Three-dimensional density and air-rock interface reconstruction with muography: Application to the TianQin tunnel

This paper presents an optimized Metropolis-Hastings algorithm and an inverse distance weighting approach to improve 3D density reconstruction and air-rock interface mapping in muography, demonstrating significantly enhanced accuracy and artifact reduction through both Monte Carlo simulations and field data from the TianQin Tunnel experiment.

The Gist

The Big Picture: X-Raying the Earth with Invisible Arrows

Imagine you want to see what's inside a mountain, but you can't drill into it or cut it open. You need a way to "see" through the rock without touching it.

This paper describes a technique called Muography. Think of cosmic rays as a constant rain of invisible, super-fast arrows (called muons) falling from space. When these arrows hit the Earth, they punch through the atmosphere and into the ground.

• The Rule of Thumb: If the muons hit a thick, heavy wall of rock, many of them get stopped or slowed down. If they hit a hollow cave or a lighter patch of dirt, most of them sail right through.
• The Goal: By counting how many muons make it through from different angles, scientists can build a 3D map of what's inside the mountain. It's like figuring out the shape of a gift inside a box by seeing how much of a flashlight beam gets blocked.

The Problem: The "Blurry Photo" Effect

The researchers tried to use this method on a tunnel called the TianQin Tunnel. However, they ran into a common problem with these 3D maps: Smearing.

Imagine taking a photo of a sharp, clear statue, but your camera lens is dirty or out of focus. The edges of the statue look fuzzy, and the shadows stretch out into weird shapes. In the world of muography, when data is sparse (not enough muons are counted), the computer algorithms get confused. They try to guess where the rocks are, but they end up creating "ghost" shapes or blurring the edges of real caves and dense rocks.

The Solution: A Smarter Guessing Game

To fix this blur, the team developed a new computer algorithm called an Optimized Metropolis–Hastings (M-H) algorithm.

The Analogy:
Imagine you are trying to guess the layout of a dark room by throwing darts at a board.

• Old Methods (L-BFGS and SART): These are like a robot that throws darts in a straight line, calculates the average, and stops. It's fast, but if the room is complex, the robot might draw a blurry, messy map.
• The New Method (Optimized M-H): This is like a smart explorer. It starts with the robot's rough map, then takes small, random steps to test different possibilities.
  • If a new guess makes the map look sharper and fits the data better, it keeps it.
  • If a guess makes it worse, it usually rejects it, but sometimes it keeps it just in case it leads to a better spot later (this is the "Monte Carlo" part).
  • Over time, this explorer "wiggles" the map until the blurry edges snap into sharp, clear lines.

The Result: In their computer simulations, this new method turned a blurry 42% accurate detection of heavy rocks into a 100% accurate detection. It cleaned up the "ghosts" and made the boundaries of caves and rocks much crisper.

The Second Trick: Mapping the Ceiling

The paper also tackled a second problem: figuring out exactly where the rock meets the air (the ceiling of the tunnel).

Usually, you need to know the density of the rock to find the cave, or know the cave to find the rock density. The team used a clever math trick called Inverse Distance Weighting (IDW).

• The Analogy: Imagine you have a bunch of laser pointers shooting up from the tunnel floor. Each laser stops when it hits the ceiling. You don't know the exact height of the ceiling, but you have many laser tips hitting different spots. The IDW method acts like a smart averaging tool. It looks at all the laser tips in a small area and calculates the most likely height of the ceiling for that spot, weighting the closer lasers more heavily.

The Real-World Test: The TianQin Tunnel

The team took their new "smart explorer" algorithm and their custom-made detector (called MuGrid-v2, which is like a high-tech, 3D-printed muon camera) into the TianQin Tunnel.

1. The Setup: They placed the detector in three different spots inside the tunnel and waited for muons to rain down for a few weeks.
2. The Check: They compared their muon map of the tunnel ceiling against a LiDAR scan (a super-accurate laser map taken from the surface).
3. The Outcome:
   • The Ceiling Map: Their muon map matched the laser map very well (within about 5 meters of error). This proved their method works even without drilling.
   • The Density Map: They looked for hidden caves or strange pockets of heavy rock inside the mountain above the tunnel. They found nothing. The mountain above the tunnel is solid and uniform. This is actually good news for the tunnel's safety!

Summary

The paper shows that by using a smarter, "wiggling" computer algorithm, scientists can take blurry, fuzzy 3D X-rays of mountains and turn them into sharp, clear pictures. They proved this works by successfully mapping the ceiling of a real tunnel and confirming that the rock above it is solid and safe, with no hidden surprises.

