Multi-input multi-output self-organizing detection and filter design

A suboptimal auto-regression moving average detection filter is proposed using only input and output measurements to monitor the failures occurring in a multi-input multioutput stochastic linear time invariant system with unknown system parameters. Several self-organizing procedures including system updating construction of a discriminate function, and suboptimal detection filter-output estimation are incorporated in the augmented detection system in order to enhance its performance. Advantages of the proposed detection system include: (1) maintain continuous failure detection capability and at the same time provide suboptimal output estimation; (2) enhance output residue in order to ease the task of failure decision; and (3) preserve failure detectability so long as failures are output separable. The effectiveness of the filter and simplicity of design are illustrated through an example