TL;DR
This paper tackles the challenge of improving the accuracy of north-seeking gyroscope (GAT) data corrupted by vibration. They propose an optimized wavelet transform (WT) filtering method that dynamically determines the best decomposition level based on the signal's characteristics, significantly reducing errors in tunnel surveying.
What Problem Does This Paper Address?
North-seeking gyroscopes, like the magnetic levitation gyroscope (GAT), are crucial for precise orientation measurements, especially in environments where GPS is unavailable, such as long tunnels. However, these sensors are highly susceptible to vibration from external sources like construction, blasting, and passing vehicles. This vibration introduces non-stationary noise into the gyro signals, distorting orientation results and reducing accuracy. While wavelet transform (WT) is a known technique for signal denoising, its effectiveness for GAT data is limited because determining the optimal decomposition level for filtering is difficult and often based on empirical values that don't account for varying environmental conditions and signal types. This paper aims to address this by developing an optimized WT filter that adapts to different signal characteristics caused by vibration.
Methods
The core of this paper is the development and optimization of a wavelet transform (WT) filter for GAT data. Here's a breakdown of their approach:
- Wavelet Transform (WT) Basics: They utilize the continuous wavelet transform (CWT) and the Mallat algorithm for signal decomposition. The Symlets wavelet function (specifically Sym10) is chosen for its suitability in signal processing and its ability to reduce phase distortion.
- Decomposition and Denoising: The WT decomposes the signal into low-frequency and high-frequency coefficients. The high-frequency coefficients are then processed using a soft thresholding function and the Rigrsure adaptive threshold setting method to remove noise.
- Optimization of Decomposed Level: This is the key innovation. Instead of using a fixed decomposition level, they propose a method to find the optimal level for different signal types. This involves:
- Signal Classification: Observed GAT signals are categorized into four types based on their waveform: steady, periodic, jitter, and jumping.
- Evaluation Metrics: Two primary metrics are used to evaluate the filtering performance at different decomposition levels:
- H (Composite Index): A normalized composite index combining RMSE, SNR, correlation coefficient (ρ), and smoothness (r) to assess the overall de-noising quality. A higher H indicates better de-noising.
- DF (Absolute Difference): The absolute difference between the filtered gyro grid bearings and the high-precision GNSS grid bearings, representing the corrective effect on the north-seeking result. A lower DF indicates better correction.
- Optimal Level Determination: The optimal decomposed level is determined by finding the level that maximizes H and minimizes DF for each signal type.
- Time-Frequency Analysis: 3D time-frequency spectra are used to visualize how the signal's characteristics change with different decomposition levels, aiding in understanding the optimization process.
- Feasibility Verification: The optimized filtering scheme is compared against unfiltered data and data filtered with a fixed, commonly used decomposition level (level 5) using two indicators:
- DT (Difference in Orientation): The difference between filtered and unfiltered gyro grid bearings.
- LBE (Lateral Breakthrough Error): A practical indicator of the filter's effectiveness in tunnel surveying, measuring the lateral error at a breakthrough point.
Datasets
The study uses real-world observed gyro signals collected from an experimental network designed to mimic tunnel conditions. This network includes:
- Gyro Calibration Line: Used for initial gyro calibration in a stable environment.
- GNSS Reference Baseline: A high-precision GNSS baseline established in the experimental area, serving as a ground truth for evaluating the accuracy of the filtered gyro data. This baseline has an orientation accuracy better than 0.5".
- Five Gyro (Traverse) Lines: These lines were used for collecting observed gyro signals under various environmental conditions, including wind-induced vibration and ground vibration caused by passing vehicles.
The data was collected under conditions simulating those encountered in tunnel surveying, specifically referencing the Hong Kong-Zhuhai-Macao Bridge project.
Key Insights and Ideas
- Signal Type Matters: The optimal wavelet decomposition level for denoising is not universal; it's highly dependent on the specific characteristics of the observed gyro signal, which are influenced by the type of environmental disturbance.
- Balancing De-noising and Correction: While increasing the decomposition level generally improves de-noising (higher H), excessively high levels can lead to instability in the corrected orientation (higher DF) and potentially distort the signal. The goal is to find a sweet spot that optimizes both.
- Empirical vs. Optimized Levels: The study demonstrates that fixed empirical decomposition levels (like level 5 or 10) are less effective than an optimized level determined by the signal's characteristics.
- Practical Relevance for Tunneling: The LBE metric clearly shows the tangible benefit of the optimized filtering method in a real-world engineering application like tunnel breakthrough, significantly reducing errors.
- Beyond Noise Reduction: The method not only removes random noise but also helps correct systematic errors or trends introduced by environmental factors, leading to a more reliable north-seeking result.
Key Results
- Significant Error Reduction: The optimized wavelet filtering method achieved a substantial reduction in the lateral breakthrough error (LBE) by 68.7%, decreasing it from 12.1 mm to 3.8 mm compared to unfiltered data.
- Improved Accuracy: The optimized filtering scheme showed a statistically significant improvement in orientation accuracy (DT) compared to both unfiltered data and data filtered with a fixed decomposition level of 5 (p < 0.01).
- Optimal Levels Identified: Specific optimal decomposition levels were identified for different signal types:
- Steady signals: Level 6
- Periodic signals: Level 12 (though level 8 was deemed a safer choice)
- Jitter signals: Level 9 (with a warning about excessive levels)
- Jumping signals: Level 10 (with a warning about excessive levels)
- Generalizable Approach: The method for determining the optimal decomposed level is presented as applicable not only to GAT sensors but potentially to other types of gyroscope sensors as well.
Limitations and Future Work
- Signal Classification: The study classified observed signals into four typical categories. The authors acknowledge that a more comprehensive classification of signals under various environmental impacts might be needed, requiring more sophisticated experiments.
- Limited Data: The length of the article and the experimental data available limited the scope of the signal classification.
- Future Thresholding Methods: For further improvement in de-noising, the authors plan to explore different thresholding methods for high-frequency coefficients at various decomposed levels.
- Alternative Filtering Methods: They also intend to investigate and compare other filtering methods suitable for non-stationary GAT signals, such as the Hilbert-Huang transform.
- Conservative Approach: For practical applications where environmental factors are unknown, the authors suggest a conservative approach of selecting decomposition levels between 6 and 10 to ensure potential improvements without risking excessive filtering.
Summary updated