Efficient numerical frameworks for modelling ultrasonic beams propagating across interfaces

This paper develops and compares two numerical frameworks—a refined Rayleigh-Sommerfeld Integral method and a high-frequency ray tracing approach—for modeling ultrasonic beam propagation across interfaces, demonstrating that the former is optimal for generating full field images while the latter is more efficient for evaluating fields through multiple interfaces at specific points.

The Gist

Imagine you are trying to predict how sound waves travel from a speaker (a transducer) through a wall and into a room. In the world of ultrasonic testing, this is like trying to see inside a machine part without touching it. The sound waves bounce off boundaries (interfaces) between different materials, like water hitting steel, and change direction or strength.

This paper is about building two different "maps" or computer programs to predict exactly where those sound waves go and how strong they are. The authors, researchers from Imperial College London, wanted to find out which map is faster and more accurate depending on the situation.

Here is a simple breakdown of their work:

The Two Competing Maps

The researchers developed two different ways to calculate the sound field:

1. The "Huygens' Crowd" Method (Rayleigh-Sommerfeld Integral)

• The Analogy: Imagine the boundary between two materials (like water and steel) is a crowded dance floor. Every single person on that floor is a tiny speaker. To know what the sound is like on the other side of the room, you have to listen to every single person on the dance floor, calculate their individual contribution, and add them all up.
• How it works: This method treats the interface as a collection of millions of tiny point sources. It uses a mathematical trick called Quasi-Monte Carlo (QMC) integration. Instead of checking every single spot on the dance floor in a rigid grid (which is slow), it picks random spots to sample, similar to how a pollster might ask random people in a crowd rather than asking everyone in a straight line.
• The Upgrade: The authors improved an existing version of this map. They realized that previous models treated these "tiny speakers" as if they shouted equally in all directions (like a lightbulb). They corrected this to show that these sources actually shout louder in one direction (like a flashlight), which makes the prediction much more accurate, especially near the boundary.

2. The "Laser Pointer" Method (Ray Tracing)

• The Analogy: Instead of listening to a crowd, imagine shooting a laser pointer. You aim a beam from the source, it hits the wall, bounces or bends according to the rules of physics (Snell's Law), and hits a specific spot. To find the sound at a specific point, you just trace the path of the "laser" there.
• How it works: This method assumes the sound waves are very high-frequency, behaving like straight lines (rays). It calculates the path a wave takes from the source, through the layers, to the destination.
• The Catch: To find the exact path, the computer has to solve a complex math puzzle (finding a "root") for every single point it wants to check. It's like solving a riddle every time you want to know where the laser lands.

The Showdown: When to Use Which?

The authors tested these two maps in three scenarios: sound hitting a wall at an angle, sound hitting a focused lens, and sound traveling through a "sandwich" of many thin layers.

Scenario A: You need a full picture of the sound field (e.g., a full image)

• Winner: The "Huygens' Crowd" (RSI) method.
• Why: If you need to know the sound level at thousands of points to draw a complete picture, the "Crowd" method is faster. It doesn't need to solve a riddle for every point; it just sums up the contributions. The "Laser" method gets bogged down because it has to solve a riddle for every single pixel in your image.

Scenario B: You have many layers (like a thin sandwich) and only care about a few points

• Winner: The "Laser Pointer" (Ray Tracing) method.
• Why: In the "Crowd" method, to get the sound to the final layer, you have to calculate the sound at every intermediate layer first. If you have 10 layers, you have to do the heavy lifting 10 times.
• The "Laser" method is like a direct flight. You can calculate the path to the final destination without stopping to check the weather at every layover. If you only need to know the sound at a few specific spots on the other side of a thick stack of materials, the "Laser" method is much faster and avoids errors that build up in the "Crowd" method.

The "Goldilocks" Conclusion

The paper concludes that there is no single "best" method; it depends on what you are trying to do:

• Use the "Crowd" (RSI) method if you want to generate a full, detailed image of the sound field and the material isn't too complex. It's great for getting a broad view.
• Use the "Laser" (Ray Tracing) method if you are dealing with many thin layers (like a multi-layered composite) and you only need to check a few specific points. It skips the middle steps and gets straight to the answer.

Why This Matters

The researchers showed that by using a smart sampling technique (Quasi-Monte Carlo), they could make these calculations much faster than traditional methods without losing accuracy. They also proved that their improved "Crowd" method is physically more correct than older versions, especially near the boundaries where sound waves enter new materials.

In short, they built two better tools for predicting how ultrasound travels, and they gave us a clear rulebook on which tool to grab for the job at hand.

Technical Summary

1. Problem Statement

Ultrasonic Non-Destructive Evaluation (NDE) requires accurate modeling of wave fields generated by transducers as they propagate through complex media, particularly across multiple interfaces (e.g., liquid-solid or solid-solid boundaries). Existing modeling approaches face a trade-off between accuracy and computational efficiency:

• Analytical methods are fast but lack flexibility for complex geometries.
• Semi-analytical methods (e.g., paraxial approximation) offer broader applicability but often sacrifice accuracy.
• Numerical methods (e.g., Finite Element Method - FEM) provide high accuracy but are computationally expensive, especially for 3D problems involving fluids and thin layers.

