Calibration of a quality estimator for a laser cutting method
The invention relates to a calibration module for calibrating a real-time estimator, which is intended for quality estimation of a cutting method using a laser cutting machine, comprising: a load interface for loading a quality estimation result of the real-time estimator; a first processor, which is intended to provide a quality measurement result of the cutting edge of the workpiece, wherein the quality measurement result can be provided in particular by detecting measurement signals of a cutting edge of a finished cut workpiece by means of a measuring device; and wherein a second processor is intended to compare the loaded quality estimation result with the quality measurement result and based on the result: is intended to calculate a calibration data set for calibrating the real-time estimator an output interface (A) which is intended to output the calculated calibration data set.
1. A method for calibrating a real-time estimator model of a real-time estimator with a calibration data set generated by a calibration module that includes a calibration model, the method comprising steps of:
cutting a workpiece by a laser cutting machine performing a cutting method;
loading a quality estimation result generated by the real-time estimator model of cutting performed by the laser cutting machine, wherein the quality estimation result is determined on a basis of sensor signals input to the real-time estimator model, wherein the sensor signals are received from at least one sensor, wherein the at least one sensor is configured as an optical sensor;
detecting measurement signals of a cutting edge of a cut workpiece by a measuring device after cutting, wherein the measuring device is an optical measuring device;
providing a quality measurement result of the cutting edge of the workpiece, wherein the quality measurement result of the cutting edge is based on the detected measurement signals;
comparing the quality estimation result with the quality measurement result to generate a comparison result, wherein a position identifier is assigned and/or stored for the detected sensor signals and for the detected measurement signals, wherein the detected sensor signals are assigned to the detected measurement signals based on their position identifiers, and based on the comparison result, only if deviations that are determined exceed a preconfigurable threshold value, executing steps of:
calculating a calibration data set, by the calibration model, for calibrating the real-time estimator model, and
calibrating or readjusting the real-time estimator model using the calibration data set.
2. The method according to claim 1 , wherein the real-time estimator model is configured to execute an estimation algorithm, and wherein the estimation algorithm calculates the quality estimation result at least from the detected sensor signals received from the at least one sensor.
3. The method according to claim 2 , wherein the estimation algorithm of the real-time estimator model comprises a deep convolutional neural network.
4. The method according to claim 1 , wherein the detection of the sensor signals and/or the measurement signals are configured based upon an execution time and/or an execution type.
5. The method according to claim 1 , wherein metadata are assigned and/or stored with the detected sensor signals and/or the detected measurement signals.
6. The method according to claim 1 , wherein the calibration data set includes an offset which is applied to the quality estimation result generated by the real-time estimator model.
7. The method according to claim 1 , wherein the calibration data set is calculated specifically for a material and/or a cutting process.
8. The method according to claim 1 , wherein the calibration data set contains parameters of the calibration model that has been trained with the quality estimation results and the quality measurement results assigned to one another.
9. The method according to claim 8 , wherein the calibration model is selected or combined from: a regression model, a support vector machine, a random forest, a gradient boosted tree, an artificial neural network and/or in which an ensemble learning approach with a combination of different models is used.
10. The method according to claim 8 , wherein the calibration model is trained by using additional stored intermediate results of the real-time estimator model as training data, wherein the intermediate results of the real-time estimator model comprises activation values of neurons of a convolutional neural network, arranged in front of an output layer.
11. The method according to claim 10 , wherein the calibration model is trained using stored input data of the real-time estimator model and assigned the quality measurement results as training data.
12. The method according to claim 8 , wherein the calibration model and/or the real-time estimator model is adapted using an action-reward learning algorithm.
13. The method according to claim 8 , wherein the calibration model and/or the real-time estimator model is adapted to new materials by means of a transfer learning method.
14. The method according to claim 1 , wherein the detection of the sensor signals of the cutting edge is carried out on a calibration cut or without the calibration cut.
15. The method according to claim 1 , wherein the detection of the measurement signals by the measuring device takes place via a deflection mirror.
16. The method according to claim 1 , wherein the quality measurement result is provided by an input data set via a user interface.
17. The method according to claim 1 , wherein execution of the real-time estimator model and the steps of loading, providing, and comparing are carried out after a calibration cut has been performed.
18. The calibration module, comprising a first processor and a second processor, for calibrating the real-time estimator model, wherein the calibration module is configured to execute the method according to claim 1 , the calibration module further comprising:
a load interface, which includes a processor, for loading the quality estimation result generated by the real-time estimator model;
the measuring device, which is configured for detecting the measurement signals of the cutting edge of the cut workpiece;
wherein the first processor is configured to provide the quality measurement result of the cutting edge of the workpiece using the detected measurement signals received from the measuring device, wherein the quality measurement result is provided by means of the measuring device;
wherein the second processor is configured to compare the quality estimation result with the quality measurement result, wherein the position identifier is assigned and/or stored for the detected sensor signals and for the detected measurement signals, via which the detected signals are clearly identifiable and are assigned to one another for the comparison step, and based on the result, only if deviations that are determined exceed the preconfigurable threshold value:
the second processor is configured to calculate a calibration data set for calibrating the real-time estimator; and
an output interface, wherein the calculated calibration data set is output via the output interface and used to calibrate or readjust the real-time estimator model.
19. A system with the real-time estimator model and the calibration module according to claim 18 to be used for a laser cutting machine.
20. The system according to claim 19 , wherein the system comprises or is in data communication with the at least one sensor.