The overall architectural complexity of autonomous underwater vehicles continuous to increase, enlarging the probability of fault occurrence in subsystems. Estimating the thrust loss by particle filter provided a useful method to detect the fault in propeller subsystem. In order to detect the fault in propellers as early as possible, the particle filter direct prediction method could amplify the fault trend and detect the fault earlier, but at the same time increase the possibility of false diagnosis. Therefore, a more accurate fault diagnosis method was required to discover the fault early and decrease the occurrence of false diagnosis. In this paper, an improved particle filter prediction method was proposed, combining the advantage of grey prediction to forecast the motion state, reducing the uncertainty in particle filter direct prediction process. Besides, the Gaussian kernel function was applied to judge the credibility of the prediction result, decreasing the possibility of the false diagnosis. In the experiments with simulated working conditions data and a section of actual sea trial data with propeller fault, the proposed method detected the fault earlier compared with the original particle filter method, and reduced the false diagnosis rate compared with the particle filter direct prediction method. The results show that the proposed method is effective in detecting the fault early with low false diagnosis.
Rolling Bearing Fault Prediction Method Based on QPSO-BP Neural Network and Dempster–Shafer Evidence Theory
