Microstructural analysis of freeze–thaw degradation in rubber-modified cement-stabilized crushed stone using X-ray computed tomography

Specimens were tested for unconfined compressive strength after 7 and 28 days of curing, following water immersion. A WAW-300 pressure testing machine was used in accordance with standard procedures outlined in JTG E51-2009 (Figure 1).

Figure 1

Flow chart of unconfined compressive strength test: (a) sample preparation, (b) sample curing, (c) sample soaking, and (d) unconfined compressive strength test.

After 28 days of curing and subsequent water immersion, specimens were subjected to up to 30 F–T cycles in a programmable high- and low-temperature chamber. Each 24-h cycle consisted of 12 h at −20°C followed by 12 h at +20°C. The durability was evaluated using the BDR, calculated as (1) BDR = R D / R C × 100 % , \text{BDR}={R}_{D}/{R}_{C}\times 100 \% , where R _D is the unconfined compressive strength after F–T cycles, and R _C is the initial 28-day strength.

For microstructural analysis, smaller cylindrical cores (Φ50 mm × 50 mm) were extracted from the center of the large, 28-day cured specimens, as shown in Figure 2. Care was taken to minimize damage during coring by using a low-speed, water-cooled diamond drill bit. To further mitigate any potential artifacts from the coring process, all subsequent image analyses were performed on a central region of interest within each sample, avoiding the outer edges.

Figure 2

Sample preparation for CT analysis: (a) core drilling from the larger cured specimen and (b) final cylindrical sample (Φ50 mm × 50 mm).

Scans were performed on these cored specimens before F–T cycling (0 cycles) and then again after 5 and 10 cycles using an industrial CT scanner with parameters detailed in Table 3. The raw CT data were reconstructed into a series of 8-bit grayscale image slices [19,20,21]. These images were processed using Avizo software to segment the distinct material phases.

To establish objective and repeatable segmentation criteria, a grayscale histogram analysis was performed on the reconstructed images. As shown in the representative histogram in Figure 3, the material exhibits a clear multi-modal distribution, with three distinct peaks corresponding to the primary phases. The first peak, at low grayscale values (approximately 25), corresponds to the pores (air voids). The second peak (approximately 65) represents the rubber powder, and the third, and largest, peak (approximately 150) corresponds to the dense cement/aggregate matrix. The thresholds for segmentation were determined by identifying the valleys (local minima) between these peaks, which represent the grayscale values least likely to be shared between phases. This analysis confirmed that the optimal threshold between pores and rubber powder was at a grayscale value of 45, and the threshold between the rubber powder and the matrix was at 79. These data-driven thresholds were then validated through visual inspection of the segmented slices to ensure accurate phase identification.

Figure 3

Representative grayscale histogram used for phase segmentation.

The final validated thresholds used for all analyses were as follows: pores (0–45), rubber powder (46–79), and cement/aggregate matrix (80–255). This segmentation process is visually represented in Figure 4. A known limitation of this approach is that the cement paste and aggregate were not segmented as separate phases.

Figure 4

Industrial CT gray image underlying database processing: (a) extracting pore from original image, (b) extracting rubber powder from original image, and (c) the original figure distinguishes the pores and rubber powder.

To quantify the regularity of pore network evolution under F–T cycling, the POEI was developed. For each specimen, the change in porosity was calculated on a slice-by-slice basis for two distinct intervals: the initial damage phase (ΔP ₀₋₅) and the progressive damage phase (ΔP ₀₋₁₀).

ΔP ₀₋₅ = (Porosity per slice after 5 cycles) – (Porosity per slice at 0 cycles).

ΔP ₀₋₁₀ = (Porosity per slice after 10 cycles) – (Porosity per slice after 5 cycles).

These two intervals were chosen strategically. The 0–5 cycle interval captures the initial, most severe phase of damage accumulation. The subsequent 5–10 cycle interval represents the evolution of damage from an already-degraded state. By comparing the spatial patterns of pore growth in these two phases, the POEI can effectively distinguish between the expansion of existing flaws and the creation of new, independent damage sites.

The POEI is defined as the coefficient of determination (R ²) from the linear regression between these two datasets (ΔP ₀₋₅and ΔP ₀₋₁₀) across the 500 slices analyzed for each specimen. The R ² value, ranging from 0 to 1, quantifies the proportion of variance in the progressive damage that is predictable from the initial damage. A high POEI value (close to 1) indicates a strong linear relationship, signifying that the spatial pattern of damage in the second interval correlates with the initial damage – an orderly expansion of existing flaws. Conversely, a low POEI value (close to 0) indicates a weak correlation and a disordered process where new, randomly located damage is created. This analysis was performed across a continuous series of 500 slices, representing a significant vertical portion of the specimen’s core, to ensure a statistically representative assessment of the internal damage distribution and its evolution.

The evolution of internal damage was quantified by tracking the change in porosity over the F–T cycles (Figure 5). The analysis reveals a distinct, two-stage damage progression that is highly dependent on the rubber content.

