METHOD FOR CONFIGURING A CONTROL SYSTEM FOR A PROCESS PLANT

Abstract

A method for configuring a control system for a process plant using a dynamic model of the process plant, the dynamic model being based on at least one of thermo fluidic correlations, thermo dynamic correlations, phenomenological correlations, and equations, and being based on geometry and/or topology of components of the process plant, the dynamic model receiving process parameters as input values, the dynamic model being adapted to represent a transition from one to another state of the process plant and the dynamic model covering the entire operating range of the process plant wherein the dynamic model is used in an offline mode, in which the dynamic model is used in stand-alone fashion, wherein, based on input and output values of the dynamic model, a behaviour of the process plant is predicted, and wherein, based on the predicted behaviour of the process plant, the control system is configured.

Claims

1-11. (canceled)

12. A method for configuring a control system for a process plant, using a dynamic model of the process plant, the dynamic model being based on at least one of thermo fluidic correlations, thermo dynamic correlations, phenomenological correlations, and equations, and being based on geometry and/or topology of components of the process plant, the dynamic model receiving process parameters as input values, the dynamic model being adapted to represent a transition from one to another state of the process plant, and the dynamic model covering the entire operating range of the process plant, wherein the dynamic model is used in an offline mode, in which the dynamic model is used in stand-alone fashion, wherein, based on input and output values of the dynamic model, a behaviour of the process plant is predicted, and wherein, based on the predicted behaviour of the process plant, the control system is configured.

13. The method according to claim 12, wherein the dynamic model is a pressure-driven model.

14. The method according to claim 12, wherein, based on the predicted behaviour of the process plant, parameters of a controller of the control system are configured and/or tuned.

15. The method of claim 14, wherein the parameters of a controller of the control system are configured and/or tuned prior to use of the control system to control the process plant.

16. The method according to claim 12, the control system using a controller being based on model predictive control.

17. The method according to claim 16, wherein a model for the model predictive control is deduced from the dynamic model, based on the behaviour of the process plant.

18. The method according to claim 14, wherein a linear parameter-varying system is deduced from the dynamic model, based on the behaviour of the process plant.

19. The method according to claim 12, wherein the process plant includes at least one of a gas processing plant, an air separation unit, a natural gas plant, an ethylene plant, a hydrogen plant and an adsorption plant.

20. The method according to claim 12, wherein, in the offline mode, the dynamic model further is used without any online connections to the control system of the plant or history about a plant.

21. The method according to claim 12, wherein the entire operating range of the process plant includes the following operating phase of the plant: start-up, regular operation, and shut-down, and, preferably, further include plant failure and/or emergency shut-down.

22. A computing unit, configured, preferably by means of a computer program, to perform a method according to claim 12.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0055] FIG. 1 schematically shows a dynamic model used within the invention in correlation to a process plant.

[0056] FIG. 2 schematically shows a method according to the disclosure in a preferred embodiment.

[0057] FIG. 3 schematically shows process parameters used within the method shown in FIG. 2.

[0058] FIG. 4 schematically shows a method according to the disclosure in a further preferred embodiment.

[0059] FIG. 5 schematically shows a method according to the invention in a further preferred embodiment.

DETAILED DESCRIPTION OF THE DRAWINGS

[0060] In FIG. 1, a dynamic model 200 used within the invention in correlation to a process plant 100 is shown. Exemplarily, the process plant 100 is formed as an air separation unit or plant.

[0061] The process plant 100 includes, among other components, a main air supply 1, a main heat exchanger 5, an expansion turbine 6, a pump 8, a high pressure column 11 and a low pressure column 12. High pressure column 11 and low pressure column 12 as well as crude argon column 13 and pure argon column 14 and associated heat exchangers, are part of a typical distillation column system.

[0062] In the air separation unit, an air feed stream, supplied via main air supply 1, is typically compressed, pre-cooled and cleaned. Further, the air stream is separated in two streams, one of which is fully cooled in the main heat exchanger 5, the other one is only partly cooled in the main heat exchanger 5. The latter one is then expanded by means of expansion turbine 6. The fully cooled air stream is supplied to the high pressure column 12 via the heat exchanger 7, and the partly cooled air stream is supplied to the low pressure column 11.

[0063] Further, a control system in the form of a distributed control system is provided for the process plant 100. Such a distributed control system includes several controllers (in particular, PID controllers) separately arranged at specific positions within the process plant 100. Such controllers include flow controllers FIC, pressure controllers PIC, liquid level controllers LIC and temperature controllers TIC. Some of those controllers are shown for the plant 100 in FIG. 1.

