With the rapid development of railway lines, high voltage disconnectors have become the most fre- quently used high voltage equipment. The reliability of OCS isolating switch is of great significance for ensuring the stable operation of traction power supply. As a very effective tool for clustering analysis, the K-means algo- rithm processes the isolated current motor's current signals monitored by the motor current acquisition system, obtaining the cluster centers and the change of their positions, thus determining the fault isolation switch with mechanical fault classification for further processing. Because of the complexity of mechanical failure of high voltage isolation switch, and the limitations of the K-means algorithm itself, this paper proposed an improved K- means algorithm by combining the adaptive weights and GA algorithm (genetic algorithm). K-means algorithm avoided such defects as dependence on initial value and local convergence. The validity of the methods was ver- ified by simulating the actual data. It is found that the mechanical fault of the high-voltage isolating switch can be classified and used as the basis for fault classification and detection of OCS isolating switch.