Leakage Localization and Quantification in Air-Insulated Power Equipment Using Schlieren Imaging and Spatio-Temporal Neural Networks

Driven by the power sector’s green and low-carbon transition, environment friendly insulation gas equipment is being deployed at scale. With the prolonged operation of gas-insulated electrical equipment, the probability of insulating gas leakage rises substantially, introducing potential safety hazards and threatening the stable performance of power systems. However, conventional detection methods exhibit considerable limitations when applied to modern environmentally friendly insulating gases, particularly dry air, due to their compositional similarity to ambient atmosphere. To address this challenge, this study presents a schlieren imaging approach for detecting leaks of insulating gases. Taking air leakage as a representative case, A high-sensitivity schlieren system was designed and implemented to localize and visualize air-to-air leakage flows. Nevertheless, the captured schlieren images often present incomplete visualizations of the leakage plumes, making it challenging to construct accurate turbulence-based physical models for leakage quantification. To overcome this limitation, a spatio-temporal neural network was introduced to estimate leakage rates in real time based on schlieren image sequences, achieving an R2 of 0.9706 on the test set. Computational visualization of the training process indicates that the model captures salient spatio-temporal signatures of leakage dynamics. By bridging qualitative schlieren visualization with quantitative leakage diagnostics, the proposed optical–computational framework provides a non-intrusive, scalable pathway for online monitoring of insulating-gas leakage in power equipment, thereby supporting safer operation and the sector’s decarbonization objectives.