A Sparse Sensor Placement Strategy for Vibration Reconstruction of GIS Enclosure and Inverse Identification of Internal Vibration Sources

To achieve the reconstruction of gas-insulated switchgear (GIS) enclosure vibration cloud maps through limited vibration measurement points and further realize the localization and identification of vibration sources on internal conductive rods, this paper proposes a combined proper orthogonal decomposition (POD)- RRQR(RRQR) for enclosure vibration cloud reconstruction, as well as a convolutional neural networks (CNN)- kernel density estimation (KDE) model for internal vibration source inversion. First, the surface vibration field data is projected onto the POD basis, and the RRQR is used to quickly select the minimum sensor locations from candidate measurement points, enabling compressed sensing reconstruction. Second, the reconstructed vibration cloud images and measurement point data are separately used to train a CNN-KDE vibration source identification model, which outputs the confidence interval of the vibration source location. The research provides a quantitative basis for sensor deployment and vibration source localization in the online monitoring of GIS mechanical faults.