A Novel Method of Fault Detection and Identification in a Tightly Coupled, INS/GNSS-Integrated System

Fan Zhang, Ye Wang, Yanbin Gao2021 · DOI

TL;DR

This paper introduces a novel method for detecting and identifying faults in GNSS measurements within tightly coupled INS/GNSS integrated navigation systems. By employing a Variance Shift Outlier Model (VSOM), the system can adaptively down-weight or exclude faulty data, significantly improving the reliability and precision of navigation solutions, as demonstrated through static and dynamic field tests.

What Problem Does This Paper Address?

Accurate and reliable navigation is crucial for many applications, and integrated Inertial Navigation Systems (INS) and Global Navigation Satellite Systems (GNSS) offer superior performance by combining their complementary strengths. However, GNSS measurements are susceptible to faults (e.g., interference, multipath) that can severely degrade the navigation solution's precision and reliability if not properly handled. Existing methods like Fault Detection and Exclusion (FDE) often rely on complete removal of faulty data, which can be suboptimal. This paper addresses the need for a more nuanced approach to fault management in tightly coupled INS/GNSS systems, aiming to detect and identify faults in GNSS measurements and adapt the estimation process accordingly without necessarily discarding all affected data.

Methods

The core of this paper is the introduction of a Variance Shift Outlier Model (VSOM) for fault detection and identification within a tightly coupled INS/GNSS system that uses a closed-loop Extended Kalman Filter (EKF).

Here's a breakdown of the key methodological aspects:

  • Tightly Coupled INS/GNSS Architecture: The system fuses raw GNSS pseudo-range measurements directly with INS data within a single nonlinear filter (specifically mentioned as EKF, though other nonlinear filters are noted as possibilities). This allows for a more integrated estimation of INS errors.
  • Variance Shift Outlier Model (VSOM): This is the novel contribution. Instead of simply flagging a measurement as faulty, VSOM models an inflated variance for a potentially faulty measurement. This allows for a more gradual response: measurements with minor variance shifts are down-weighted, while those with significant shifts are effectively excluded.
  • Hypothesis Testing: To determine if a variance shift has occurred, the paper employs statistical hypothesis testing. Specifically, it uses:
  • Likelihood Ratio Test (LRT): Compares the likelihood of the data under a null hypothesis (no fault) versus an alternative hypothesis (variance shift).
  • Score Test: An asymptotically equivalent test to LRT, offering computational advantages as it doesn't require fitting the alternative model.
  • Parametric Bootstrap: To address the challenges of significance assessment and multiple testing (i.e., testing for faults in multiple measurements simultaneously), a parametric bootstrap procedure is utilized. This helps establish appropriate thresholds for the LRT and score test statistics.
  • Down-Weighting Mechanism: Once a fault is detected and its inflated variance quantified by the VSOM, the system implements a down-weighting process. This involves modifying the innovation covariance matrix using a weighting factor derived from the estimated variance shift. This effectively reduces the influence of the faulty measurement on the estimation process.
  • Horizontal Protection Level (HPL) Computation: The paper also details the calculation of HPL, a critical component for integrity monitoring, ensuring that the navigation solution's accuracy meets predefined safety standards.

Datasets

The paper describes field tests using real-world data:

  • Static Test: Data was collected for 21,600 seconds using a fiber-optic-gyroscope-based SINS and a Unicore UB370 GNSS receiver. The GNSS receiver tracked dual-constellation signals (GPS L1 and BDS B1).
  • Dynamic Test: A vehicular dynamic test was conducted in an urban environment using similar equipment (fiber-optic-gyroscope-based SINS and Unicore UB370 GNSS receiver tracking GPS L1 and BDS B1).

No new benchmark datasets were introduced; existing real-world data from controlled experiments was used for evaluation.

Key Insights and Ideas

  • Nuanced Fault Handling: The VSOM offers a more sophisticated approach than simple fault exclusion. By modeling variance inflation, it allows for adaptive down-weighting, which is crucial for maintaining continuity and precision when measurements are only partially degraded.
  • Statistical Rigor: The use of LRT and score tests, coupled with parametric bootstrapping for significance and multiple testing, provides a statistically sound framework for fault detection and identification. This moves beyond ad-hoc thresholding.
  • Integration Benefits: The tightly coupled architecture, where raw GNSS measurements are directly fused, is shown to be an effective platform for implementing such advanced fault detection mechanisms.
  • Practical Validation: The method's effectiveness is demonstrated through both static and dynamic field tests, showing tangible improvements in position accuracy and stability.

Key Results

  • Static Test: The VSOM method successfully detected and identified faulty GNSS measurements. The Root Mean Square Error (RMSE) of the position solution was reduced from 5.8459 m before fault detection and identification to 4.4656 m after. The paper highlights that abrupt jumps in position error (over 5 m) were eliminated.
  • Dynamic Test: In an urban dynamic scenario, the method demonstrated its ability to handle faults without disturbing the navigation trajectory. A fault occurring at the 350th second, which caused an abrupt disturbance in the trajectory before VSOM implementation, was effectively managed, resulting in a stable trajectory. The paper notes that when weight factors approached infinity, the corresponding measurements were nearly deleted.

Limitations and Future Work

  • Improvement Needed: The authors acknowledge that the fault detection and identification based on the VSOM still needs improvement.
  • Future Directions: They suggest exploring more applicable scene tests and developing more flexible test criteria to further refine the method. The paper also mentions that when weight factors approach infinity, measurements are "nearly deleted," implying that complete deletion might still occur in extreme cases, which could be an area for further investigation.

Summary updated

BibTeX
@misc{zhang2021novel,
  author = {Fan Zhang and Ye Wang and Yanbin Gao},
  title = {{A Novel Method of Fault Detection and Identification in a Tightly Coupled, INS/GNSS-Integrated System}},
  year = {2021},
  doi = {10.3390/s21092922},
}
Figure