Study On Modal Identification And Fault Monitoring Of Multimodal Chemical Process

Equipment wear,process adjustment,process load and the external environment and other factors might lead to a number of different chemical process conditions,which would change chemical process into the so-called multi-modal chemical process.In the multimodal chemical process,different modes have different process characteristics and statistical characteristics.Therefore,the traditional process monitoring model is no longer applicable,and many monitoring models need to be established according to the characteristics of different modes.However,it is a difficult problem to determine the number of modes from a number of process historical data and to accurately classify the process modality.There are special transition processes between adjacent stationary modes.Transition processes are characterized by time-varying,non-Gaussian and strong nonlinearity,which makes it difficult to establish accurate multi-mode chemical process fault monitoring model.In order to solve the above problems,this thesis focused on the above three characteristics of multimodal chemical process,and conducted the following three studies:(1)Aiming at the problem that the number of process modes was unknown under the premise of lack of prior knowledge,an adaptive multi-model fault detection method was proposed.The singular value decomposition method was used to solve the optimal clustering number,and the training data were trained by fuzzy C-means clustering algorithm.The modal soft division was carried out and the monitoring model of different stationary modes and transition modes was established by using the principal component analysis method to realize the process state monitoring.The method was applied to the propylene metering tank and compared with the method based on the validity index of the intra-cluster classification(BWP).The false positive rate was reduced by 5.59% and the false negative rate was almost 0.The results showed that this method could overcome the deficiencies of prior knowledge,accurately divide the modal and improve the accuracy of process monitoring.(2)The non-Gaussian and fixed-point independent component analysis algorithms for multimodal chemical processes had the disadvantages of high complexity and easy to fall into local extremum.A global optimal non-Gaussian multi-model fault monitoring method was proposed.The non-Gaussian variables were identified by lilliefors test normality.Fuzzy C-means clustering algorithm was used to classify the process modal.Particle swarm optimization was used to optimize the traditional independent component analysis algorithm.Non-Gaussian monitoring models with different process modal were established.monitor.In the case analysis,this method was applied to the propylene metering tank device.Compared with the fixed-point independent component analysis algorithm,the false alarm rate could be reduced by 2.42% and the false negative rate was controlled at about 0.74%.It was proved that this method can improve the accuracy of non-Gaussian variable fault monitoring in multimodal chemical process.(3)Aiming at the problem that the normal transient fluctuation of the process variable was misjudged as the process error caused by the static control limit in the traditional multivariate statistical monitoring method,a dynamic multi-point fault monitoring method was proposed.Stationary model,single-point monitoring statistic and multi-point anomaly statistics were constructed based on autoregressive model and non-Gaussian model.Dynamic control limit monitoring was adopted.Transition mode was used to monitor non-Gaussian statistic directly by dynamic control limit.The results showed that the false negative rate was less than 0.8%,and the false alarm rate was reduced by 6.33% compared with the 3? threshold value monitoring method,and controlled within 1.2%.