Academia.eduAcademia.edu

Figure 8 – uploaded by Soundar Ramchandran

See full PDFdownloadDownload figure
Figure 8 a, Response of the overhead and bottom product compositions for setpoint changes and disturbances for Column A with the neural network model-based controller; b, variations in the feed flowrate and feed composition for column A; ¢, changes in the reflux rate and boilup rate as determined by the neural network model-based controller for column A; d, steady-state targets for the overhead and bottom product composi- tions calculated by the neural network model-based controller for column A

Figure 8 a, Response of the overhead and bottom product compositions for setpoint changes and disturbances for Column A with the neural network model-based controller; b, variations in the feed flowrate and feed composition for column A; ¢, changes in the reflux rate and boilup rate as determined by the neural network model-based controller for column A; d, steady-state targets for the overhead and bottom product composi- tions calculated by the neural network model-based controller for column A

Related Figures (22)

Figure 1 Feedforward neural network architecture
Figure 1 Feedforward neural network architecture
Table 1 Design and operating conditions for the two distillation columns ent methanol-water distivation columns: one Is a seven stage column, and represents an experimental system at Texas Tech University’, and the other represents a 45- tray high-purity industrial-scale distillation column. The lab-scale column is an atmospheric column, and produces products with approximately 4-5% impurity in the overhead product, and 1-2% impurity in the bottom product. The reboiler hold-up in the lab-column is roughly 30 times more than that in the condensate receiver, and, hence, creates vastly different dynamics at the two ends. We will call it Column A. By contrast, the high-purity column is an_ industrial-scale column producing products with less than 1000 parts per million impurity, and is typical of the ‘refining’ column in a three-column industrial methanol separation process***4. We will call it Column B. These two cases will enable us to perform control studies to evaluate the use of neural network models in a process model-based control environment.
Table 1 Design and operating conditions for the two distillation columns ent methanol-water distivation columns: one Is a seven stage column, and represents an experimental system at Texas Tech University’, and the other represents a 45- tray high-purity industrial-scale distillation column. The lab-scale column is an atmospheric column, and produces products with approximately 4-5% impurity in the overhead product, and 1-2% impurity in the bottom product. The reboiler hold-up in the lab-column is roughly 30 times more than that in the condensate receiver, and, hence, creates vastly different dynamics at the two ends. We will call it Column A. By contrast, the high-purity column is an_ industrial-scale column producing products with less than 1000 parts per million impurity, and is typical of the ‘refining’ column in a three-column industrial methanol separation process***4. We will call it Column B. These two cases will enable us to perform control studies to evaluate the use of neural network models in a process model-based control environment.
Figure 3. Predictions for reflux rate from the neural network model for column B Figure 2. Predictions for reflux rate from the neural network model for column A
Figure 3. Predictions for reflux rate from the neural network model for column B Figure 2. Predictions for reflux rate from the neural network model for column A
Figure 4 Open-loop response for the overhead and bottom product compositions for a +10% change in boilup rate for column A
Figure 4 Open-loop response for the overhead and bottom product compositions for a +10% change in boilup rate for column A
Figure 5 Process disturbances affecting column A during the +10% change in boilup rate
Figure 5 Process disturbances affecting column A during the +10% change in boilup rate
Figure 6 Open-loop response for the overhead and bottom product compositions for a +1% change in boilup rate for column B
Figure 6 Open-loop response for the overhead and bottom product compositions for a +1% change in boilup rate for column B
Table 5 Relative gain array for column B using average gains Control strategy
Table 5 Relative gain array for column B using average gains Control strategy
Table 4 Relative gain array for column A using average gains
Table 4 Relative gain array for column A using average gains
Table 3 Process gains for overhead and bottoms compositions for column B
Table 3 Process gains for overhead and bottoms compositions for column B
Figure 7 Schematic for the neural network model-based control strategy
Figure 7 Schematic for the neural network model-based control strategy
Figure 9 a, Response of the overhead and bottom product composi- tions for setpoint changes and disturbances for column A with the conventional feedback PI with feedforward controller; b, changes in the reflux rate and boilup rate as determined by the conventional feed- back PI with feedforward controller for column A
Figure 9 a, Response of the overhead and bottom product composi- tions for setpoint changes and disturbances for column A with the conventional feedback PI with feedforward controller; b, changes in the reflux rate and boilup rate as determined by the conventional feed- back PI with feedforward controller for column A
* Valve travel for the manipulated variable (reflux rate, L and boilup rate, V) Table 6 Comparison of controller performances for column A the approach is not the most optimal, it reflects the industrial practice in tuning controllers on-line and is simple to implement.
* Valve travel for the manipulated variable (reflux rate, L and boilup rate, V) Table 6 Comparison of controller performances for column A the approach is not the most optimal, it reflects the industrial practice in tuning controllers on-line and is simple to implement.
