US 12,140,914 B2
Method for the closed-loop control of a chemical process in an industrial-scale chemical installation
Marcel Schulz, Bochum (DE); and Dirk Scheckreiter, Schwerte (DE)
Assigned to THYSSENKRUPP INDUSTRIAL SOLUTIONS AG, Essen (DE); and thyssenkrupp AG, Essen (DE)
Appl. No. 17/272,017
Filed by THYSSENKRUPP INDUSTRIAL SOLUTIONS AG, Essen (DE); and thyssenkrupp AG, Essen (DE)
PCT Filed Sep. 19, 2019, PCT No. PCT/EP2019/075193
§ 371(c)(1), (2) Date Feb. 26, 2021,
PCT Pub. No. WO2020/064506, PCT Pub. Date Apr. 2, 2020.
Claims priority of application No. 10 2018 123 792.8 (DE), filed on Sep. 26, 2018.
Prior Publication US 2021/0286347 A1, Sep. 16, 2021
Int. Cl. G05B 13/04 (2006.01); G05B 19/418 (2006.01); G06F 18/214 (2023.01); G06N 20/00 (2019.01); H04W 4/80 (2018.01)
CPC G05B 13/04 (2013.01) [G05B 19/41865 (2013.01); G06F 18/214 (2023.01); G06N 20/00 (2019.01); H04W 4/80 (2018.02)] 10 Claims
OG exemplary drawing
 
1. A method for closed-loop control of a chemical process performed in an industrial-scale chemical installation, the method comprising:
acquiring process data of the industrial-scale chemical installation with sensors, the industrial-scale chemical installation having at least one subsystem having the sensors;
transferring the process data to a control system via a fieldbus;
transferring at least a subset of the process data from the control system to a computer system, wherein the computer system comprises a simulation program for stationary and dynamic process simulation of the chemical process, a closed-loop control program for implementing a closed-loop controller for the chemical process, and a memory for storing simulated state variables;
cyclically, repeatedly calculating the simulated state variables of the chemical process by the simulation program from the at least the subset of the process data and storing the simulated state variables in the memory;
transferring a setpoint value of a control variable of the chemical process to the closed-loop control program;
reading at least a subset of the simulated state variables from the memory for input into the closed-loop control program;
ascertaining a manipulated variable to achieve the setpoint value by the closed-loop control program through processing the simulated state variables that have been read from the memory; and
transferring the manipulated variable that has been ascertained to the control system,
wherein the computer system includes a machine learning module, wherein training data sets are used to train the machine learning module, which training data sets include at least the subset of the simulated state variables and the manipulated variable calculated therefrom by the control program,
wherein the training data sets contain the response of the chemical installation to the manipulated variable contained in the process data,
wherein after training of the machine learning module, the control program is switched over to the machine learning module, wherein the closed-loop control program is replaced by the machine learning module, and so that the manipulated variable is determined from the read state variables by the machine learning module; entering the at least the subset of the simulated state variables into the machine-learning module to ascertain a machine-learning manipulated variable;
combining the machine-learning manipulated variable and the manipulated variable ascertained by the closed-loop control program to ascertain a resultant manipulated variable; and
transferring the resultant manipulated variable to the control system.