u*th unless the effective cohesion and/or aerodynamic scaling parameter are increased beyond values typically used for dust and sand. This behaviour is consistent with TWPs experiencing stronger effective adhesion than smooth, spherical grains of similar size, due to their irregular morphology and multiple contact points with the substrate. The density differences between TWPs (∼1300 kg m−3) and microspheres (∼1025 kg m−3) showed negligible influence within the studied size range (106 to 125 μm). We conclude that particle morphology, incorporating both size and shape, plays a dominant role in controlling the aerodynamic detachment of TWPs on the idealised glass substrate, while density effects are secondary under the tested conditions. Because controlled laboratory studies using well defined particles and simplified surfaces are a neccessary step towards isolating these fundamental mechanisms, our findings provide insights for improving MP and TWP resuspension models and highlight the need for future studies on more realistic environmental surfaces and broader particle sizes and density ranges.
The paper describes a careful experimental study of the suspension of tyre-wear particles from a glass substrate. The experiments were carried out in a small wind tunnel and the results used to investigate critical flow conditions for suspension as a function of particle size and shape. The paper is clearly written, though reading it is spoilt by the excessive use of abbreviations and acronyms, even the humble car gets reduced to PC. Most of this can be avoided by using plain language, pronouns, etc. I must admit that I found this annoying and distracting.
The work was carried out in a small wind tunnel, working section 170x54x27 (it might be 27x54) cm, at free stream speeds from 1 to 9.5 ms-1. The floor of the wind tunnel was covered with roughness elements, except for the glass slide on which the particles to be studied were deposited. Tyre-wear particles come in a variety of shapes and sizes and are, generally, non-spherical, and a key question is whether suspension from a smooth, glass surface provides any guidance for suspension from, say, road surfaces, where surface irregularities will play an important role – as will the action of vehicles moving over the road. Might not the latter be the dominant factor? If that is true, then what is the practical value of this work? These matters deserve some discussion, so that the results can be placed in context.
The boundary layer in the wind tunnel was probably quite shallow, perhaps about 20 cm – it would be helpful to provide the values in the paper along with shear stress profiles. There is no discussion of the mismatch in flow scales (compared with full scale) and its likely consequences, nor of particle inertia and Reynolds number effects. The friction velocity and the probability distribution of the wall shear stress are probably similar to those at full scale, but the associated time and length scales are not and that could be important. Saltation is not discussed – it probably plays no role in the experiments, but I don’t know. What might occur if the glass substrate were longer? Some discussion of these issues is essential.
A question: The term ‘detachment’ is used. I would use suspension – is there a reason for using detachment rather than suspension? How do you separate sliding from detachment?
Other comments on a page-by-page basis follow.
Line 56. What is meant by ‘collision’?
Lines 73-75. Perhaps: “However, for particles with non-spherical geometries, such as those from tyre wear, it is important to evaluate theoretical models of the critical friction by using experimental results for different particle sizes and shapes.”
Line 81. “... the deposited particles often settle in clusters ...”
Line 101. “... water, and were able to ...”
Line 112. Please make clear what is meant by: ‘instance segmentation’?
Line 116. Why is bold used here?
Line 132. Suggested wording: “Conventional image processing techniques have been widely adapted to detect particle images and analyse detachment under turbulent flow conditions in the wind tunnel.” Do you mean in general, or in your laboratory?
Section 2.1. The wind tunnel features a 54 cm × 27 cm cross-section and is 730 cm long. This is confusing as the cross-section dimensions are for the working section, which we learn later to be 170 cm long. It would be useful to add here the dimensions of the substrate and its location in the working section. At present, I have to wait for line 395 to learn: “The detachment experiments were conducted on standard laboratory glass slides (Thermo Fisher Scientific), measuring 75 mm × 25 mm.”
The wind tunnel flow exhausts to the atmosphere. How is the make-up flow provided that ultimately provides the inflow to the tunnel?
Use of a three-wire hot-wire probe requires a yaw and pitch calibration, either assumed or obtained from local calibrations. Assuming a wire responds to the flow component normal to it is generally considered inadequate. This is not discussed.
Line 156 “... second interval, lit by ...”
Figure 3. “Scanning electron microscopy image of passenger car tire wear particles.”
Line 213. “In particular, it adopts a CSP-style that replaces the C3 block with the C2f module and uses a spatial pyramid pooling fast (SPPF) block for efficient multi-scale feature aggregation.” Am I supposed to understand this sentence? There are other, similar examples.
