A Study of Condensation and Frost Dynamics on Metallic and Silicon Substrates Evaluated Via a Computer Vision Model

Abstract

Water droplets condensing and freezing on surfaces are ubiquitous phenomena that often must be managed rather than eliminated. Condensation and frosting on engineered surfaces play an important role in the performance and safety of many systems, including refrigeration, aerospace components, environmental controls, heat exchangers, and electronics applications. When water vapor condenses and subsequently freezes on a surface, the resulting droplet growth, coalescence, ice nucleation, and frosting mechanisms are strongly influenced by the physical and thermal characteristics of the substrate. Surface geometry, wettability, droplet mobility, and droplet pinning can all alter how condensed water accumulates and freezes. Understanding how these variables interact and affect condensation and frost behavior is key to the development of effective ice-phobic surfaces. Two different types of surfaces, laser-patterned aluminum and silicon wafers coated with MoS₂, were the subject of investigation. Condensation and freezing experiments were performed on both surface types using the same experimental apparatus. For the metallic surfaces, experiments were conducted at ambient pressure, room temperature (i.e., 20 °C–23 °C), and at humidity levels of 39%–47% RH. The aluminum samples were cooled with a Peltier cooler set to -10°C. Two K-type OMEGA thermocouple probes (040G-6) were placed into precision, 1-mmdiameter holes machined in the edge of the sample to obtain a continuous, accurate surface temperature reading. The 1 mm holes were drilled to a depth of 10 mm. The average surface temperature was -8°C ± 0.5°C. Hole locations were spaced ±5 mm off the vertical centerline and centered on the horizontal centerline, 3.5 mm below the condensing surface. Condensation from moist air was observed until the full field of view froze. The experiment was recorded using high-magnification video microscopy. Videos were then reduced to 1 frame per second for processing by a purpose-built YOLOv11object detection model. The model extracted data on droplet population, average droplet diameter, droplet phase, and total surface coverage across the entire field of view. The YOLOv11-based architecture was trained on 1,300 1 1 Portions of this chapter were published in: Turner, Jane C., Aryan Singh Dalal, Hande Küçük McGinty, Melanie M. Derby, and Amy R. Betz. "Examining condensation and freezing behavior on a laser-patterned metallic surface using an AI neural network model." Frontiers in Thermal Engineering 6 (2026): 1808591. manually annotated images and evaluated using standard object detection metrics. Following training, the model achieved a mean average precision at an intersection-over-union threshold of 0.50 of 78%, with a precision of 75% and a recall of 76.5%. A bar of 6061 aircraft-grade aluminum with a mirror finish was used to create the substrate. 30 mm x 30 mm x 7mm sections were cut. A smooth portion of the unaltered bar stock was used as the control surface, and microtextured surfaces were created with a Pharos Laser and a 125 mm focal lens. The microtextured surface was generated using 15% of the total laser power. The laser created a pattern of parallel lines of a width of 100 ± 5 μm spaced 150 ± 5 μm apart. On the microtextured sample, the laser redeposited the material onto the surface, making a positive ridgeline. Tittling the stage of a scanning electron microscope at 70° to enable a clear side view of the microtextured sample. After imaging ten separate lines, sampled 5 mm apart, an average of 100 of the tallest peaks is found to be 6.8 µm ± 0.98 µm. The measured peak range is 3.3–9.5 µm. Freezing behavior is strongly affected by metallic surface texture [9]. The smooth aluminum control surface reached full frost coverage in an average time of 419.5 s, while the microtextured aluminum surface froze completely in an average time of 139.5 s at a freezing rate 3× times faster than the control. The faster freezing rate is attributed to the increased droplet pinning on the microtextured surface. Increased surface roughness leads to more pinning. This likely altered the dominant growth mechanisms, resulting in a smaller droplet size. It also provided more heterogeneous nucleation sites, improving heat transfer, and lowering the energy barrier to phase change. Freezing behavior is strongly