Polar Transversal Initial Alignment Algorithm for UUV with a Large Misalignment Angle

Zheping Yan, Lu Wang, Tongda Wang, Honghan Zhang, Zewen YangSensors 2018 · DOI

TL;DR

This paper introduces a novel "polar transversal initial alignment algorithm" for Unmanned Underwater Vehicles (UUVs) operating in polar regions. It addresses the limitations of conventional alignment methods in high-latitude areas by employing a transformed coordinate system and a simplified Unscented Kalman Filter (UKF) to handle large initial misalignment angles, crucial for accurate UUV navigation.

What Problem Does This Paper Address?

Accurate initial alignment is critical for Strapdown Inertial Navigation Systems (SINS) used in UUVs, as it provides the initial conditions for navigation. Conventional alignment algorithms, typically designed for non-polar regions, struggle in polar environments due to the rapid convergence of Earth's meridians near the poles. This convergence leads to calculation overflows and error amplification, rendering these algorithms invalid. Furthermore, UUVs often experience large initial misalignment angles, which standard linear models and Kalman filters cannot effectively handle. This paper tackles the challenge of achieving accurate initial alignment for UUVs in polar regions, specifically considering the complexities of high latitudes and significant initial orientation errors.

Methods

The core innovation of this paper is the introduction of a polar transversal coordinate system. This system is derived by rotating the conventional latitude and longitude coordinate system by 90° around the y-axis of the Earth-Centered Earth-Fixed (ECEF) frame. This transformation effectively maps the polar region to an equatorial-like region in the new coordinate system, mitigating the issues caused by meridian convergence.

For the actual alignment process, the paper proposes a polar transversal initial alignment algorithm. This algorithm is designed to handle large misalignment angles, which necessitate a nonlinear dynamics model. To fuse the data from the inertial measurement unit (IMU) within this nonlinear framework, a simplified Unscented Kalman Filter (UKF) is employed. The simplification arises from the assumption of additive observation noise, avoiding the need for state augmentation typically required in standard UKF implementations.

The paper also details the derivation of attitude and velocity error equations within this transversal frame and establishes the dynamics and observation models for the UKF. Observability analysis is performed using Singular Value Decomposition (SVD) to confirm the system's ability to estimate the states.

Datasets

The paper utilizes simulated data for initial testing and a semi-physical simulation for experimental validation. The semi-physical simulation uses experimental data obtained from a rectangular pool experiment with a White Dolphin-100 UUV. This experimental data includes measured gyro drifts and accelerometer biases.

Key Insights and Ideas

  • Coordinate System Transformation is Key: The fundamental insight is that by transforming the coordinate system, the unique challenges of polar navigation (meridian convergence) can be circumvented, making established alignment principles applicable.
  • Large Misalignment Angles Require Nonlinearity: The paper emphasizes that large initial misalignment angles necessitate nonlinear dynamics models, moving beyond the limitations of linear Kalman filtering.
  • Simplified UKF for Efficiency: The use of a simplified UKF is a practical approach to data fusion in this context, offering a balance between accuracy and computational complexity, especially when observation noise is additive.
  • Polar Alignment is Fundamentally Different: The research highlights that polar alignment isn't just a variation of existing methods but requires a distinct approach due to the geometric properties of the Earth at high latitudes.

Key Results

The proposed polar transversal initial alignment algorithm demonstrates superior performance compared to a conventional initial alignment algorithm in simulations and a semi-physical experiment.

  • Simulation Results:
    • The algorithm achieved significantly smaller attitude errors after initial alignment. For example, $\phi_x$, $\phi_y$, and $\phi_z$ errors were 92.04', -121.4', and 485' respectively, compared to 442.6', 857.7', and 1769' for the conventional method.
  • Attitude errors converged to near zero much faster and remained stable.
  • Simulations with varying misalignment angles showed that smaller angles led to faster convergence and better accuracy.
  • Experiment Results:
    • The experimental results mirrored the simulation findings, with Algorithm 1 (proposed) yielding attitude errors of 208.4', -152.9', and 448.8' for $\phi_x$, $\phi_y$, and $\phi_z$, respectively, compared to 442.7', 858', and 1773' for Algorithm 2 (conventional).
  • The proposed algorithm effectively achieved initial alignment in the polar region, especially under large misalignment conditions.

Limitations and Future Work

  • Limitations:
    • The paper acknowledges that smaller misalignment angles lead to faster alignment speeds and better accuracy, implying that very large angles might still pose challenges, though the algorithm is designed to handle them.
  • The experimental validation was performed using a semi-physical simulation due to geographical restrictions, rather than a full at-sea polar experiment.
  • Future Work:
    • The paper implicitly suggests that further refinement of the algorithm for even more extreme conditions or integration with other sensors could be explored.
  • The study focuses on initial alignment; future work could investigate the performance of this transversal frame during dynamic navigation in polar regions.

Summary updated

BibTeX
@article{yan2018polar,
  author = {Zheping Yan and Lu Wang and Tongda Wang and Honghan Zhang and Zewen Yang},
  title = {{Polar Transversal Initial Alignment Algorithm for UUV with a Large Misalignment Angle}},
  journal = {Sensors},
  year = {2018},
  doi = {10.3390/s18103231},
}
Figure