Research On Fault Diagnosis Method Based On Particle Filter Under Complicated Environment

With the rapid development of information and intelligent control technology,the scale of the modern industrial control systems become huger,its structure become more complex and the internal relation is more closely,and the requirements to reliability and safety are increasingly higher.Fault detection technology can provide methods and channels to ensure the normal operation of the control system and detect faults in time after a failure,and it already turns into an indispensable technology to ensure the normal operation of the system.But the actual industrial control system has a certain extent of nonlinear characteristics.At present,the fault detection technology for nonlinear systems has not yet formed a perfect and mature method.Particle filter has a unique advantage in dealing with nonlinear system problems,because of that it is not constrained by the assumption of noise and model,which gradually has become a hot spot in the field of investigation.But the traditional particle filter algorithms exist some problems that particle degeneracy and sample impoverishment,and it was struggling to deal with unknown noise,which lead to a low accuracy of fault state estimation,and affect the detection accuracy.Therefore,this paper conduct a research on these problems,and the main research contents are as follows:Aiming at solving the problem that the lower estimation accuracy caused by the degeneracy and diversity weakening of the traditional particle filter in the fault state estimation.This paper researches on weight-jittered firefly algorithm and incomplete resampling method.Firstly,due to the strong search characteristic of the firefly algorithm,so weight-jittered firefly algorithm is used to guide the whole movement of the particles in the optimization process,which makes the particles tend to the true values around quickly and increases the estimation accuracy.Secondly,the incomplete resampling method is used to alleviate particle degradation and better maintain particles diversity.This paper research that the lower estimation accuracy caused by the divergence and failure of the nonlinear systems filtering under unknown noise.First of all,due to the Sage-Husa estimator that estimating the noise characteristics in real time directly,it uses the combination of the Sage-Husa estimator and unscented Kalman filter to achieve the real-time estimation of noise characteristics,and take advantage of the latest measurement information to produce a new proposal distribution function.Andthen this paper makes use of the weight-jittered firefly algorithm to optimize the particles into resampling process.The improved algorithm not only can effectively alleviate the degeneracy and increase the diversity of the particles,but also improve the fault state estimation accuracy under unknown noise.Aiming at proving the validity of the improved fault state estimation method based on particle filter,two groups of nonlinear system models that the one-dimension single-variable nonlinear systems and the three-tank DTS200 system are selected for this paper as research object.The fault detection experiment is carried by different fault models respectively,which the one-dimension single-variable nonlinear system is set to the abrupt faults model,and the three-tank DTS200 are set to the sensor and actuator faults models.Ultimately,simulation results prove that the improved method can effectively improve the accuracy of fault detection.