Data-driven aided parity space-based approach to fast rate residual generation in non-uniformly sampled systems

Abstract

The existing parity space-based fault detection approaches for non-uniformly sampled systems are mostly based on the known system models, and residual signals are generated and evaluated to reflect the inconsistency between the expected behavior and the actual mode of operation. For the system with unknown model parameters, system identification method is required to identify model first and then calculate the corresponding parity vector. In this paper, a novel nonuniformly sampled-data-driven approach to fault detection is proposed directly from test data instead of system identification, based on it, to achieve fast residual-generation as well as dimensionality reduction of parity matrix. Firstly, according to the input-output train data, a linear time invariant subspace lifting model is built for non-uniformly sampled system by use of the lifting technology and subspace method. Then, the parity space-based residual generation is designed by introducing instrumental variable to eliminate the unknown disturbances and faults in training set. Meanwhile, a causal residual system with reduced order is obtained according to non-uniqueness of the solutions of parity matrix. Furthermore, a fast synchronization of residual can be realized by inverse lifting computing. A simulation is given to show the effectiveness of the proposed method.