[0064] The dynamic model 200 is based on thermo fluidic equations 210 (e.g., thermos fluidic mechanical (cubic) equations of state) and/or equations of state and/or physical properties 220 and material and components used therein. Further, the dynamic model 200 is based on geometry and/or topology of components of the process plant 100 and on different design correlations for gas and liquid hold up, for pressure drops 230, and for heat transfer coefficients 240. Further, the dynamic model 200 is preferably based on material constants like metal heat capacity 250. This model represents the geometry and topology of equipment in the same level of detail that is being used in process design.

[0065] Design correlations are used to derive a mechanical design from thermodynamic process data. Design correlations can be such as heat transfer (which depends on the actual design of the equipment such as heat transfer area, fin sizes, stream pattern, etc.) between different media inside a heat exchanger or heat and mass transfer on a sieve tray inside a distillation column (which depend e.g. on phase boundary area, fluid velocities due to the actual design, etc.). In other words, such design correlations can be equations for calculating, e.g., heat transfer in a heat exchanger or columns.

[0066] Heat exchangers are represented, in the dynamic model 200, in one space dimension, i.e., in 1-D spatial resolution, denoted by reference numeral 260, as shown, e.g., for the main heat exchanger, whereas distillation columns are modelled as a sequence of trays or comparable simulation approaches. This approach enables the use of design correlations for pressure drops, heat transfer coefficients and so on. However, unlike the equipment design tools, the first principles dynamic model models holdups for gas, liquid and heat and hence allows for modelling transient plant states.

[0067] The dynamic model 200 is also adapted to receive controller set points of process parameters as input values, in particular, it includes base layer control 270 like for flow, pressure and temperature (PID) controllers as mentioned with respect to the process plant above. Exemplarily, the correlation to two of those controllers is shown. This implies that (PID) controller set points are inputs to the dynamic model 200 instead of valve positions.

[0068] In FIG. 2, a method according to the disclosure in a preferred embodiment is shown. The dynamic model 200, described above, is used during operation of the process plant 100, for which the dynamic model 200 is made.

[0069] For operating the process plant 100, a control system 300 in the form of a distributed control system including (at least) individual, separately arranged controllers for temperature, flow and liquid level as shown in FIG. 1 and described above. For sake of clarity, these individual controllers are not shown in FIG. 2. The dynamic model 200 is operated or executed on a computing unit 400 which can be part of the control system 300 or a separate computer. This kind of operation is an aspect of the online mode mentioned before.

[0070] In the method, signals from the control system 300 are received and fed into the dynamic model 200. The signals represent values of at least one first process parameter, indicated with reference numeral 310. This at least one first process parameter includes, e.g., valve positions, and, in particular, flow, pressure and temperature set points for the individual controllers mentioned before. The values of these first parameters can include (currently) measured data and/or historic data.

[0071] Based on and using the dynamic model 200, values of at least one second process parameter are determined. This at least one second process parameter includes, in the embodiment shown in FIG. 2, a column profile 410.

[0072] Typical column profiles and the process parameter to be used or determined within this method are shown in FIG. 3. In particular, FIG. 3 shows concentration profiles v1, v2, and v3 for different values of a manipulated variable MV (MV1, MV2, MV2), e.g. air flow. These profiles show, e.g., the mole fractions y of the component of interest as a function of the packing height or stage number x in a distillation column.

[0073] For the purpose of process control, a location (e.g. packing height) corresponding to a characteristic concentration of the component of interest (e.g. y*) is chosen as the controlled variable. The parameters x1, x2 and x3 are these locations corresponding to the values of the manipulated variables MV1, MV2, and MV3.

[0074] This allows significantly improved control performance of or compared to using a (single) temperature measurement. Temperature measurement locations are typically selected during process design. Here, the goal is to select a location that provides large temperature sensitivity with respect to the manipulated variables such as air or column reflux flow. The temperature profile in the plant, however, can deviate to a certain extent from the design profile and/or it can be strongly dependent on the operating case. Therefore, a single temperature measurement may be problematic or insufficient for control purposes.

[0075] This new controlled variable or parameter can directly be taken from the composition profiles estimated by the dynamic model. The controlled variables “profile location” can be used within SISO controllers, or within (in particular, linear or non-linear) MPC approaches. For example, such a model predictive controller 320 which is to be supplied with the values or set points obtained from the dynamic model is shown. Such model predictive controller could include one, several or all of the individual controllers of the control system 300 mentioned above.

[0076] This offers the following advantages over the typical temperature control approach: The controlled variable does not depend on a single temperature measurement but is based on all the input signals to the dynamic model 200. A correction with respect to pressure is no longer necessary and the dynamic response of the control variable's “profile position” with respect to manipulated variables is essentially linear, as changes in the typical manipulated variables (air, reflux flow) simply move the column profiles up or down along the column height. Therefore, the linear controllers based on these controlled variables can be tuned much more aggressively, reacting faster to disturbances or set point changes. Tighter control of the column profile in the low pressure column of the air separation unit allows for operation with lower air flow, and consequently, a lower required compressor power.

