Research On Design Method For Diagnosis Systems Based On Structural Analysis

In model based diagnosis,the structural analysis approach is an important branch for fault diagnosis.In structural analysis,the structural model of a diagnosis system is investigated to obtain useful structural information.By analyzing the relation between unknown variables and equations in the structural model,the structural properties of the system are revealed for identifying fault diagnosability,obtaining the Minimal Structurally Overdetermined(MSO)sets which can be used to derive the residuals,and so on.Structural analysis can be used for fault diagnosis,the design of fault diagnosis systems and the fault-tolerant control.In the field of structural analysis,there are still some shortages in the research and application,such as obtaining the MSO sets which usually causes a great deal of computation;identifying the fault isolability properties of a diagnosis system is still based on pairwise comparisons between the faults,which causes large amounts of computation;in optimal sensor placement,the effectivity analysis of adding sensors to enhance model based fault isolability properties that is not discussed,and a few amount of literature addresses how to identify redundant sensors based on perfect consistency of fault isolability properties.Aiming at the above problems,the thesis has an in-depth investigation and the related contributions of this thesis are concluded as follows.Based on the properties of equivalence classes based on fault isolation,an improved method to derive the MSO sets is presented in this thesis.First,the proper structurally overdetermined(PSO)sets which has high degree of redundancy are partitioned into equivalence classes based on fault isolation.Second,the intermediate PSO sets are derived by removing two different equivalence classes at the same time,then the degree of redundancy of the intermediate PSO sets are quickly reduced.The iteration is performed until all the MSO sets are obtained.The new approach helps speed up the solution process,and the computational cost for the sets of intermediate overdetermined equations with the high degree of redundancy is lower.It offers a new way to derive all the MSO sets.For the isolability properties of a diagnosis system,this thesis presents a new approach based on an augmented system model to efficiently determine the properties.Unlike conventional structural analysis approaches,the fault signals are regarded as additional unknown variables,then the augmented system model can be built.Then,by using only one time structural model decomposition and analyzing the computational dependencies between the fault variables,some faults can be isolated,and then the fault equations in the model are classified,which significantly reduces the search space for non-isolable faults.Later,in the corresponding monitored system model,the sets of structurally linked equations associated with the search space can be derived,and then all sets of non-isolable faults can be fast extracted.Finally,the isolability properties of the diagnosis system are disclosed.Compared with previous approaches,the new approach significantly improves the efficiency of identifying fault isolability properties,and the computational cost is lower.In the field of sensor placement,the isolability properties of the diagnosis system based on adding sensors are investigated,especially the effectivity analysis of adding sensors for improving the fault isolability properties has been addressed.First,the model of a diagnosis system and the updated model with the new sensors are decomposed,and it can be determined whether the considered fault equations belong to the same set of structurally linked equations.Then,according to the properties of the set of structurally linked equations,the effects on the isolability properties are presented when different variables are measured,and then all the state variables are divided into three types of variable sets.Finally,some variables in one of the three variable sets are measured by new sensors,which can be used to distinguish two non-isolable faults,and then the isolability properties are improved.The new approach need not compute all possible MSO sets for determining the fault isolability properties,and the selection space of adding sensors is significantly reduced.Therefore,the approach has low computational cost.A new approach for identifying redundant sensors based on perfect consistency of fault isolability properties is proposed in this thesis.By using structural model decomposition,the set of common variables of a diagnosis system is derived,and the criterion,i.e.if one(or some)sensor(s)is(or are)removed from the diagnosis system,and the set of common variables of system is the same with the one of the updated system,then,the fault isolability properties of the two associated systems are identical.Therefore,there exist redundant sensors based on perfect consistency of fault isolability properties in the diagnosis system.This approach provides a new train of thought for solving the problem of optimal sensor placement.The problem of optimal sensor placement has been addressed by using structural analysis in the thesis.A benchmark four-tank system is used as a case study to illustrate the principle of the approach.According to the diagnostic requirements,all possible combinations of the permitted state variables are measured by additional sensors,then all possible four-tank diagnosis systems are built.Each of those system models can be partitioned respectively into equivalence classes based on fault isolation,and the fault isolability properties can be disclosed.Comparing the properties with the required properties,all of the desired sensor placement schemes are obtained.Finally,the optimal sensor placement scheme can be extracted,in which the number of sensors is least,and then simulation results validate that this method is feasibility and effectiveness.