Fault Diagnosis And Predictive Maintenance Based On System Identification

Fault diagnosis is a technology which analyzes the running state of the system,detects,separates and evaluates the abnormal conditions,and provides guidance for fault recovery.Predictive maintenance is a task type which analyzes the change trend of system status through algorithm,and implements active maintenance measures before the fault brings to great loss,so as to improve the service life of equipment.With the widespread application of DCS in the industrial process,more and more production data can be collected and saved.Using these data to make accurate fault diagnosis and to implement early maintenance can save production costs,improve product quality and ensure process safety.However,due to the characteristics of high complexity,strong nonlinearity and time-varying in modern process industry,traditional fault diagnosis methods have become more and more difficult to meet the requirements.On the basis of previous research work,aiming at common fault problems,the thesis proposes some method for fault diagnosis and predictive maintenance based on system identification,which are summarized as follows:(1)A fault detection and isolation method based on residuals of identified model is proposed.Under normal operating conditions,the input and output data of system are used to identify the dynamic model of the process.Then,the residual error is used to detect whether there is a fault in the system and separate the input fault,the output fault,or the change of process parameter under the online monitoring operating condition.Residuals are based on the results of consistency tests between process output observations and model outputs.The thresholds for model residuals are set according to the distribution of residual statistics under normal operating conditions.Finally,the effectiveness of the proposed approach is illustrated by examples(2)A two-step identification method based on mechanism model is proposed for the nonlinear characteristics of industrial control valves.Firstly,the nonlinear characteristics of the control valve are systematically described with a small number of parameters according to the input-output relationship.Then the high order ARX model is used to describe the linear dynamic process,and the parameters to be estimated for the nonlinear part of the control valve are obtained by traversing,which aims to minimize the output error.The output signal of valve position is reconstructed,and the model parameters of the linear dynamic process are estimated by a low order Box-Jenkins model.Finally,the consistency of the algorithm is proved mathematically,and the effectiveness of the proposed approach is illustrated by examples.(3)An iterative identification method based on phenomenon model is proposed for the nonlinear characteristics of industrial control valves.the nonlinear input-output relationship is described as a linear combination of known basis functions by using a form of cubic spline,multi-relay or multi-backlash structure.An iterative identification algorithm is proposed to estimate the weight of the nonlinear basis function and the model parameters of the linear dynamic process based on the theory of the asymptotic identification,which minimizes the prediction error approximately.The iteration procedure is divided into two steps,each step solves a least square problem,and an unitization technique is introduced to improve the accuracy.Finally,the effectiveness of the proposed approach is illustrated by examples.(4)A method for fault diagnosis and predictive maintenance based on model confidence interval is improved for the change of process parameter.The dynamic model of the process is identified by the input and output data of the system on a regular basis,and the corresponding confidence interval is calculated.Constantly compare the process model under the online operating condition with the process model under the normal operating condition,evaluate the status of the system through some specific performance indexes such as the step response and frequency response of the model,and make maintenance warning when the change of process parameter reaches a certain degree.Finally,the effectiveness of the proposed approach is illustrated by examples.Finally,on the basis of summing up the whole thesis,the contents of the study are summarized in detail,and the future research and development direction are illustrated.