An optimal fault detection (FD) approach for a class of networked control systems (NCSs) is concerned in this work. To improve the accuracy of the FD, a new event-triggered scheme (ETS) is addressed, where the output measurement transmitted or not is determined by a set of conditions instead of a single condition. Moreover, output measurements in these conditions are not only the last transmitted data, but also the data in the past period of time. The structure of fault detection and isolation (FDI) consisting of the residual generation and residual evaluation can detect faults as accurately as possible, and show robustness to disturbances in the meantime. Optimal observer gains can be obtained by utilizing the discrete-time Riccati equation (DTRE), and the time-varying threshold can be specified by employing the linear matrix inequality (LMI). Finally, a numerical simulation and the application on closed-loop continuous stirred-tank reactor (CSTR) are adopted to prove the superiority of the proposed approach.