The specific challenge addressed in this paper is the efficient and accurate simulation of 3D ultrasonic wave fields interacting with multiple interfaces. The authors aim to determine which numerical framework is most suitable depending on the specific application constraints (e.g., number of field points vs. number of interfaces).

2. Methodology

The authors developed and compared two distinct numerical frameworks, both utilizing a Quasi-Monte Carlo (QMC) integration scheme (specifically the Halton sequence) to evaluate surface integrals. This approach replaces traditional meshing with pseudo-random sampling to accelerate computation while maintaining accuracy.

A. Rayleigh-Sommerfeld Integral (RSI) Model

• Foundation: Based on the superposition principle and diffraction theory. The interface is treated as a collection of secondary sources (dipoles) rather than simple monopoles.
• Improvements: The authors improved upon previous literature (Zhang et al.) by explicitly incorporating:
  1. Dipole behavior: Correcting the radiation profile of interface sources to account for directivity.
  2. Boundary Conditions: Ensuring continuity of pressure and normal stress across interfaces by using density ratios and plane wave reflection coefficients.
• Mechanism: It calculates the transmitted field by integrating over the transducer face and every interface. For $N$ interfaces, this requires nested surface integrals.
• Limitation: It is a "multiple surface integral" model; to find the field at a point beyond an interface, the field must be propagated through all preceding interfaces.

B. Ray Tracing Model

• Foundation: Based on high-frequency approximations (Kirchhoff approximation) and Fermat's Principle of Least Time.
• Mechanism:
  1. The incident field is expanded into an angular spectrum of plane waves.
  2. Each plane wave is treated as an acoustic ray.
  3. Snell's Law is satisfied using a variational approach (solving for a conserved quantity $C$ via root-finding algorithms like Ferrari's method for isotropic media).
  4. The field is calculated by summing contributions from rays traveling directly from the source to the observation point, applying transmission coefficients at interfaces without needing to propagate the field between every intermediate interface.
• Advantage: It avoids nested integrals; adding more interfaces only requires calculating the correct phase difference and transmission coefficient for the specific ray path.

3. Key Contributions

1. Dual Framework Development: The paper presents two robust frameworks (RSI and Ray Tracing) specifically optimized with QMC integration for ultrasonic beam modeling.
2. Theoretical Refinement: The RSI formulation was corrected to include dipole source behavior and strict adherence to boundary conditions, addressing inaccuracies found in prior works where pressure fields were discontinuous across interfaces.
3. Efficiency Analysis: A comprehensive comparative study was conducted to define the "sweet spot" for each method based on problem parameters (number of interfaces vs. number of field points).
4. Validation: The models were validated against:
   • Each other (showing high agreement in simple cases).
   • Finite Element Method (FEM) simulations near the interface (where analytical models often fail due to singularities).
   • Experimental parameters (using water-steel and multi-layered aluminum systems).

4. Results

The study evaluated three scenarios: oblique incidence, normal incidence with a focused transducer, and propagation through a multi-layered material.

• Single Interface / Full Field Imaging:

  • The RSI model is more efficient when calculating a large number of field points (e.g., generating a full 2D/3D image of the wave field).
  • It avoids the computational overhead of root-finding algorithms required by ray tracing for every point.
  • Both methods showed excellent agreement with FEM results near the interface, validating the corrected RSI formulation.

• Multi-Layered Media / Sparse Sampling:

  • The Ray Tracing model significantly outperforms RSI when dealing with multiple interfaces (e.g., thin-layered materials) or when only a few specific field points are needed.
  • In the RSI model, errors (random fluctuations due to QMC sampling) propagate and accumulate from one interface to the next, degrading accuracy in thin layers unless the sample size is drastically increased (which kills efficiency).
  • Ray tracing calculates the field at a point independently of intermediate layers, avoiding error accumulation and nested integrals.

• Computational Time:

  • RSI: Time scales linearly with the number of field points but is heavily penalized by the number of interfaces.
  • Ray Tracing: Time scales with the number of field points (due to root searching) but is largely independent of the number of interfaces.
  • Comparison: For a single point in a multi-layer system, Ray Tracing was orders of magnitude faster. For a full image in a single-layer system, RSI was faster.

5. Significance and Implications

• Application Guidance: The paper provides a clear decision matrix for researchers and engineers:
  • Use RSI for full-field imaging in systems with few interfaces.
  • Use Ray Tracing for inspection of multi-layered composites, thin films, or when only specific receiver points are required.
• Handling Complex Media: The Ray Tracing framework is highlighted as more adaptable for future extensions to anisotropic and inhomogeneous materials, as it only requires modulating slowness vectors, whereas RSI would require complex changes to Green's functions and additional nested integrals.
• Shear Waves: The Ray Tracing method is noted as naturally extending to shear wave propagation (post-critical angle), whereas RSI requires distinct Green's functions for longitudinal and shear modes.
• Hybrid Potential: The authors suggest that these QMC-based methods could be combined with FEM (using Ray Tracing to propagate to a scatterer and FEM for local scattering) to solve complex scattering problems efficiently.

In conclusion, the paper demonstrates that while QMC integration improves the efficiency of both methods, the choice between a wave-based (RSI) and a ray-based approach is critical and depends entirely on the specific geometry and output requirements of the ultrasonic inspection task.