Research On Fault Detection Methods Based On Subspace Identification

Fault detection techniques play an extremely important role in improving system safety and have matured over decades of development.Model-based fault diagnosis methods are more effective when models are easy to build or are more accurate,but models of real complex systems are difficult to build or are not as accurate.Multivariate statistical analysis methods based on data,such as Principal Component Analysis(PCA),Canonical Variate Analysis(CVA)can achieve good results under ideal steady-state conditions,but are not as effective when applied to dynamic production processes.Fault diagnosis methods that only consider model or data information have certain drawbacks.Therefore,the study of subspace identification methods that take into account both the process data and the known information of the model is of great theoretical and practical importance to improve the reliability and safety of industrial processes.In this paper,fault detection based on the Subspace identification method(SIM)is studied.Different from the traditional subspace identification method,which needs to identify the model parameters A,B,C,D,and then use the model-based method for fault detection,the subspace identification fault detection method used in this paper can use the input and output data to identify the parameters required by the residual matrix in one step,which reduces the calculation steps and reduces the complexity of the calculation compared with the traditional subspace identification fault detection method.After obtaining the residual and statistics,the chi-square distribution lookup table is used to obtain the threshold,and the failure occurs by comparing the statistics and the threshold.Build an event trigger mechanism to effectively reduce the consumption of communication resources.Design moving window algorithm to solve the problem that small faults are difficult to detect.The subspace identification algorithm is optimized by performance indicators to solve the problem of insensitivity between actuator failure and sensor fault.The main research contents of this paper are summarized as follows:First,a fault detection method based on moving window subspace identification is proposed to address the difficulty in detecting small faults.The optimal window length is obtained by using the Sparrow Search Algorithm(SSA)to achieve the maximum fault detection rate.To improve the utilization of communication resources,an event-triggered strategy is applied to subspace identification fault diagnosis,effectively reducing the loss of communication resources.The effectiveness of the proposed method is verified by simulation of the Tennessee Eastman(TE)process.Then,a subspace identification fault detection method based on performance indicators is proposed to solve the problem of insensitivity to sensor failure and actuator fault.Based on this,the moving window algorithm is optimized to solve the problem that small faults are difficult to detect.The effectiveness of the algorithm is verified by the Traction Drive Control System(TDCS).Finally,the main work done in this paper is summarised and the problems and future development trends of subspace identification fault detection algorithms are pointed out,as well as the outlook for future research work.