Diagnosing mechanical faults in Gas Insulated Switchgear (GIS) disconnectors is crucial for ensuring stability of the power grid. Existing methods focus on features extracted from the mechanical process of GIS disconnectors, which are susceptibility to interference in substations, leading to an unsatisfied accuracy. We propose the Very Fast Transient Current (VFTC) is strongly related to the gap distance between contacts. Thus, the feature, peak value of VFTC in each breakdown, is extracted. Next, A three-stage model comprised of feature extraction, off-line modeling and online monitoring is established to achieve online diagnosis. The experiments show that the overall discrimination accuracy of this method is 96.3%.