Online　dictionary　learning　(ODL)　is　an　emerging　and　efficient　dictionary　learning　algorithm,　which　can　extract　fault　features　information　of　fault　signals　in　most　occasions.　However,　the　typical　ODL　algorithm　fails　to　consider　the　interference　of　noise　and　the　structural　features　of　the　fault　signals,　which　leads　to　the　fault　features　of　weak　fault　signals　that　are　difficult　to　extract.　For　that,　a　novel　feature　enhancement　method　based　on　an　improved　constraint　model　of　an　ODL　(ICM-ODL)　algorithm　has　been　proposed　in　this　paper.　For　the　stage　of　dictionary　learning,　the　elastic-net　constraint　is　used　to　promote　the　anti-noise　performance　of　the　dictionary　atoms.　For　the　stage　of　signals　sparse　coding,　the　l(2,1)　norm　constraint　is　added　to　learn　the　structural　features　of　fault　signals.　In　addition,　a　variational　mode　decomposition　algorithm　is　used　to　reduce　the　impact　of　noise　on　the　signal　initially.　Taking　the　weak　fault　signals　of　bearing　as　examples　for　analysis,　the　results　show　that　the　feature　enhancement　of　the　weak　fault　signals　is　fulfilled　by　using　the　ICM-ODL　algorithm.　Compared　with　the　typical　ODL　method,　the　ICM-ODL　algorithm　can　not　only　improves　the　anti-noise　performance　of　the　dictionary　atoms,　but　also　removes　the　noise　compositions　of　the　reconstructed　signal　significantly.