復雜非線性系統(tǒng)的智能故障診斷與容錯控制

1、南京航空航天大學博士學位論文 V摘 要 論文針對具有多變量、時滯、不確定性或模型未知的復雜非線性系統(tǒng)，將狀態(tài)觀測器理論、魯棒控制控制理論、自適應控制理論與先進的智能技術包括神經網絡和 Takagi-Sugeno（T－S）模糊模型有機結合，研究了系統(tǒng)狀態(tài)不可測時的故障診斷和主動容錯控制。 首先，給出了非線性系統(tǒng)能觀的有關概念，針對部分模型已知的非線性系統(tǒng)，當系統(tǒng)相對階為 n（n 為系統(tǒng)的階數）時，提出了一種觀測器設計方案，利用非線性

2、變換，將含有建模誤差的非線性系統(tǒng)變換為僅依賴原系統(tǒng)的輸入、輸出的規(guī)范形式，從而可利用可測數據構造觀測器。當系統(tǒng)的相對階小于 n 但為最小相位系統(tǒng)時，用 RBF 神經網絡逼近系統(tǒng)的不確定性，解決了具有不確定性非線性系統(tǒng)的觀測器設計問題。 在解決了狀態(tài)觀測器設計問題之后，針對一類含有部分模型已知的非線性系統(tǒng)，利用非線性變換的方法研究狀態(tài)不可測時的故障診斷問題。RBF 神經網絡作為故障估計器，其輸入為系統(tǒng)的估計狀態(tài)，輸出為估計的故障，既可用作

3、構造容錯控制律也可用作報警。 針對一類模型未知非線性系統(tǒng)，提出了基于自適應神經網絡的故障診斷策略，用一個 RBF 神經網絡構造狀態(tài)估計器，用另一個權值和中心可以在線調整的 RBF 神經網絡構造故障估計器，解決了模型未知且系統(tǒng)狀態(tài)不可測時的故障診斷問題。 本文與常規(guī)的應用 LMI（Linear Matrix Inequality）研究時滯系統(tǒng)的方法不同，針對一類狀態(tài)不可測且模型未知的非線性時滯系統(tǒng)，提出了基于神經網絡的故障診斷新方法。直接

4、估計系統(tǒng)的狀態(tài)及時滯狀態(tài)，然后作為神經網絡的輸入對故障進行估計，并對系統(tǒng)的穩(wěn)定性進行了論證。仿真結果表明該方法簡單有效。 非仿射非線性系統(tǒng)的研究比一般的仿射非線性系統(tǒng)更復雜。 針對一類非仿射非線性系統(tǒng)，研究了基于狀態(tài)觀測器的魯棒自適應 H∞ 跟蹤控制問題。RBF 神經網絡用來在線抵消非線性模型誤差， 高增益觀測器用來估計不能直接測量的輸出導數。 在系統(tǒng)沒有擾動時， 確保跟蹤誤差漸近趨于零且系統(tǒng)的所有信號有界； 當擾動存在時，能取得預期的

5、 H∞ 跟蹤性能。由于磁懸浮系統(tǒng)的非線性模型為非仿射非線性系統(tǒng)，本文方法針對磁懸浮系統(tǒng)進行了大量的仿真研究。 南京航空航天大學博士學位論文 VIIAbstract The intelligent fault diagnosis and active tolerant control for a class of complex uncertain nonlinear dynamic systems or model-unkown ti

6、me delay systems with unmeasured states are studied in this dissertation. Based on the states observer design theory, robust control theory , adaptive control theory and combined with advanced intelligent techniques i

7、ncluding the neural network and T-S fuzzy model, a set of fault diagnosis and reconfigurable control methods are proposed.The main contents of the dissertation are as follows: At first, the observability definitions f

8、or the nonlinear system are given. A state observer design method for a class of uncertain nonlinear systems whose reletive degree equals the system order n is addresed.The system is transformed diffeomorphically into

9、 a canonical system with the modelling error only depending on the measurable input and output data. When the systems reletive degree is less its orders, a new observer structure is introduced, and a neural network i

10、s applied to approximate the uncertainty . The observer guarantees that the state estimate error converges to zero provided that system zero dynamics is asymptotically stable. Based on estimated states, a fault diagno

11、sis architecture for a class of uncertan nonlinear syetems is proposed, a diffeomorphism is applied to transform the nonlinear system into a new coordinate system . The estimated states are input to the fault approxi

12、mator whose outputs are estimated fault model.When a system model is unknown and the states are unavailable for measurement, the states are estimated on-line by employing a general RBF neural network, while the fault o

13、f system is estimated by an adaptive RBF neural network where center and width vectors of Gaussian function are on-line updated. Further the fault diagnosis scheme for a class of nonlinear time-delay systems with unme

14、asured states is studied. Unlike the usual method which is within framework of Linear Matrix Inequalities techniques(LMI) , in this paper,the estimated states and time-delay states are used the input to the neural net

15、works in order to approximate the fault model.The stability of the error system are analyzed using Lyapunov stability theory. A robust adaptive tracking control architecture with state observer is proposed for a class

16、 of nonaffine nonlinear systems. A high-gain observer is used to estimate the derivatives of system output, a RBF neural network is used to cancel nonlinear uncertainties. Applying estimate states , the track controll

17、er is designed , the fixed control law and adaptive law are derived. It is shown that the tracking error is guaranteed to be asymptotically convergent with the aid of an additional robustifying control term when there

18、 not exists externel disturbance, and a tracking perfomance is achieved with a externel disturbance. A magnetic levitated ball system is used as a simulation example. Furthemore, the active fault-tolerant control te