Figure 5

Change in porosity of specimens during F–T cycling. The black bars represent the porosity increase after 5 cycles (compared to 0 cycles), and the red bars represent the additional increase after 10 cycles (compared to 5 cycles).

Initial Damage Stage (0–5 F–T Cycles): In the first five F–T cycles, the increase in porosity was non-monotonic with respect to rubber content. The porosity gain was most severe for the C5R10 specimen, which aligns perfectly with its low BDR. This suggests that a small amount of rubber introduces interfacial weak points that promote extensive, disordered microcracking under initial F–T stress. As rubber content increased beyond this point, the initial porosity gain decreased significantly.

Damage Progression Stage (5–10 F–T Cycles): The analysis of damage from 5 to 10 cycles shows a different, more telling trend. The incremental increase in porosity during this period was significantly reduced with increasing rubber content. The high-content specimens (C5R30 and C5R45), which suffered less initial damage, also showed a remarkable ability to resist further degradation. Conversely, the low-content specimens continued to accumulate damage, albeit at a slower rate than in the initial stage.

This two-stage analysis provides a powerful insight: High rubber content not only mitigates the initial F–T damage but also fundamentally improves the material’s long-term stability by arresting the progression of damage in subsequent cycles.

Figure 6 shows the volume ratio of rubber powder extracted from 500 slices to the total sample volume. The results show that the volume ratio of each slice in the 500 slices is relatively stable, which is consistent with the theoretical value of the volume ratio, indicating that the distribution of rubber powder in the sample is very uniform.

Figure 6

Volume proportion of rubber powder in each slice.

To quantitatively characterize the regularity of pore expansion, the relationship between initial damage (porosity change from 0 to 5 F–T cycles) and progressive damage (porosity change from 5 to 10 F–T cycles) was analyzed for each of the 500 CT slices. The results are presented as a series of scatter plots in Figure 7.

Figure 7

Correlation analysis of progressive versus initial pore expansion under F–T cycling for specimens with varying rubber content. Each point represents a single CT slice. The POEI, defined as the coefficient of determination (R ²), is shown for each specimen, along with the corresponding p-value indicating statistical significance. (a) C5R0 (0% rubber), (b) C5R10 (10% rubber), (c) C5R20 (20% rubber), (d) C5R30 (30% rubber), and (e) C5R45 (45% rubber).

This visualization clearly illustrates the fundamental shift in the damage mechanism. For the reference specimen (C5R0), there is a moderate correlation between initial and progressive damage stages, yielding a coefficient of determination (R ², or POEI) of 0.41. At low rubber dosages (C5R10 and C5R20), the correlation weakens significantly, reaching a minimum POEI of 0.09 for the C5R20 specimen. This low value indicates a disordered and random damage process where new microcracks form independently of initial flaws.

In stark contrast, as the rubber content increases to higher levels (C5R30 and C5R45), a strong, statistically significant linear correlation re-emerges. The points cluster more tightly around the regression line, yielding high R ² values of 0.55 and 0.61, respectively (p < 0.001 for both). This demonstrates a shift back to an “ordered” damage progression, where damage accumulates primarily through the expansion of existing pores. This robust statistical and visual evidence confirms that the mode of damage evolution is fundamentally dependent on rubber content and validates the use of POEI as a key metric for frost resistance.

To establish the intrinsic link between macroscopic BDR and microstructural evolution, we performed a coupled analysis. Figure 8 combines the 28-day compressive strength and the BDR after 30 F–T cycles. The chart clearly illustrates the trade-off: While strength consistently decreases with added rubber, the BDR exhibits a non-monotonic U-shaped trend, with the lowest durability observed at 10–20% rubber content and the highest durability at 45%.

Figure 8

Comparative trend chart illustrating the trade-off between 28-day unconfined compressive strength and BDR after 30 F–T cycles.

This macroscopic behavior is directly explained by the microstructural damage mechanism. Figure 9 plots the POEI (R ²) against the BDR after 10 cycles. The two parameters exhibit a highly consistent trend, confirmed by a strong, statistically significant positive correlation (p < 0.05). This strong correlation confirms that the POEI is a powerful descriptor of BDR: A higher POEI signifies more ordered pore expansion and, consequently, superior material durability.

Figure 9

Correlation between the POEI (R ²) and the BDR (ten cycles) for materials with varying rubber content.

Based on these findings, a model correlating POEI (R ²) with rubber content was established (Figure 10). The model predicts a minimum POEI at a rubber content of 23.5%, aligning with the peak strength loss zone. Crucially, when rubber content exceeds 27.4%, the POEI of the modified specimens surpasses that of the traditional CSCS, demonstrating that a sufficient quantity of rubber fundamentally optimizes the pore expansion mode from disordered to ordered, thereby enhancing BDR.

Figure 10

Quadratic model of the POEI (R ²) as a function of rubber content.

