Thermal Properties Reconstruction and Temperature Fields in Asphalt Pavements: Inverse Problem and Optimisation Algorithms

Zhonghai Jiang, Qian Wang, Liangbing Zhou, Chun XiaoFluid 2022 · DOI

TL;DR

This paper presents a novel approach to predicting temperature distributions in asphalt pavements by combining a high-order implicit difference scheme with a hybrid swarm intelligence algorithm. By treating thermal diffusivity as a temperature-dependent variable rather than a constant, the researchers achieved significantly higher prediction accuracy, reducing mean absolute deviation by approximately 0.3°C compared to traditional linear models.

What Problem Does This Paper Address?

Asphalt pavement performance is inextricably linked to temperature: high temperatures lead to rutting and instability, while extreme cold causes cracking. To design durable roads and plan maintenance, engineers need to accurately predict the "temperature field" within the pavement layers.

The challenge lies in the physics. Most existing models assume that the thermal properties of asphalt—specifically thermal diffusivity—are constants. In reality, these properties shift as the temperature changes. Calculating these shifting properties based on observed temperature data is known as an "inverse problem," which is notoriously difficult to solve because it is mathematically "ill-posed" (small errors in data can lead to massive errors in the result).

Methods

The authors tackle this problem using a two-pronged mathematical strategy:

  1. Numerical Simulation: They employ a two-layer implicit difference scheme (the Douglas format) to solve the one-dimensional nonlinear heat conduction equation. Unlike simpler "explicit" schemes that can become unstable if the time steps are too large, this implicit method is highly stable and achieves fourth-order accuracy in space.
  2. Hybrid Optimization (CS-GWO): To solve the inverse problem (finding the relationship between temperature and thermal diffusivity), they developed a hybrid algorithm combining:
  3. Cuckoo Search (CS): Excellent at "global search" and jumping out of local optima thanks to its use of "Levy flights" (random walks with occasional long jumps).
  4. Grey Wolf Optimizer (GWO): Highly efficient at "local search" and converging quickly on a solution by mimicking the social hierarchy and hunting behavior of wolves.

By blending these two, the CS-GWO algorithm balances exploration (finding the right neighborhood) and exploitation (pinpointing the exact answer) better than either algorithm could alone.

Datasets

The study validated its model using real-world experimental data:

  • Field Monitoring: Data was collected from a test asphalt pavement in September 2020.
  • Conditions: The researchers focused on high-temperature, no-rain weather cycles, collecting continuous 24-hour data.
  • Scope: They specifically analyzed data points where temperatures exceeded 55°C to test the model under high-stress thermal conditions.

Key Insights and Ideas

  • The "Aha" Moment in Optimization: The researchers found that while the Grey Wolf Optimizer is fast, it often gets stuck in "premature convergence" (finding a good-but-not-best answer). By introducing the "nest-abandoning" behavior of Cuckoos into the wolf pack's logic, the algorithm gained the ability to reset and find the true global optimum.
  • Dynamic vs. Static Physics: The study confirms that thermal diffusivity ($\alpha$) follows a linear relationship with temperature ($T$) in the form $\alpha(T) = AT + B$. Accounting for this slope ($A$) is the key to closing the gap between theoretical models and actual field measurements.
  • Computational Efficiency: Despite the complexity of the nonlinear equations, the implicit scheme required only four iterations per layer to achieve convergence, making it practical for real-time engineering applications.

Key Results

  • Algorithm Performance: In benchmark tests (using Sphere and Rastrigin functions), the hybrid CS-GWO algorithm outperformed standard CS and GWO by several orders of magnitude in terms of accuracy and convergence speed.
  • Prediction Accuracy: When applied to the pavement temperature field, the numerical solution using the optimized parameters was consistently closer to measured values than the "exact" linear solution.
  • Error Reduction: At a depth of 0.05m, the numerical model's Mean Absolute Deviation (MAD) was roughly 0.78°C to 0.85°C, whereas the standard model sat at 1.06°C to 1.14°C. This represents a meaningful improvement in precision for pavement engineering.

Limitations and Future Work

The authors acknowledge a few hurdles:

  • Boundary Conditions: There are still inherent errors in the upper boundary conditions (where the pavement meets the air), as solar radiation and wind speed create highly complex heat exchange patterns that are difficult to simplify.
  • Weather Variability: The current study focused on high-temperature, dry days. The authors suggest that future work should test the model under a wider variety of climatic conditions, including rainfall, snow, and extreme sub-zero temperatures, to ensure the CS-GWO algorithm remains robust across all seasons.

Summary updated

Abstract

A two-layer implicit difference scheme is employed in the present study to determine the temperature distribution in an asphalt pavement. The calculation of each layer only needs four iterations to achieve convergence. Furthermore, in order to improve the calculation accuracy a swarm intelligence optimization algorithm is also exploited to inversely analyze the laws by which the thermal physical parameters of the asphalt pavement materials change with temperature. Using the basic cuckoo and the gray wolf algorithms, an adaptive hybrid optimization algorithm is obtained and used to determine the relationship between the thermal diffusivity of two types of asphalt pavement materials and the temperature. As shown by the results, the prediction accuracy achievable with this approach is higher than that of the linear model.
BibTeX
@inproceedings{jiang2022thermal,
  author = {Zhonghai Jiang and Qian Wang and Liangbing Zhou and Chun Xiao},
  title = {{Thermal Properties Reconstruction and Temperature Fields in Asphalt Pavements: Inverse Problem and Optimisation Algorithms}},
  booktitle = {Fluid},
  year = {2022},
  doi = {10.32604/fdmp.2023.025270},
}
Figure