“…In the first place, they can be adapted to utilize the residuals of an existing filter with relative ease, and a wide range of procedures--from the simple statistical tests of Mehra and Peschon [26] to the more complex GLR method of Wiilsky et al [28], [29], [38], [43]---are available, offering various tradeoffs between performance and complexity. In addition, since we implement a Kalman filter based on no failures, we suffer no performance degradation prior to detection, unlike, for example, the Beard-Jones approach.…”