To detect instrument failures for the secondary system, the FDI (Failure Detection \& Isolation) method based on the Kalman filter is developed. Each filter is designed to be insensitive to measurements, decreasing the Kalman gain artificially and using only averaged value of measurements. Because it is mainly dependent upon the dynamic model and little influenced at the moment of failure, it can exactly indicate the direction of failures and identify the common-mode and multiple failures. This concept not only minimizes the number of filters but also performs a role of analytic redundancy like the original Kalman filter. As soon as the residual (difference between estimated values and measurements) exceeds the pre-determined bound, i.e., more likely to fail, the Kalman filter is stopped. But since this measurement may be false owing to abrupt noises and sudden transients, several times it must be confirmed by the MCM (Multiple Consecutive Miscomparison) counter strongly dependent on measurement history. Then, if it is not in accord with other measurements for many times, detailed information is provided to help operator``s decision. Utilizing the very simple logic in failure detection, it is possibly implemented in real-time for practical system. And the results of simulation prove that it utterly identifies common-mode and multiple failures regardless of the number of direct redundancies.