[0077] In FIG. 4, a method according to the disclosure in a further preferred embodiment is shown schematically. The dynamic model 200, described above, is used during operation of the process plant 100, for which the dynamic model 200 is made.

[0078] For operating the process plant 100, a control system 300 in the form of a distributed control system including (at least) individual, separately arranged controllers for temperature, flow and liquid level as shown in FIG. 1 and described above. For sake of clarity, these individual controllers are not shown in FIG. 4. The dynamic model 200 is operated or executed on a computing unit 400 which can be part of the control system 300 or a separate computer. This kind of operation is also an aspect of the online mode mentioned before.

[0079] In the method, signals from the control system 300 are received and fed into the dynamic model 200. The signals represent values of at least one first process parameter, indicated with reference numeral 310. This at least one first process parameter includes, e.g., valve positions, and, in particular, flow, pressure and temperature set points for the individual controllers mentioned before. The values of these first parameters can include (currently) measured data and/or historic data.

[0080] Based on and using the dynamic model 200, values of at least one second process parameter are determined. This at least one second process parameter includes, in the embodiment shown in FIG. 4, a temperature profile 411 for, e.g., the main heat exchanger. This temperature profile 411 is then used, within operating the process plant, for estimating or determining a lifetime 430 of that main heat exchanger. This can be performed, e.g., by means of a machine learning or AI model 420. Such model is, e.g., disclosed in WO 2019/015805 A1.

[0081] Temperature profiles in (plate-fin) heat exchangers are a key towards health monitoring of such heat exchangers and estimating remaining life time. Using a machine learning approach, stress levels are estimated from temperature profiles, where the data set for learning is obtained, e.g., from detailed process and FEM modelling. The stress estimation hinges upon the availability of accurate temperature profiles, which are typically not available for heat exchangers in production facilities. Detailed temperature profiles are provided by the dynamic model. As outlined above, the predictions of the dynamic model can be regularly aligned with the available temperature and other measurements by applying data reconciliation methods.

[0082] In typical cases, there is a high sensitivity on the simulated temperature profiles with respect to the process input (e.g. small deviations in flow could have significant impact on temperature profiles). Also, direct measurement of metal temperatures (by means of, e.g., smart equipment) can be used. Nevertheless, for specific applications it is possible to estimate lifetime consumptions based on simulated temperature profiles (if heat exchanger profiles are one dimensional).

[0083] In FIG. 5, a method according to the invention in a further preferred embodiment is shown schematically. The dynamic model 200, described above, is used to configure the control system 300, which afterwards can be used to operate the process plant 100. It is to be noted that the (physical) process plant 100, although shown in FIG. 5, is not necessary for performing the method.

[0084] The control system 300 includes, as also mentioned with respect to FIGS. 2 and 4, controllers which, in turn, can be linear or non-linear model predictive controllers, exemplarily shown with reference numeral 320. Such model predictive controller is based on a model of the process to be controlled, the model of the model predictive controller denoted with reference numeral 325.

[0085] In the method, based on input values and output values of the dynamic model 200, a predicted behaviour of the process plant is predicted. Then, based on the predicted behaviour of the process plant, the control system 300 is configured.

[0086] The input values 510 can include, e.g., process parameters like step changes for air flow, products, flows, reflux and the like. Corresponding output values 520 can include, e.g., step responses for temperatures, product purities and the like.

[0087] Based on these output values 520, parameters of the model predictive controller 320 of the control system 300 can be configured and/or tuned. Also, based on these output values 520, the model 325 for the model predictive controller 320 can be deduced from the dynamic model 200.

[0088] The dynamic model 200 (in the offline mode) is ideally suited for the pre-configuration of advanced process control systems such as linear model predictive control (LMPC): Dynamic relationships between selected inputs and outputs can easily be identified from the dynamic model, independently of the plant operation. Step tests on the real plant can be planned and conducted efficiently using the dynamic model. With this approach, disturbances that might be present in the plant are not affecting the step responses, and production losses or interruptions due to on-going tests on the plant can be avoided.

[0089] Performing the step test on the dynamic model translates into significant time and cost savings, as the test can be performed much faster than real time, and without the need for on-site travel. Since the LMPC can be configured even before the plant is started up, the advanced control system will be available from the first day of the plant operation. This in turn can reduce overall commissioning time.

[0090] This pre-configuration of model-based controllers is not limited to LMPC. In a similar manner to the steps required in setting up an LMPC, a piecewise linear MPC or a fully nonlinear MPC strategy (NMPC) can be pre-configured based on the dynamic model. The advantage of using the dynamic model for the configuration of a nonlinear controller is even more significant than for the linear case, as more rigorous and more detailed system response testing is required to set up a fully nonlinear model with a large range of validity. It can be expected that such extensive testing would be extremely time-consuming on the actual plant, if feasible at all.