Shandong Science ›› 2023, Vol. 36 ›› Issue (4): 1-9.doi: 10.3976/j.issn.1002-4026.2023.04.001

• Oceanographic Science, Technology and Equipment •     Next Articles

Wave sensor fault diagnosis method based on t-SNE reduction and KNN algorithm

TAI Penga,b(), SONG Miaomiaoa,b,*(), WANG Boa,b, CHEN Shizhea,b, FU Xiaoa,b, HU Weia,b, GAO Saiyua,b, CHENG Kaiyua,b, ZHENG Shanshana,b, JIAO Zixuana,b, WANG Longfeia,b   

  1. a. College of Marine Technology and Science, Qilu University of Technology (Shandong Academy of Sciences), Qingdao 266300, China
    b. Institute of Marine Instrumentation, Qilu University of Technology (Shandong Academy of Sciences), Qingdao 266300, China
  • Received:2022-09-10 Online:2023-08-20 Published:2023-08-03


This study proposes an efficient wave sensor fault diagnosis method based on wavelet packet decomposition, dimension reduction, and k-nearest neighbor algorithm(KNN) classification network to address the difficulty of wave sensor fault diagnosis, unidentifiable fault types, and time-consuming diagnosis. First, the standard deviation of the original signal is normalized. The normalized data is then subjected to a three-layer wavelet packet decomposition. The extracted feature vectors represent normalized data from the eight bands on layer 3. The second step involves using the t-distributed stochastic neighbor embedding (t-SNE) algorithm to reduce the dimension of the feature data. Finally, the dimension-reduced feature data is input into the KNN classification network for fault classification and detection. Experimental results show that the proposed method can improve the accuracy and diagnosis speed of the wave sensor fault diagnosis, with a diagnosis accuracy of up to 93.55% for normal and six faulty conditions.

Key words: wave sensor, wavelet packet decomposition, t-SNE dimension reduction, k-nearest neighbor algorithm, fault diagnosis

CLC Number: 

  • TP83