研究目的
To develop a method for online identification of AUV dynamics via in-field experiments to update the dynamic model for better control and guidance law design.
研究成果
The developed method enables rapid online identification of AUV dynamics, with parameters converging within 12 seconds. The identified models explain 78% to 92% of the output variation and outperform conventional offline methods in prediction accuracy, computational cost, and training time. The models are useful for estimating the AUV's turning radius at different speeds and designing gain-scheduled controllers.
研究不足
The method requires the AUV to perform specific maneuvers for identification, which may not be feasible in all operational scenarios. The accuracy of the identified model depends on the quality of the excitation signals and the AUV's sensor measurements.
1：Experimental Design and Method Selection:
The AUV is commanded to execute a compact set of maneuvers under doublet excitation. The identification process uses state variable filter and recursive least square (SVF-RLS) estimator to estimate unknown parameters.
2：Sample Selection and Data Sources:
Training and validation data sets are collected from the AUV's response to rudder inputs during the experiments.
3：List of Experimental Equipment and Materials:
STARFISH AUV equipped with sensors including a compass module, pressure sensor, and doppler velocity log (DVL).
4：Experimental Procedures and Operational Workflow:
The AUV performs straight runs at constant depth and thrust, followed by excitation signals to excite yaw dynamics. Parameters are estimated online during the training stage and validated using fresh data.
5：Data Analysis Methods:
The SVF-RLS estimator is used for parameter estimation, and the coefficient of determination is calculated to validate the model's prediction capability.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