Figure 10 a, Response of the overhead and bottom product composi- tions for setpoint changes for column B with the neural network model-based controller; b, changes in the reflux rate and boilup rate as determined by the neural network model-based controller for column B
Figure 10 a, Response of the overhead and bottom product composi- tions for setpoint changes for column B with the neural network model-based controller; b, changes in the reflux rate and boilup rate as determined by the neural network model-based controller for column B
Figure 11 a, Response of the overhead and bottom product compo- sitions for setpoint changes and disturbances for column B with the conventional feedback PI with feedforward controller; b, changes in the reflux rate and boilup rate as determined by the conventional feed- back PI with feedforward controller for column B
Figure 11 a, Response of the overhead and bottom product compo- sitions for setpoint changes and disturbances for column B with the conventional feedback PI with feedforward controller; b, changes in the reflux rate and boilup rate as determined by the conventional feed- back PI with feedforward controller for column B
Figure 12 a, Response of the overhead and bottom product compo- sitions for feed flowrate changes for column B with the neural network model-based controller; b, changes in the reflux rate and boilup rate as determined by the neural network model-based controller for column B
Figure 12 a, Response of the overhead and bottom product compo- sitions for feed flowrate changes for column B with the neural network model-based controller; b, changes in the reflux rate and boilup rate as determined by the neural network model-based controller for column B
* Valve travel for the manipulated variable (reflux rate, L and boilup rate, V) Table 7 Neural network model-based controller performance for column B For satisfactory control of any process it is essential that the model used to infer the control action be able to predict the process gains with sufficient accuracy®. Figure 15 shows the process gains for the reflux rate with respect to the overhead product composition, AL/Axp. Similar behaviour was observed for the process gain for the boilup rate with respect to the bottom product composition, AV/Ax,. Despite the fact that the ‘process’ and the ‘model’ are distinctly different from each other, the model is able to describe the process
* Valve travel for the manipulated variable (reflux rate, L and boilup rate, V) Table 7 Neural network model-based controller performance for column B For satisfactory control of any process it is essential that the model used to infer the control action be able to predict the process gains with sufficient accuracy®. Figure 15 shows the process gains for the reflux rate with respect to the overhead product composition, AL/Axp. Similar behaviour was observed for the process gain for the boilup rate with respect to the bottom product composition, AV/Ax,. Despite the fact that the ‘process’ and the ‘model’ are distinctly different from each other, the model is able to describe the process
Figure 13 Response of the overhead and bottom product composi- tions for feed flowrate changes for column B with the conventional feedback PI with feedforward controller be required to achieve the steady-state condition defined by the values of F, z, xp and x,. The predicted values for L and V were then used in the dynamic simulators along with values for F and z. The dynamic simulators were allowed to run until ‘near’ steady-state conditions were identified, and the steady-state values for x, and Xp were noted. The fact that the ‘process’ (the dynamic simulators) and the ‘model’ (the neural networks trained using CAD data) are different is illustrated in Figure 14, which compares the steady-state values of the overhead product composition obtained from the dynamic process simulator for Column A with that used in the neural network model for Column A.
Figure 13 Response of the overhead and bottom product composi- tions for feed flowrate changes for column B with the conventional feedback PI with feedforward controller be required to achieve the steady-state condition defined by the values of F, z, xp and x,. The predicted values for L and V were then used in the dynamic simulators along with values for F and z. The dynamic simulators were allowed to run until ‘near’ steady-state conditions were identified, and the steady-state values for x, and Xp were noted. The fact that the ‘process’ (the dynamic simulators) and the ‘model’ (the neural networks trained using CAD data) are different is illustrated in Figure 14, which compares the steady-state values of the overhead product composition obtained from the dynamic process simulator for Column A with that used in the neural network model for Column A.
Neural network control of distillation: S. Ramchandram and R. R. Rhinehart
Neural network control of distillation: S. Ramchandram and R. R. Rhinehart
Figure 14 Comparison of the steady-state values for the overhead product composition obtained from the dynamic process simulator and that used in the neural network model for column A * Valve travel for the manipulated variable (reflux rate, L and boilup rate, V)
Figure 14 Comparison of the steady-state values for the overhead product composition obtained from the dynamic process simulator and that used in the neural network model for column A * Valve travel for the manipulated variable (reflux rate, L and boilup rate, V)
Figure 15 Process gain for the reflux rate with respect to the over- head product composition for column A
Figure 15 Process gain for the reflux rate with respect to the over- head product composition for column A
Related topics:
Chemical EngineeringDistillation columnsDynamic SimulationProcess SimulationPI ControllerReference SystemsDynamic SimulationsNeural network control
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved