Interval Fault Detection Based On Set-membership Estimation Method

Fault diagnosis technology can effectively improve the security and reliability of the system and it has great application value in practical engineering,so the study about fault diagnosis is a hot topic.Fault diagnosis based on analytical model requires an accurate model of the system,when the system model uncertainty,the set-membership estimation method can be used for fault diagnosis.In this thesis,the state estimation and fault diagnosis are studied under this background.The ellipsoidal algorithm and the zonotope algorithm are studied.The advantages of the two algorithms are summarized,then a new set-membership estimation algorithm is designed,this algorithm combines the advantages of the two algorithms that simple thinking and high precision.With the increase of the output order,the accuracy of the estimation will be improved.The new set-membership estimation algorithm is used to estimate the state of the uncertain linear discrete-time systems that parameters,noise and disturbances are unknown but bounded.Design of adaptive double observers for fault diagnosis,the first observer is a general observer that does not contain the disturbance and uncertain parameters,and the residual is obtained by the output of the observer and the output of the system.Because the system contains the disturbance and uncertain parameters,it may leads to misjudge only by the residual that the first observer.So the second adaptive observer has disturbance and uncertain parameters,the threshold value of the output of the second observer is calculated by the new set-membership algorithm,then the residual threshold is calculated.When the residual exceeds the threshold value,fault is detected.In order to estimate the state of nonlinear system by the new set-membership estimation algorithm,it is required to use the Taylor formula to linearization the system.The linear error bound is estimated by DC programming method,then,the new set-membership estimation algorithm is used to estimate the state.For fault detection,the double observers are designed for linearization system.The threshold value of the output of the second observer is calculated by the new set-membership method.At the same time the DC programming method is improved for set membership estimation to improve the estimation accuracyA great number of simulations are given to illustrate the efficiency of the proposed method.