In this paper, the author develops a new design method of a reduced-order unknown input observer(UIO) for estimating states of a system which is subject to unknown inputs. The proposed UIO estimates state and unknown input in parallel. Unknown inputs are modeled to be present in the state and output equation. The major advantage of the proposed method is that the unknown inputs of the state equation and those of the output equation can be handled simultaneously without state augmentation. Also, no a priori knowledge on modes of unknown inputs is required. Since some faults of sensors and actuators can be modeled as additive unknown inputs, the proposed UIO can be used to diagnose a control system. This method is based on a general system model, so it is shown that results of previous studies can be obtained by specializing the proposed method.
This paper also deals with the use of output anticipation signals to relax restrictions of existing UIO design methods. The anticipation signals are treated as additional outputs of the original system, and a modified existence condition is presented. The modified condition is shown equal to the condition of system inversion. Numerical examples show that the proposed UIO using the anticipation signals extends the applicability of UIO to a larger class of dynamic systems.
There can be the problem that a system designer selects sensors to diagnose a system. In that case, this paper propose a criterion of selecting sensors which give good estimation performance. The proposed criterion is cost functions of robustness problem. Several applications show that this criterion is useful and practical to choose sensors for good estimation performance.