Research On System Monitoring Method Based On Multivariate Statistics And Information Entropy

In the 21 st century,the rapid development of science and technology greatly pushes forward the modern industrial system towards the intelligent direction.industrial systems are becoming more and more integrated and more complex,furthermore,easily affected by the working environment,device aging or human error,it is likely to cause great economic losses and serious safety problems.System monitoring plays an important role in improving the reliability of system operation and reducing the rate of accident.Based on above mentioned,this thsis studies fault detection and performance monitoring of closed-loop control system.The main research contents of the thesis are as follows:1.The commonly used principal component analysis method does not consider the correlation between sampled data when constructing the principal component model.Moreover,it can not integrate the system dynamic characteristics into the principal component model.A dynamic principal component analysis method is proposed in this work.The fault detection method restores the dynamic characteristics of the system itself by constructing the sampled data into a dynamic time series augmentation matrix.The dynamic time series augmented matrix is used for the projection dimension reduction.Then the statistical value obtained by the projection is compared with the control limit established by the normal data.The fault will occur when projection statistics beyond the limit,thereby realizing the significant fault detection.2.Aiming at the characteristics of small faults and insignificant features in industrial systems,a incipient fault detection method based on dynamic principal component analysis and KullbackLeibler Divergence is proposed.Through the constructed PCA model incorporating dynamic characteristics,the model score vector is obtained,and the probability distribution of the score vector is calculated.Then the similarity between the probability distributions is quantized by KLD,the corresponding statistical limit is established according to the 3? rule.The KL divergence value is compared with the statistical limit,and the over-limit is a fault.In order to verify the effectiveness of the proposed method,the method is applied to the Tennessee Eastman Process(TEP).The results show that the proposed method can effectively detect small faults.3.In allusion to the phenomenon that the performance of the system control loop will gradually deteriorate with the passage of time,a new control loop performance monitoring index is proposed in this paper.By means to using the time series generated by the controller deviation,the weighted permutation entropy is calculated to monitor the performance of the control loop.The simulation results show that the proposed performance monitoring method has better performance than the method based on permutation entropy.