Method for controlling a laser processing operation by means of a reinforcement learning agent and laser material processing head using the same
The present invention relates to a method for controlling a processing operation of a workpiece by means of a Reinforcement Learning (RL) agent unit, comprising the steps of: (a) observing an interaction zone in the workpiece by means of at least one radiation sensor to generate at least one sensor signal s t , wherein the workpiece is processed using an actuator having an initial actuator value a t ; (b) determining a basis function Φ(s t ) from the set of sensor signals s t ; (c) determining a reward function r t giving the probability of good results of the processing operation; (d) choosing a next actuator value a t+1 on the basis of a policy π depending on the reward function r t and the basis function Φ(s t ); and (e) repeating the steps (a) to (d) for further time points to perform a RL controlled processing operation.
1. Method for controlling a processing operation of a workpiece by means of a Reinforcement Learning (RL) agent unit, comprising the steps of:
(a) observing an interaction zone in the workpiece by means of at least one radiation sensor to generate at least one sensor signal s t ; wherein the workpiece is processed using an actuator having an initial actuator value a t ;
(b) determining a basis function φ(s t ) from the set of sensor signals s t ;
(c) determining a reward function r t giving the probability of good results of the processing operation;
(d) choosing a next actuator value a t+1 on the basis of a policy π depending on the reward function r t and the basis function φ(s t ); and
(e) repeating the steps (a) to (d) for further time points to perform an R L controlled processing operation, and wherein steps (a) and (b) comprise the steps of:
recording a pixel image at an initial time point of an interaction zone in the workpiece by means of a camera, wherein the workpiece is processed using an actuator having an initial actuator value a t ;
converting the pixel image into a pixel vector s t ; and
representing the pixel vector s t by a sum of predetermined pixel mappings each multiplied by a corresponding feature value, wherein the set of feature values φ(s t ) represents the basis function φ(s t ).
2. The method according to claim 1 , wherein step (c) comprises the step of classifying the set of feature values φ(s t ) on the basis of learned feature values to determine the reward function r t .
3. The method according to claim 1 , wherein the step (c) comprises the step of classifying a set of sensor data measured by a post-process sensor and/or a pre-process sensor.
4. The method according to claim 1 , wherein the actuator value a t is the laser power of a processing laser beam interacting with the workpiece in the interaction zone, the processing velocity, an x/y/z position/movement of a processing head relative to the workpiece, processing gas type and pressure, feed-rate of added materials in case of other cladding, cutting, welding, soldering, or material processing techniques.
5. The method according to claim 1 , wherein the processing operation is a laser processing operation comprising a laser welding process or a laser cutting process.
6. The method according to claim 1 , wherein the predetermined pixel mappings are obtained by the steps of:
recording a training set of pixel images at a predetermined first number of time points displaying the interaction zone having a good processing result, and at a predetermined second number of time points displaying the interaction zone having a had processing result;
converting the pixel images into pixel vectors and generating a covariance matrix from the pixel vectors; and
calculating the eigenvectors of the covariance matrix to determine the predetermined pixel mappings.
7. The method according to claim 6 , wherein learned feature values are obtained by the following steps:
representing each pixel vector of the training set by a sum of predetermined pixel mappings multiplied by corresponding feature values; and
training a classifier with obtained feature values by discriminating feature values at the first number and the second number of time points into a good class and a bad class.
8. The method according to claim 7 , wherein the classifier is selected from a group comprising Support Vector Machines (SVM), Artificial Neural Networks (ANN), or a Fuzzy-K Nearest Neighbor (KNN).
9. The method according to claim 6 , wherein the predetermined pixel mappings are calculated by means of isometric feature mapping (ISOMAP), linear discriminant analysis (LDA) or principal component analysis (PCA).
10. The method according to claim 1 , wherein a further feature value φ(s t ) is obtained by measuring the temperature of the interaction zone by means of an optical temperature sensor.
11. The method according to claim 1 , wherein further feature values φ(s t ) are obtained by sensor data of three photodiodes recording process emissions on different wavelengths.
12. The method according to claim 1 , wherein the recorded pixel image of the camera is a pixel image of the processed workpiece illuminated by LEDs or laser sources having a different wavelength than a processing laser, wherein further optical filter systems are employed to prevent disturbances generated by the light of the processing laser beam or generated by emission of the molten material of the workpiece.
13. A laser material processing head having a laser and a Reinforcement Learning agent unit being adapted to perform a method according to claim 1 .
14. A laser material processing head according to claim 13 , comprising a high-speed camera, sensors for solid-borne and air-borne acoustics, laser-triangulation, a temperature sensor and three photodiodes recording process emissions on different wavelengths for generating sensor data to be used as feature values.