TL;DR
This paper introduces a novel method, HSA-KS, to automatically detect and remove sudden "jumps" in maglev gyroscope (GAT) signals caused by external disturbances like wind or ground vibrations. This significantly improves the accuracy of north-seeking measurements, outperforming existing techniques.
What Problem Does This Paper Address?
Maglev gyroscope total stations (GATs) are crucial for precise directional measurements in challenging environments like tunnels and mines, especially where GPS is unreliable. However, their accuracy is compromised by instantaneous disturbance torques (e.g., from strong winds or ground vibrations). These disturbances cause sudden "jumps" in the GAT's rotor current signal, disrupting the north-seeking process and leading to biased results. While existing methods can handle general noise, they struggle to accurately identify and eliminate these specific, transient jump events, leaving a gap in robust GAT signal processing.
Methods
The core of this paper is the proposed HSA-KS method, a two-stage approach:
- Heuristic Segmentation Algorithm (HSA): This algorithm automatically identifies potential "change points" in the GAT signal by recursively dividing it into subsequences with different mean values. It's designed to handle non-stationary signals, which is characteristic of GAT data affected by disturbances.
- Two-Sample Kolmogorov-Smirnov (KS) Test: After HSA segments the signal, the KS test is used to distinguish between stationary and jumping subsequences. A "categorical stationary subsequence" (CSS) is identified as a benchmark (a stable subsequence with sufficient samples and low standard deviation). Then, each other subsequence is compared to the CSS using the KS test. If the distributions differ significantly (PKS ≤ 0.01), the subsequence is classified as a "jumping" subsequence and is excluded from further calculations.
The mean value of the remaining stationary subsequences is then used to calculate the processed gyro north azimuth.
Datasets
The study utilized data from a field experiment conducted at the 5th sub-tunnel of the Qinling water conveyance tunnel in Shaanxi Province, China. The experiment involved measuring north azimuths using a GAT and comparing them against high-precision GPS measurements. Ten sets of GAT signals, each containing approximately 20,000 data points and affected by instantaneous disturbance torques from passing vehicles, were collected.
Key Insights and Ideas
- Jumps are a distinct problem: The paper clearly differentiates the impact of instantaneous disturbance torques (causing "jumps") from general high-frequency noise, highlighting the need for specialized detection methods.
- HSA's strength in segmentation: The HSA algorithm proves effective at automatically breaking down complex, non-stationary GAT signals into more manageable segments, a crucial first step.
- KS test for classification: The KS test provides a statistically rigorous way to classify these segments as either stable (stationary) or disturbed (jumping), enabling targeted data cleaning.
- Stationary segments are key: The method's success hinges on identifying and averaging only the stable segments, effectively discarding the corrupted data.
- Improved accuracy is significant: The proposed method demonstrates a substantial improvement in north-seeking accuracy, outperforming established techniques.
Key Results
- The HSA-KS method successfully detected and eliminated jumping subsequences in GAT signals.
- For "Signal 1" (affected by disturbance and then recovering), the D-value (absolute difference between GAT and GPS north azimuths) decreased from 9.2" to 3.5".
- For "Signal 2" (affected by disturbance and not fully recovering within the sampling period), the D-value decreased from 3.5" to 1.4".
- Across 10 experimental datasets, the HSA-KS method achieved an average improvement (reduction in D-value) of 53.5%, significantly outperforming optimized HHT (28.2%) and optimized WT (which showed mixed results, sometimes amplifying jumps).
- The autocorrelograms of the processed signals using HSA-KS resembled those of steady GAT signals, indicating successful restoration of stationarity.
Limitations and Future Work
- Reduced sample size: While effective, the HSA-KS method reduces the total number of usable samples. However, the authors argue that the remaining samples are sufficient to maintain nominal north-seeking accuracy.
- Continuous disturbance: The paper focuses on instantaneous disturbance torques. The authors suggest future research should investigate how to address GAT signals affected by continuous disturbance torques (like persistent wind vibration) and develop corresponding compensation algorithms.
Summary updated