Industrial　Control　System(ICS)　security　is　one　of　the　lifebloods　of　national　development.　Fully　understanding　of　its　vulnerabilities　plays　an　important　role　in　the　actual　application　scenarios.　Meanwhile,　an　attacker　may　also　exploit　multiple　vulnerabilities　to　achieve　the　final　malicious　purpose,　such　as　the　Stuxnet　worm.　In　order　to　solve　the　above　problems,　we　construct　a　Knowledge　Graph(KG)　of　heterogeneous　ICSs,　and　propose　a　potential　relationship　mining　method　(R-HetGNN)　based　on　this　graph.　The　method　solves　the　multi-modality　problem　in　KG　aggregation　and　KG-heterogeneity　problem.　Besides,　we　use　random　walk　algorithm　to　solve　the　ulti-level　neighbor　problem.　Experimental　results　on　a　real-world　dataset　show　that　R-HetGNN　achieved　83.0%　on　the　F1　score,　superior　to　other　knowledge　reasoning　modules,　such　as　GAT　and　TransE.