Research On Data-driven Fault Monitoring Methods Of Complex Systems Based On Subspace Technology

With the rapid development of science and technology and the continuous improvement of industrial level,industrial systems are becoming increasingly complex,with large system scale,complicated process flow,high uncertainty,and wide distribution.In order to improve system reliability and safety and ensure production efficiency,real-time and effective fault monitoring of complex industrial systems is also more challenging.In related research,the subspace-based method has gradually become one of the important development directions in the field of fault monitoring research due to its advantages such as easy analysis and design,high reliability,inherent connection with system model and the ability to reflect the dynamic characteristics of the process.Therefore,this thesis conducts in-depth research on the subspace-based fault monitoring method for complex system with problems in switched system,large-scale interconnected system,large-scale interconnected switched system,and online optimization.This thesis introduces the background and research significance of fault monitoring of complex dynamic systems,summarizes the basic concepts and research methods of fault monitoring and provides a detailed introduction to the relevant research status and the problems that need to be studied in fault monitoring of switched system,fault monitoring of large-scale interconnected system,fault monitoring of large-scale interconnected switched system,and online optimization fault monitoring.Based on this,the main research content and chapter arrangement of this thesis are introduced.This thesis studies the subspace-based fault monitoring method for switched system with unknown switching signals.Considering the mode mismatching problem caused by unknown switching signals,mode identification is combined with the subspace-based fault monitoring method.A stable kernel representation is constructed for different modes of the switched system,and the relationship between gap metric of kernel representation and cluster radius is discussed.Then a criterion is proposed for mode differentiability of the switched system.Both in the open-loop and closed-loop cases,the normalized stable kernel representations are constructed.Via exploiting the truncation operator,the gap metric is transformed into the maximum singular value problem based on the normalized stable kernel representations,by which the data-driven realization of mode differentiability is derived.Furthermore,a decision logic including mode differentiability is presented to achieve the effective fault monitoring for switched system with unknown switching signals.This thesis proposes a distributed subspace-based fault monitoring method for largescale interconnected systems with unknown communication.Considering the interconnection between subsystems,a distributed fault monitoring structure describing the information coupling of subsystems is established.By utilizing the local process data with unknown communication information,the Hankel matrices are constructed for single subsystem and the observer-based residual generator is obtained.Meanwhile,for each distributed node,an adaptive algorithm is established to handle the changes of processing parameter and operation point.The distributed subspace-based fault monitoring method not only avoids the burden of central computation relying on global information,but also eliminates the impact of unknown communication on fault monitoring,which provides a feasible scheme for the fault monitoring research of large-scale interconnected system.This thesis studies the distributed subspace-based fault monitoring method for largescale interconnected switched systems.Considering the influence of switching behavior of subsystems on global fault monitoring,a mode differentiability criterion is obtained for the global system modes based on the gap metric constraints of a single subsystem.Based on the global closed-loop extended model of subsystem interconnection,the stable kernel representation of the global system is transformed into its distributed expression on the subsystem by using the permutation matrix and the corresponding kernel space is constructed.By employing the local information matrix,a multi-mode distributed residual generator is constructed for each subsystem node using process data.Based on the local data on the subsystem and the information interaction between subsystems,a distributed subspace fault monitoring method with mode estimation window is established to realize the global fault monitoring of large-scale interconnected switched systems.This thesis explores the subspace-based fault monitoring method for dynamic system under the online optimization demand.By constructing the stable kernel representation of the observer-based residual generator and establishing its normalized form,an optimization problem with the gap metric of kernel representation as the optimization objective is proposed.By virtue of the gradient descent iterative optimization method,the online optimization of residual generator is obtained with the available input and output data.Further,considering the asymptotic stability demand,the parameterized matrix satisfying the column orthogonal constraint is constructed,and the online optimization strategy based on input-output normal form is developed.The proposed online optimization method avoids the parameter updating at each sampling time and realizes the real-time optimization of the residual generator parameters with the computational burden of the optimization process reduced.