This　article　studies　the　performance-oriented　fault　detection　(FD)　for　nonlinear　control　systems　in　the　data-driven　framework.　Unlike　the　traditional　residual-　or　test　statistic-based　FD　methods,　the　proposed　approach　focuses　on　evaluating　the　system　performance　changes/degradations　caused　by　the　faults.　Specifically,　the　regulation　performance　described　by　an　infinite-time　cost　function　is　first　introduced　as　a　control　performance　index　of　the　feedback　system.　Using　only　the　process　input　and　output　(I/O)　data,　a　Takagi-Sugeno　fuzzy　dynamic　model　is　constructed　with　the　aid　of　subspace　identification　method.　Then,　the　internal　system　state　variables　are　expressed　in　terms　of　the　I/O　data　and　a　fuzzy　cost　function　is　proposed　to　approximate　the　performance　index.　On　this　basis,　the　temporal　difference　error　induced　by　the　Bellman　equation　is　adopted　as　the　process　evaluation　indicator,　whose　threshold　is　determined　through　the　randomized　algorithm.　To　demonstrate　the　effectiveness　of　the　proposed　FD　approach,　case　studies　are　performed　in　the　end　on　a　ship　propulsion　system　and　a　three-tank　system.