研究目的
To propose a novel calibration method for star sensor installation error in a stellar-inertial navigation system using a regularized backpropagation (BP) neural network, aiming to improve calibration accuracy and tolerance for large installation errors.
研究成果
The proposed method using a regularized BP neural network effectively calibrates star sensor installation errors with high accuracy and strong tolerance for large errors. It outperforms traditional Kalman filter methods in terms of accuracy and calculation time, making it suitable for both integrative and separated installation modes in Stellar-INS.
研究不足
The method's performance under extremely large installation errors beyond the tested range (up to 2°) is not verified. Additionally, the training process is time-consuming and requires a simulation environment.
1：Experimental Design and Method Selection:
A regularized BP neural network is designed for star sensor installation error calibration. The network structure includes one hidden layer with improved regularization through adjusting the performance function.
2：Sample Selection and Data Sources:
Simulation data are collected from a digital platform simulating a stellar-inertial navigation system, including IMU and star sensor measurements.
3：List of Experimental Equipment and Materials:
A single star simulator, a turntable, and the Stellar-INS are included in the simulation environment.
4：Experimental Procedures and Operational Workflow:
The calibration procedure involves observing a star simulator at three different positions to satisfy the observability of three-axis installation errors.
5：Data Analysis Methods:
The neural network is trained with simulation data to estimate installation errors, and its performance is compared with traditional Kalman filter methods.
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
