Multiple AUVs cooperative localization as a new kind of underwater positioning technology, not only can improve the positioning accuracy, but also has many advantages the single AUV does not have. It is necessary to detect and isolate the fault to increase the reliability and availability of the AUVs cooperative localization system. In this paper, the Extended Multiple Model Adaptive Cubature Kalmam Filter (EMMACKF) method is presented to detect the fault. The sensor failures are simulated based on the off-line experimental data. Experimental results have shown that the faulty apparatus can be diagnosed effectively using the proposed method. Compared with Multiple Model Adaptive Extended Kalman Filter and Multi-Model Adaptive Unscented Kalman Filter, both accuracy and timelines have been improved to some extent.
Since a good knowledge of MEMS gyro stochastic errors is important and critical to MEMS INS/GPS integration system. Therefore, the stochastic errors of MEMS gyro should be accurately modeled and identified. The Allan variance method is IEEE standard method in the filed of analysis stochastic errors of gyro. This kind of method can fully characterize the random character of stochastic errors. However, it requires a large amount of data to be stored, resulting in large offline computational burden. Moreover, it has a painful procedure of drawing slope lines for estimation. To overcome the barriers, a simple linear state-space model was established for MEMS gyro. Then, a recursive EM algorithm was implemented to estimate the stochastic errors of MEMS gyro in real time. The experimental results of ADIS16405 IMU show that the real-time estimations of proposed approach are well within the error limits of Allan variance method. Moreover, the proposed method effectively avoids the storage of data.