Technical Summary

Technical Summary: Three-Dimensional Density and Air-Rock Interface Reconstruction with Muography

Problem Statement
Transmission muography is a non-invasive imaging technique used to reconstruct density distributions and air-rock interfaces by measuring the attenuation of cosmic-ray muons. However, the reconstruction process involves solving an ill-posed inverse problem. Widely used deterministic algorithms, such as damped least squares, Limited-memory Broyden–Fletcher–Goldfarb–Shanno (L-BFGS), and Simultaneous Algebraic Reconstruction Technique (SART), often produce severe "smearing" artifacts when measurement data are sparse or detector locations are limited. Furthermore, existing specialized methods (e.g., seed algorithms or back-projection) often require auxiliary prior information, such as manual seed points or gravity anomaly data, which may not be available in many field scenarios.

Methodology
The authors propose a two-pronged approach to address these limitations:

1. Optimized Metropolis–Hastings (M-H) Algorithm for Density Reconstruction:

   • The authors developed an optimized Markov Chain Monte Carlo (MCMC) algorithm based on the Metropolis–Hastings method.
   • Initialization: To avoid manual initialization, the algorithm uses results from deterministic methods (L-BFGS-B and SART) as initial samples.
   • Perturbation: A 3×3 Gaussian convolution kernel is applied to the current sample to generate a vector $G$. The iteration step size for each voxel is determined by this vector and a predefined bias rate ($b$).
   • Acceptance: New samples are generated by randomly perturbing the current density values within the calculated bounds. The acceptance of new samples is governed by a likelihood function comparing experimental opacity ($X_i$) with theoretical opacity ($X_{the}$), utilizing a system temperature parameter ($T_s$) and an acceptance rate ($B_p$).
   • Output: The final density distribution is the element-wise mean of the effective samples after discarding burn-in samples.

2. Inverse Distance Weighting (IDW) for Air-Rock Interface Reconstruction:

   • When the density distribution is known (or assumed uniform), the method solves for the path length matrix ($L$) to reconstruct the air-rock interface (elevation map).
   • The algorithm projects ray endpoints onto a 2D grid. For grid cells containing multiple endpoints, an Inverse Distance Weighting (IDW) method calculates the weighted average elevation, where weights are inversely proportional to the square of the distance from the cell center.

Key Contributions

• Artifact Mitigation: The optimized M-H algorithm effectively mitigates smearing artifacts common in sparse measurement scenarios without requiring auxiliary data or manual seed points.
• Robustness: The method demonstrates robustness regardless of the initialization algorithm used (L-BFGS-B or SART).
• Integrated Pipeline: The work integrates the reconstruction algorithms with the MuGrid-v2, an in-house developed muon scintillator detector featuring segmented light guides and silicon photomultipliers (SiPMs).
• Dual-Mode Reconstruction: The paper provides a unified framework for both 3D density reconstruction (solving for $\rho$) and interface reconstruction (solving for $L$) based on the same muon flux attenuation principles.

Results

• Monte Carlo Simulations:
  • In a simulated scenario with lead, air, and chalcopyrite anomalies, the optimized M-H algorithm significantly improved detection precision compared to baseline methods.
  • For high-density anomaly detection at a threshold of 5.1 g/cm³, precision improved from 42% (L-BFGS-B) to 100% (optimized M-H).
  • Across other thresholds and low-density scenarios, the M-H algorithm showed consistent precision gains ranging from 6% to 42%.
  • Visual inspection of reconstructed volumes showed sharper boundaries and the elimination of artifacts beneath density anomalies.
• TianQin Tunnel Field Experiment:
  • The MuGrid-v2 detector was deployed at three locations within the TianQin Tunnel to image the overlying rock structure.
  • Density Reconstruction: No significant density anomalies were detected above the tunnel, a conclusion supported by the absence of abnormal voxel clusters.
  • Interface Reconstruction: The reconstructed air-rock interface was compared against Light Detection and Ranging (LiDAR) measurements. The results showed an average error of 5.121 m and a relative error of 16.1%. The discrepancy was attributed to the assumption of uniform rock density, yet the reconstruction successfully captured the overall topographic features.

Significance and Claims
The paper claims that the proposed approach advances muography in two primary ways:

1. Algorithmic Improvement: It demonstrates that the optimized M-H algorithm significantly reduces artifacts near density anomalies, achieving up to a 58% precision improvement over baseline algorithms in high-density detection.
2. Methodological Expansion: It provides a complementary tool for geological exploration by enabling air-rock interface reconstruction when density distributions are known or assumed.

The authors conclude that the TianQin Tunnel experiment validates both the capabilities of the MuGrid-v2 detector and the expanded scope of muography applications. They modestly note current limitations, specifically that the hyperparameters of the M-H algorithm require manual tuning without well-defined selection criteria, and that performance in more complex geological scenarios requires further validation with larger datasets.