Designed inputs for detection and isolation of failures in the state transition matrices of dynamic systems

Abstract

A new failure detection and isolation algorithm is presented for linear time-invariant deterministic systems. In particular, the technique exploits the system inputs to distinguish between a change in the state transition matrix and actuator failures, including output equivalent events. It is assumed that a failure has been detected previously by some other method, such as by a failure detection filter. Previous methods that attempted to isolate changes in the state transition matrix did not use the input in any way and required explicit knowledge of the changed parameter values (e.g., the multiple model method). The designed input method presented in this paper requires only the knowledge of the null spaces of the changed parameter matrices, not of the explicit values of the changed parameters. This feature allows the failure detection and isolation (FDI) algorithm based on designed inputs to isolate partially unknown changes in the state transition matrix, a distinct advantage over other methods in terms of computational efficiency. Detailed design procedures are accompanied by an example that illustrates the input design method. The example uses designed inputs to distinguish between changes in the state transition parameters of an AFTI/F-16 model from nearly output equivalent actuator calibration errors.