An incipient fault diagnosis method for substation equipment based on multidimensional temperature gradient features

Infrared thermography is widely used for fault diagnosis of substation equipment. However, existing temperature-based methods mainly rely on absolute temperature information and are highly sensitive to environmental disturbances, such as ambient heat sources and wind speed, which limits the reliable identification of incipient faults. To address this issue, this paper proposes an incipient fault diagnosis method for substation equipment based on multidimensional temperature gradient features. The key innovation is to shift the diagnostic basis from absolute temperature analysis to multidimensional thermal variation analysis by introducing temperature gradient as a more robust indicator of local thermal state changes. First, the physical significance of temperature gradient for condition characterization under environmental interference is analysed based on Fourier’s law. Second, a YOLOX based detection method is developed to automatically localize substation equipment from infrared images and extract the corresponding diagnostic regions. Third, a multidimensional diagnosis strategy is established by jointly using the hotspot temperature gradient M ∇ T and the relative temperature gradient difference R ∇ T , enabling incipient faults to be characterized from both local thermal intensity and spatial thermal variation. By integrating automatic equipment detection with multidimensional temperature gradient analysis, an automatic diagnosis framework for incipient faults is established. Experimental results show that the proposed detection method attains a mean average precision of 97.76%. The diagnosis thresholds derived for incipient faults are 0.88 °C/m for M ∇ T and 33.00% for R ∇ T .