Method for classifying a multitude of images recorded by a camera observing a processing area and laser material processing head using the same

1. 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 , wherein the position and orientation of the keyhole in the recorded pixel image is normalized according to a method for classifying a multitude of images recorded by a camera observing a processing area of a workpiece processed by a processing beam, comprising the steps of recording a first pixel image and a multitude of subsequent pixel images by the camera during a processing operation, detecting mismatches of a position and orientation of a keyhole generated by the processing beam in the workpiece within an image plane of the subsequent pixel images in comparison to the first pixel image, compensating the mismatches of the position and orientation of the respective keyholes in the subsequent pixel images with regard to the first pixel image, to produce a set of pixel images having each a normalized keyhole position and orientation, classifying the set of normalized pixel images into at least two classes by means of a classifier;

(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 a 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 RL controlled processing operation.

2. A method for closed-loop controlling a processing operation of a workpiece, comprising the steps of:

(a) recording a pixel image at an initial time point of an interaction zone by means of a camera,

wherein the workpiece is processed using an actuator having an initial actuator value, wherein the position and orientation of the keyhole in the recorded pixel image is normalized according to a method for classifying a multitude of images recorded by a camera observing a processing area of a workpiece processed by a processing beam, comprising the steps of recording a first pixel image and a multitude of subsequent pixel images by the camera during a processing operation, detecting mismatches of a position and orientation of a keyhole generated by the processing beam in the workpiece within an image plane of the subsequent pixel images in comparison to the first pixel image, compensating the mismatches of the position and orientation of the respective keyholes in the subsequent pixel images with regard to the first pixel image, to produce a set of pixel images having each a normalized keyhole position and orientation, and classifying the set of normalized pixel images into at least two classes by means of a classifier;

(b) converting the pixel image into a pixel vector;

(c) representing the pixel vector by a sum of predetermined pixel mappings each multiplied by a corresponding feature value;

(d) classifying the set of feature values on the basis of learned feature values into at least two classes of a group of classes comprising a first class of a too high actuator value, a second class of a sufficient actuator value and a third class of a too low actuator value at the initial time point;

(e) performing a control step for adapting the actuator value by minimizing the error e t between a quality indicator y e and a desired value; and

(f) repeating the steps (a) to (e) for further time points to perform a closed-loop controlled processing operation.