To further investigate the physical mechanisms underlying the different R ² values, a detailed analysis of the CT scanning results was conducted from three dimensions: pore number, volume, and 3D morphology.

As shown in Figure 11, the evolution patterns of pore numbers differ significantly among specimens with varying rubber content. For low-content specimens with poor BDR (e.g., 10 and 20%), F–T cycles (particularly the first 5) lead to a continuous and significant increase in the number of small pores. This indicates that new, discrete micro-cracks are constantly generated within the material under frost-heaving pressure, which corresponds to the aforementioned disordered expansion mode (low R ²). In contrast, for high-content specimens with superior BDR (e.g., 45%), the number of small pores tends to stabilize or even decrease during the F–T process, accompanied by an increase in the number of medium pores. This reveals a more stable pore evolution mode where damage is primarily manifested as the gradual expansion and interconnection of existing small pores to form medium pores, rather than the nucleation of numerous new cracks. This is a clear manifestation of ordered pore expansion (high R ²).

Figure 11

Change of the proportion of each pore number under the action of the F–T cycle.

From the perspective of pore volume fraction (Figure 12), a decisive difference lies in the emergence of super-large pores. For this analysis, “super-large pores” (>1.00 × 10⁹ µm³ or 1 mm³) are defined as voids whose size indicates the transition from distributed micro-damage to localized macroscopic cracks, which govern structural failure. As shown in Figure 12b, F–T cycles induce the formation and continuous development of these super-large pores in traditional CSCS and low-content rubberized specimens. However, in the high-content (45%) specimens, no super-large pores were detected throughout the entire F–T process (Figure 12a). This provides strong evidence that the high content of rubber particles effectively absorbed the frost-heaving stress, suppressed catastrophic macroscopic crack propagation, and thereby maintained the structural integrity of the material.

Figure 12

Change of the pore volume ratio: (a) under F–T cycles and (b) proportion of ultra-large pore volume.

The 3D visualizations in Figure 13 intuitively illustrate the evolution of the pore network. While a full quantitative analysis of the network topology (e.g., using metrics like fractal dimension or connectivity index) remains a valuable direction for future work, these visualizations clearly reveal distinct differences in damage morphology. In the traditional CSCS specimens (Figure 13a and b), the pore network exhibits drastic expansion and increased connectivity between 5 and 10 F–T cycles, with smaller, isolated pores coalescing into larger, more tortuous crack networks. This signifies a widespread and interconnected damage path. In contrast, for the 30% rubber content specimen (Figure 13c and d), the changes are markedly more moderate and localized. The growth appears to be concentrated around existing larger pores, with less formation of new, independent cracks, visually confirming the “ordered” expansion quantified by the high POEI. For instance, a future quantitative analysis using metrics like fractal dimension would likely confirm a greater increase in network complexity and tortuosity for the control specimen, corresponding to widespread damage, compared to the more stable and localized pore growth observed in the C5R30 specimen.

Figure 13

Changes in microstructure of CSCS before and after F–T cycles: (a) Cement stabilized macadam material five times F–T and non-F–T 3D pore diagram comparison chart. (b) Comparison of 3D pore diagram of cement stabilized macadam material between ten F–T cycles and five F–T cycles. (c) The 3D pore diagram comparison diagram of cement stabilized macadam with 30% rubber powder content after five F–T cycles and no F–T cycles. (d) The 3D pore diagram comparison of ten F–T cycles and five F–T cycles of CSCS materials with rubber powder content of 30%.

The central finding of this study is the quantitative link between the spatial regularity of micro-damage and macroscopic durability. While previous studies, such as those by Jie et al. [9] and Yasser et al. [12], have documented the trade-off between compressive strength and durability in rubberized materials, they often lacked a microstructural explanation for the non-monotonic behavior of BDR [10]. Our work bridges this gap by demonstrating that the orderliness of damage evolution, not just the total increase in porosity, is the critical determinant of performance. The strong correlation between the POEI and BDR validates the POEI as a powerful analytical tool and provides a new perspective for material characterization.

The implications are twofold. From an engineering perspective, they provide clear guidance for designing durable, rubber-modified materials [18]. A critical threshold (in this study, ∼27.4% rubber content, where the POEI surpasses the control) must be exceeded to activate the beneficial network effect. Scientifically, the POEI offers a new paradigm for damage analysis that could be adapted for other forms of material degradation under cyclic loading.

Despite the clear findings, several limitations should be acknowledged. First, while mechanical tests were extended to 30 F–T cycles to confirm long-term performance trends, the CT analysis was concluded at 10 cycles due to the significant time and computational resources required for high-resolution scanning and reconstruction. Therefore, longer-term microstructural studies are needed to assess the ultimate service life of these materials. Second, this study used a specific blend of rubber particle sizes; the influence of rubber gradation warrants further investigation. Finally, future work should include chemical characterization of the rubber-cement ITZ to further elucidate damage mechanisms and incorporate quantitative 3D metrics like fractal dimension to analyze pore network evolution.