Fault Detection And Diagnosis Of Dividing Wall Column Based On Convolutional Neural Network

In recent years,divided wall column has received more and more attention due to its advantages of economy and energy consumption.However,due to its more complex and higher coupling than the traditional distillation column,it is accompanied with great security risks in the industrial application of the divided wall column.When the design,optimization and dynamic control of the process of the adjacent distillation column and the reactive distillation column are carried out,the research on fault diagnosis in the operation process of the distillation column is essential.Fault diagnosis and detection can reduce the occurrence of accidents in chemical production process,improve the economy and safety of chemical process,and provide guidance for the safe operation of distillation columns in chemical plants.In this thesis,three-component separation process of n-hexane / n-pentane / n-heptane and synthesis process of isobutyl acetate by transesterification were taken as examples to study the fault diagnosis of the divided wall distillation column and the reactive distillation divided wall column.Firstly,software Aspen was used to establish the steady-state process model,which was compared with the economy of a conventional distillation sequence.The dynamic control structure is established based on the optimal steady-state design,and the control structure with better control properities is screened based on disturbance analysis method.Then,the process faults of the distillation column are simulated based on the control structure with better control properities,and the trend of the selected detection variables is analyzed,and the dynamic safety under different faults is analyzed.Finally,the convolutional neural network model based on MATLAB platform is used to diagnose and detect different types of faults.DWDC process can save 21.68%,16.25% TAC,33.01%,25.69% energy consumption,RDWDC can reduce 26.92% energy consumption,26.586% TAC compared with conventional distillation separation process and conventional reactive distillation process.The feasible dynamic control structure was designed based on the optimized steady-state model.The control properities was quantified by IAE and Abs Error,and the control properities of the four-point temperature control structure of the divided wall column was slightly better than that of the three-point temperature control structure.For reactive divided wall column,the control structure with tray temperature control feed flow has better anti-interference ability.In the fault simulation based on optimal control structure,the condenser failure at the top of the tower has the greatest impact on the safety of the process.The dynamic changes of selected characteristic variables are closely related to the control structure.Among the five faults simulated by the divided wall distillation column,five CNN models were builted for fault diagnosis.The CNN neural network based on the convolution pooling layer of two layers alternately has good fault diagnosis performance,and the accuracy of DWDC process fault diagnosis can reach 90.3% of iteration number 6000.The accuracy of RDWDC fault diagnosis can reach 91.6 % when the number of iterations reaches 7000.The research results show that the model can have a good application prospect in the fault diagnosis of the divided wall distillation column and the reactive distillation dividing wall column.