A Robust and Efficient IMU Array/GNSS Data Fusion Algorithm

• Published in: IEEE sensors journal 2024-08, Vol.24 (16), p.26278-26289
• Main Authors: Zhang, Tisheng, Yuan, Man, Wang, Liqiang, Tang, Hailiang, Niu, Xiaoji
• Format: Article
• Language: English
• Subjects: Arrays; Classification algorithms; Computational complexity; Data fusion; Data integration; Global navigation satellite system; inertial measurement unit (IMU) array; inertial navigation system (INS)/global navigation satellite system (GNSS); micro-electromechanical system (MEMS) IMU; Navigation; Sensors
• ISSN: 1530-437X; 1558-1748
• DOI: 10.1109/JSEN.2024.3418383

The inertial measurement unit (IMU) array, composed of multiple IMUs, has been proven to be able to effectively improve the navigation performance in inertial navigation system (INS)/global navigation satellite system (GNSS) integrated applications. The conventional IMU-level fusion algorithm, using IMU raw measurements, is straightforward and highly efficient but yields poor robustness when the IMU array is not rigidly installed. On the contrary, the classic INS-level fusion algorithm, using navigation results from each IMU, is immune to the nonrigid installation of the IMU array but suffers a heavy computation load. Here, we propose a robust and efficient INS-level fusion algorithm for IMU array/GNSS (eNav-Fusion). Each IMU in the array shares the common state covariance (P matrix) and Kalman gain (K matrix), and the navigation solutions of all IMUs are eventually fused to produce a more accurate solution. The proposed eNav-Fusion was fully evaluated with rigidly and nonrigidly installed IMU arrays. For a rigid 16-IMU array, the processing time of eNav-Fusion was close to that of the IMU-level fusion and only 1.22\times to that of the INS/GNSS algorithm for a single IMU; and the navigation performance was improved by 2.51\times , as expected for such scale of array. For a nonrigid 6-IMU array, in which case the traditional IMU-level fusion does not work, eNav-Fusion still maintained the same accuracy as the classic INS-level fusion algorithm, while the computation load is still close to that of the IMU-level fusion. In conclusion, the proposed eNav-Fusion achieves the same robustness as the INS-level fusion, while only consuming comparable computational complexity to the IMU-level fusion.