Fault diagnosis of free-conducting particles within GIL based on vibration signals

Accurate quantitative diagnosis of free-conducting particle faults plays an important role in improving the reliability of the gas insulated line (GIL) system. However, the existing fault diagnosis methods cannot accurately identify the free-conducting particle faults with different quantities and sizes. Motivated by this, this paper proposes a novel fault diagnosis method based on vibration signals, which integrates variational mode decomposition (VMD), self-adapting whale optimization algorithm-multiscale permutation entropy (SAWOA-MPE), and deep forest (DF). First, the raw vibration signals of free-conducting particle faults are decomposed via VMD, and the decomposed signals are reconstructed based on the correlation degree. Afterwards, SAWOA is employed to optimize the critical parameters of MPE, and the optimized MPE is further utilized to extract the fault features of the reconstructed signals. Finally, the extracted feature vectors are trained and tested to construct a valid DF classification model that identifies the free-conducting particle faults. The experimental results indicate that the identification accuracy of the proposed method can reach 99.5%. Moreover, comparative tests based on various feature vector extraction methods and classification models further validate the superiority of the proposed method.