Data-driven Based Process Condition Monitoring And Fault Diagnosis

With the ongoing development of modern industry,control systems becoming more and more large and complex,so the factor of affecting system run has increased abruptly,and the probability of fault occurrence becomes higher and higher,the intelligent monitor and diagnosis technology emerges as the times require.Meanwhile,a lot of process data are kept in the process of production,the analysis and use of massive data has become more convenient under the development of computer technology.In this paper,we focus on process status monitoring and fault diagnosis based on statistical information of data,and put forward new random filtering and fault diagnosis methods with considering the non-Gaussian noises in industrial.The main work and contribution are as follows:First of all,particle filter inherited from the Bayesian estimation and the Monte Carlo method is applied to realizing fault detection and estimation in a concentric-pipe counter-flow heat exchanger in this paper.Secondly,this paper uses the survival information potential to describe the uncertainty of the stochastic system and constitute a new performance index.We introduce a new filter gain updating algorithm based on this performance index,so as to the shape of the estimation error distribution as narrow as possible.To guarantee that the estimation error is stochastically exponentially ultimately bounded in the mean square,a suboptimal stochastic stability filter gain updating law is used b ased on linear matrix inequality.The effectiveness of the proposed method is demonstrated by the simulation application of the manipulator.Finally,according to the main problem of traditional weighted least squares method may affected by the gross errors which could result in higher deviations and poorer state estimation results.A robust power system state estimation method is proposed here under generalized maximum correntropy criterion.The novel algorithms are numerically evaluated using the IEEE 14-bus benchmark,the simulation result indicate it can automatically suppress bad data and its calculation is fast,which lays a solid foundation for the further theoretical research and engineering practice.