Line 230. “Manual creation of polygons around microparticles often exceeds 50 particles mm−2 and is both time-consuming and sometimes inaccurate.” Do you mean that the process of manual creation of polygons around microparticles is both time-consuming and sometimes inaccurate because the density of particles often exceeds 50 mm-2?
Line 243 “Finally, the fine-tuned model was for a new set of data collected individually from the three seeding approaches.” I don’t understand this sentence.
Table 2. Some entries should have units. Have all the entries been defined? Are they all necessary?
Line 274. One example of the unnecessary use of an abbreviation: “The DSC is defined as:” – “It is defined as:” There are many others that could be resolved in a similar manner.
Line 277. “By computing the DSC for each individual TWP, we were able to obtain a detailed evaluation of the model’s ability to accurately delineate particle boundaries, quantifying the degree at which individual particles of different size and shape detach, and very efficient and sensitive to resolve TWPs capture on different surfaces.” Please rewrite and clarify the final clause, beginning 'and very’.
Equs. 5 and 6. These are properties of 2-dimensional shapes. ‘A’ is the plan surface area of the particle. Similarly, (6) applies to the surface area in plan. It is not a measure that demonstrates a spherical shape (lines 302, 305).
Line 306. Suggested wording: “This made the metric instrumental in determining how elongated or irregular the particles are, and how they exhibit different detachment behaviour due to variations in contact area and adhesion forces.”
Fig. 5. The lines joining the points have no meaning.
Fig 6. Dice or dice?
Fig. 7. Which is which, density or relative frequency?
Line 354. I don’t see the need of the references: Hancock, line 354; Foken et al., line 361; Zhou et al., 375; Fernholtz etc., lines 375-376. The first is about simulation of stable boundary layers, the second the well-established the log-law, the third and fourth effectively the same.
Equ. 7. How was ‘d’ determined?
Line 378. Measured at what height?
Line 385. How was u* adjusted? 2% uncertainty is remarkably low for hot-wire measurements.
Fig. 8. Suggested caption: “Figure 8. Calibration of u∗ from the profile measurement and the eddy-covariance as a function of the surface wind speed.” Add the value of z. As it stands, the axis has a label of the velocity profile, u(z).
Line 431. Would linear regression be preferable to ordinary least square?
Equ. 9. Define the symbol, N*.
Equ. 10. ‘a’ is an off-set, implying suspension at zero u*. That’s odd.
Line 436. Please define more fully: A (the maximum detachment fraction) and b (the rate of detachment change).
Line 442. C has no units.
Equ. 11. Is ‘d’ the diameter of a sphere?
Fig.9. No side forces are shown, likewise no saltation effects. Can there be direct lift-off? L acts near the leading edge and has along lever arm to the trailing edge, greater than that of the weight. Not obvious that a sphere is more likely to lift off, all else equal.
Section 3.1. Do you need to discuss sliding as well as lifting?
Fig 10b. What do the numbers in the boxes mean?
Lines 483-500. There’s a great deal of the obvious and well known here that could well be moved to introduction.
Line 486. What is meant by ‘acceleration’?
Line 489. How were the speed changes managed?
Fig 11. What is meant by ‘normalised’ in the caption?
Line 523. “For particles larger than 75 µm, gravity term dominates over cohesive forces and the uth∗ increases with size.” Provide a reference, or is this your inference?
Fig. 12. There are no error bars in (a) and (b), just two points in (c). Should the results in (c) be plotted against the mean diameter for each group?
Line 534. “Consequently, the drag force (Fd) and lift force (Fl) can more easily pivot a more spherical particle off the surface,” – not obvious really, see comments above on Fig. 9.
Fig. 13. Note u* scales change. Plot (d) against mean C for each group?
Fig. 14. Tyre-wear results for particles of equivalent diameter around 20 micro-m would be very useful.
Conclusions, Line 586-589. This is true. Note though that De and C are actually measures of the top or plan view of the particles, not of the whole, three-dimensional shape.
App. A1 is about Yolo. Is it needed? Please add brief explanations within A2, A3 and B1.
Fig. C1. Move to main text. Plot using dimensionless speed, U/Uref and (ln((z-d)/H), where H is the boundary layer depth, for the whole boundary layer and add the associated dimensionless shear stress profiles.
Fig. D1. How was acceleration from one u* to the next managed? What was the time-speed profile?