以特徵結構設計 UIFIO 錯誤診斷器

Abstract

隨著系統安全性與可靠度的日益提高，一個好的控制系統設計必須同時兼 顧輸出表現與安全考慮。本論文首先提出 UIFIO (unknown input fault isolation observ- er)演算法，藉由特徵結構設計法 (eigenstructure assignment)，特徵向量(eigenvetor) 的可塑性，使診斷系統在可結構化 之不確定因子 (st- ructured uncertainty) 干擾下，能把外界干擾效應 除去，而將所有可能發生的不同錯誤組合分辨出來 (isolation)。針對非 結構性之不確定因子對診斷系統的影響，配合 UIFIO 演算法，我們進一 步提出兩種「限制條件下，最小狀態之特徵向量近似法 (constrained approximation for minimizing the condition number of eigenvetor assignment)」。 藉由演算法，我們可以再無限多組解中，逼近一組次 佳 (locally optimal solution) 的設計值，使得診斷系統不僅不受結構 性不確定因子干擾，更能將非結構不確定因子的影響盡可能減低。 As the demands on safety and reliablity of automatically co- ntrolled system become more stringent,a good control system has to not only achieve the performacne requirements but also guara- tee the safety concern. In this thesis, an UIFIO (unknown input fault isolation obs- erver) will be firstly proposed with eigenstructure assignment and the corresponding numerical algorithm will also be summari- zed. With the flexiblity provided by output feedback in multiva- riable systems beyond closed-loop eigenvalure assignment, the f- ault diagnostic observer decouples the structured uncertanity f- rom residual outputs and performs isolation for any possible co- mbination of failure occurrences. Furthermore, since the robust- ness of the diagnostic observer relative to unstructured uncert- ainty depends on the condition number of closed-loop eigenvecto- r set, two constrained approximations for minimizing condition number of closed-loop eigenvector assignment based on the UIFIO are proposed. With these two approximation methods, a locally optimal solution of closed- loop eigenvector set can be found to suppress the effects induced by unstructured uncertanity.