Fault Diagnosis For Dynamic Systems Based On Adaptive PCA And Temporal Logic

In order to ensure that the system runs under normal and safe operation conditions,the research on fault diagnosis is much concerned.Usually,in complex and dynamic systems,mathematical mechanism and knowledge of processes are difficult to obtain in advance,while process data are rich and easy to obtain.Thus,the research on the pure-data-model based fault diagnosis of dynamic systems is of great significance.Our work focuses on the model updating problem and the feature expressing problem in the fault diagnosis of dynamic systems.From the aspects of the adaptive extraction of global and local information,adaptive and sparse model,the introduction and extension of temporal logic,by using adaptive principal component analysis and temporal logic,our research work mainly includes:1.A fixed and global model may fail to extract the time-varying and local information on dynamic systems.To address this limitation,an adaptive partitioning principal component analysis method for fault diagnosis is proposed.First,according to each system operation,the process variables are partitioned into several blocks.The different system operations have their corresponding partitioning results.Then,considering the local information within the blocks and the global information between the blocks,different global and local models for fault diagnosis are constructed.Finally,the proposed method is demonstrated in the application of the TE process,as a result,both the fault detection rate and the fault isolation accuracy are improved.2.The research on the adaptive sparsity of a time-varying model is studied.First,an operation-related sparse penalty term is introduced to the objective function of the original optimization problem of principal component analysis.In this way,the optimization problem of the adaptive sparse principal component analysis is obtained.Next,for the solution of this optimization problem,an iterative interior point algorithm is presented.Then,an adaptive sparse model for fault diagnosis is constructed,and two monitoring statistic QT2 and SPE are defined.Moreover,a fault isolation algorithm is proposed.The algorithm first reconstructs the most-fault-related variables.Finally,the adaptive sparse method is demonstrated in the application of the TE process and the waveform system.3.The problem of feature expression and extraction of time-varying processes is studied.A hierarchical framework for fault detection and identification is proposed based on principal component analysis and temporal logic.As a nature language,temporal logic is used to express time-varying process features in three layers,namely,process variable layer,principal component layer and monitoring statistic layer.A new piece-wise linear fitting algorithm is proposed for the statistical learning of temporal logic.Then,according to normal and fault system operations,normal and fault temporal logic bases are constructed.Next,the fault detection is carried out by checking whether the current system behavior satisfies the temporal logic formulae in the normal base or not.When a fault is detected,according to the fault base,the similarity measure algorithm is used to identify the type of the detected fault.Finally,the proposed framework for fault detection and identification is applied to the TE process,as a result,the accuracy of fault detection and identification is improved.4.In order to solve the problem that the existing temporal logics lack of the expression power,a new temporal logic is extended.The extended temporal logic is successfully used for online fault detection.First,principal component analysis is used to select the first several principal components as the process feature variables.Then,according to the feature variables,a series of steps are conducted for the statistical learning of the extended temporal logic,including region-of-interest discovery,generation of a pair of region-of-interest and time,time automaton based frequent feature pattern extraction,and the translation of the frequent feature pattern.Next,a fault diagnosis model is constructed in an automaton form.Finally,the online fault detection is carried out in the robot arm system and the waveform system.