Abstract
Neural networks and deep learning for fault diagnosis requires gathering a significant amount of data to train the network model. Continuous parameter tuning is essential. Despite employing regularization techniques to improve the algorithm's performance on the training set, it lacks strong generalization capabilities for fault detection in the variable operating conditions of underwater thrusters. Additionally, the algorithm exhibits high computational complexity, leading to suboptimal real-time fault detection and limited adaptability. To address real-time detection in the case of entanglement, the correlation analysis method with variable sliding window is proposed, a kind of online adjustable gain fuzzy clustering method with adjustable gain is carried out to deal with the entanglement detection. The electrical current and velocity collecting application is designed to gather electrical current and velocity from submerged propellers in various operating states using the variable sliding window method, and then standardize the gathered information. The correlation matrix is used to detail the cross-correlation and autocorrelation between standardized sensors data using a sliding window for data-collecting series. An adjustable gain fuzzy clustering method with improved distance index and adjustable gain is used to detect the entanglement based on the correlation matrix elements. During the process of real-time correlation analysis and entanglement detection, the variable sliding window method is used for capturing accurately of variant electrical current and velocity information. The adjustable gain is adopted in the fuzzy clustering to adjust the convergence speed. The effectiveness of the proposed fault detection method is verified using collecting data from a case of propeller entanglement, and it is compared and validated with the support vector machine with better real-time computing performance. The results indicate that in ensuring the accuracy of fault detection, the proposed electrical current and velocity information method can extract the multi-sensor cross-correlation information of the submerged propeller with lower computational burden.
DOI: 10.61416/ceai.v26i3.8995