Kemeng Ren, Daqi Zhu, Tianhong Zeng, and Simon X. Yang
Fault detection, restricted Boltzmann machine (RBM), grey model, unmanned underwater vehicles (UUVs), residual processing
A fault detection algorithm consisting of a grey model-restricted Boltzmann machine (GM-RBM) prediction model and a dual- criterion residual processing model is proposed for unmanned underwater vehicle sensors. Firstly, in the offline training phase, data under fault-free conditions are collected to train the GM- RBM prediction model. Secondly, in the online deployment phase, the trained network calculates the predicted data. The residual sequence is obtained by subtracting the predicted data from the actual data. Finally, the residuals are input into the dual- criterion residual processing model to detect sensor faults. Simulation experiments demonstrate that the prediction model achieves high prediction accuracy and training speed. In contrast, the residual processing model excels in analysing residuals with high accuracy and robustness in parameter selection. The effective combination of these two models accurately identifies faulty data and ensures comprehensive fault detection.
Important Links:
Go Back
