US 12,285,866 B2
Method and device for ascertaining control parameters in a computer-assisted manner for a favorable action of a technical system
Martin Bischoff, Großhelfendor (DE); and Michel Tokic, Tettnang (DE)
Assigned to SIEMENS AKTIENGESELLSCHAFT, Munich (DE)
Appl. No. 17/256,128
Filed by Siemens Aktiengesellschaft, Munich (DE)
PCT Filed Jun. 19, 2019, PCT No. PCT/EP2019/066145
§ 371(c)(1), (2) Date Dec. 24, 2020,
PCT Pub. No. WO2020/002072, PCT Pub. Date Jan. 2, 2020.
Claims priority of application No. 18180479 (EP), filed on Jun. 28, 2018.
Prior Publication US 2021/0122038 A1, Apr. 29, 2021
Int. Cl. B25J 9/16 (2006.01)
CPC B25J 9/163 (2013.01) [B25J 9/1671 (2013.01)] 14 Claims
OG exemplary drawing
 
1. A method for ascertaining control parameters in a computer-assisted manner for a favorable action of a technical system, the method comprising:
detecting an initial state and a surrounding area of the technical system by at least one sensor;
generating a physical simulation model of the technical system and the surrounding area on the basis of sensor data detected by the at least one sensor, wherein the surrounding area and the initial state of the technical system are continuously detected by the at least one sensor and the physical simulation model is continuously updated by the sensor data;
adapting the physical simulation model based on the sensor data as input data;
ascertaining a specified target state for the technical system, wherein various combinations of action steps on the basis of the initial state are configured to lead to the specified target state;
simulating the various combinations of action steps of the technical system by the physical simulation model, on a basis of the initial state and with respect to the specified target state, control parameters of the technical system for carrying out the action steps being varied, and outputting respective resultant simulation data;
training a machine learning routine on a basis of the respective resultant simulation data and by an evaluation of a respective action step;
ascertaining an optimized combination of action steps on a basis of the trained machine learning routine, the optimized combination of action steps being assigned a favorable evaluation; and
outputting the control parameters of the optimized combination of action steps for controlling the technical system.