affected by metallic surface texture. The smooth aluminum control surface reached full frost coverage in an average time of 419.5 s, while the microtextured aluminum surface froze completely in an average time of 139.5 s. A freezing rate 3x times faster than the control. The faster freezing rate is attributed to the increased droplet pinning on the microtextured surface. Increased surface roughness leads to more pinning. This likely altered the dominant growth mechanisms, resulting in a smaller droplet size. It also provided more heterogeneous nucleation sites, improving heat transfer and lowering the energy barrier to phase change. The second part of this work examined how molybdenum disulfide application methods influenced condensation and frost formation on silicon (Si)substrates. Two different methods were used to coat the Si substrate with MoS₂. One was via chemical vapor deposition (CVD) with coating, and the other by mechanical exfoliation (ME). An uncoated silicon oxide (SiOx) was used as a control surface. It is highly suspected that the Si wafers form an oxide later during the repeated cooling and thawing cycles. Contact angle measurements taken with a goniometer showed that the oxidized SiOx control had a before contact angle of 83.02° and a contact angle of 91.33°. While MoS₂-ME and MoS₂-CVD had average contact angles of 92.74° and 104.37°, respectively. Visually, the CVD coating produced a more uniform distribution of small individual hexagonal MoS₂ flakes, while the ME surface contained comparably massive regions of irregularly distributed flake regions. These differences in surface structure produced distinct droplet growth and mobility behaviors. The MoS₂ experiments were conducted under ambient conditions of approximately 19–21 °C and 45–51% RH. The Peltier cooler wasset to −5 °C. Six trials were conducted for each surface. Microscope videos were reduced to 1 frame per second and processed using the vision model for analysis. Although early nucleation and late-stage phase identification remained challenging, the model reliably captured droplet population and diameter once droplets reached approximately 10 µm in diameter. The model reports a single confidence value for all parameters examined at each instance. The model reached approximately 75% confidence at average times of 37 s for the control, 47 s for MoS₂-CVD, and 24 s for MoS₂-ME. The average maximum droplet populations were 1510 for the control, 1753 for MoS₂-CVD, and 1170 for MoS₂-ME. The average droplet diameters at initial freezing were 25.68 µm for the control, 39.92 µm for MoS₂-CVD, and 33.6 µm for MoS₂-ME. The average surface coverage values were 55.98% for the control, 64.40% for MoS₂-CVD, and 50.30% for MoS₂-ME. The control surface had the longest average freezing time at 621 s, followed by MoS₂-CVD at 425 s and MoS₂-ME at 242 s. Thus, MoS₂-ME froze ~ 2.5× faster than the control and ~ 1.7× faster than MoS₂-CVD. The power law values are calculated averages across each test series. The population exponents are as follows: pME α t -1.3 , pCVD α t -1.3, and pControl α t -0.91. However, this pattern is not inverted in the diameter's exponent order as would be expected and is: dCVD α t 0.73 , dControl α t 0.69 , and dME α t 0.63 . The MoS2-CVD surface produced the maximum average values for droplet population, average droplet diameter at initial freezing, and surface coverage before freezing. These results suggest that the more uniform CVD coating induced a Cassie state, thereby promoting droplet mobility and increasing coalescence events. The irregular distribution of mechanically exfoliated MoS2 flakes likely induced a Wenzel state, thereby increasing droplet pinning, the solid–liquid contact area, and the number of heterogeneous nucleation sites. The above results may suggest that the faster freezing rate of the ME was influenced by greater pinning and increased nucleation sites. Across both substrate sets, the results show that engineered surfaces can strongly alter condensation and freezing dynamics. On metallic aluminum, laser texturing increased droplet pinning and accelerated freezing relative to the control. On silicon substrates, different MoS2 application methods influenced droplet mobility and freezing time. Together, these studies further the collective understanding of the effects that surface topography, coatings, and surface wetting properties play in governing condensation growth regimes and frost dynamics.