PROCESS SIGNAL RECONSTRUCTION AND ANOMALY DETECTION IN LASER MACHINING PROCESSES

Abstract

A method and a system for monitoring a laser machining process includes the steps of: inputting at least one process signal data set of the laser machining process into an autoencoder formed by a deep neural network; generating a reconstructed process signal data set by means of the autoencoder; determining a reconstruction error based on the at least one process signal data set and the at least one reconstructed process signal data set; and detecting an anomaly of the laser machining process based on the determined reconstruction error. A laser machining method includes the method and a laser machining system includes the system.

Claims

1. A method for monitoring a laser machining process, said method comprising the steps of: inputting at least one process signal data set of the laser machining process into an autoencoder formed by a deep neural network; generating a reconstructed process signal data set by means of said autoencoder; determining a reconstruction error based on the at least one process signal data set and the at least one reconstructed process signal data set; and detecting an anomaly of the laser machining process based on the determined reconstruction error.

2. The method according to claim 1, further comprising: measuring at least some of the process signals of the process signal data set; and/or transmitting at least some of the process signals of the process signal data set from a control.

3. The method according to claim 1, further comprising the steps: determining a quality feature of the laser machining process; and evaluating the quality feature as valid when no anomaly is detected in the step of detecting an anomaly of the laser machining process; and evaluating the determined quality feature as not valid when an anomaly of the laser machining process is detected in the step of detecting an anomaly.

4. The method according to claim 3, wherein: the step of determining a quality feature of the laser machining process is by means of a regressor formed by a neural network, and a value for the quality feature is determined; and/or the step of determining a quality feature of the laser machining process is by means of a classifier formed by a neural network, and a classification value for the quality feature is determined.

5. The method according to claim, wherein: said autoencoder and at least one of said regressor and said classifier have a common encoder; and/or said regressor and/or said classifier determines the quality feature based on data from an encoder of said autoencoder.

6. The method according to claim 4, wherein: said autoencoder and at least one of said regressor and said classifier are parallel to each other; and/or said regressor and/or said classifier determines the quality feature based on the at least one process signal data set; and/or said autoencoder and at least one of said regressor and said classifier have a common input layer.

7. The method according to claim 4, wherein said autoencoder and at least one of said regressor and said classifier are trained with the same data.

8. The method according to claim 1, wherein the step of determining a reconstruction error comprises: determining a deviation of the at least one process signal data set from the at least one reconstructed process signal data set; and/or determining a mean absolute or squared deviation of the at least one process signal data set from the at least one reconstructed process signal data set; and/or determining a signed, absolute or squared deviation summed up along the time axis; and/or determining a Mahalanobis distance.

9. The method according to claim 1, wherein the step of determining a reconstruction error comprises determining a Mahalanobis distance with respect to: a deviation of the at least one process signal data set from the at least one reconstructed process signal data set; and/or individual characteristic values of the reconstruction error; and/or encoding of a process signal data set.

10. The method of claim 8, wherein said method comprises determining parameters mean vector and covariance matrix of the Mahalanobis distance using defect-free or labeled data sets.

11. The method according to claim 1, wherein the reconstruction error for individual dimensions is determined separately and/or based on a metric and/or by means of a fast Fourier transformation and/or a wavelet transformation.

12. The method according to claim 1, wherein: said method comprises normalizing the reconstruction error with respect to the process signal data set; and/or the step of determining a reconstruction error comprises: filtering at least part of the process signal data set and/or the reconstructed process signal data set; and based thereon, determining the reconstruction error.

13. The method according to claim 1, wherein said method comprises: determining a degree of abnormality; wherein detecting an anomaly of the laser machining process is based on the determined degree of abnormality.

14. The method according to claim 13, wherein the determining a degree of abnormality is based on a weighted summation or on a Mahalanobis distance with respect to individual characteristic values for the reconstruction error.

15. A laser machining method, comprising the steps of: machining a workpiece by means of a laser beam; and monitoring the laser machining process according to the method according to claim 1.

16. A system for monitoring a laser machining process, said system comprising: at least one sensor assembly configured to sense process signals of the laser machining process; at least one autoencoder formed by a deep neural network; and at least one processor configured to carry out the method for monitoring the laser machining process according to claim 1.

17. A laser machining system for machining a workpiece by means of a machining laser beam, said laser machining system comprising: a laser machining head for radiating the machining laser beam onto said workpiece; and a system according to claim 16.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0091] Embodiments of the present invention are described in detail below with reference to figures. The figures depict various features of embodiments, the features not being limited only to the embodiments. Rather, all features that are not mutually exclusive may also be combined with one another or features may be removed from embodiments if they are not essential for the implementation of the invention.

[0092] FIG. 1 shows a schematic diagram of a laser machining system for machining a workpiece by means of a laser beam and a system for monitoring a laser machining process according to an embodiment;

[0093] FIG. 2 shows a schematic diagram of a deep autoencoder for processing process signal data sets of the laser machining process according to an embodiment;

[0094] FIG. 3 schematically shows a process signal data set and a reconstructed process signal data set;

[0095] FIG. 4 schematically shows a process signal data set and a reconstructed process signal data set in case of an anomaly in the laser machining process;

[0096] FIG. 5 shows a schematic diagram of a deep autoencoder in parallel with a regressor or classifier according to an embodiment;

[0097] FIG. 6 shows a schematic diagram of an extended deep autoencoder together with a regressor or classifier according to an embodiment; and

[0098] FIG. 7 shows a schematic diagram of a method for monitoring a laser machining process according to an embodiment.

DETAILED DESCRIPTION OF THE INVENTION

[0099] Unless otherwise noted, the same reference symbols are used below for the same elements and those with the same effect.

[0100] FIG. 1 shows a schematic diagram of a laser machining system 100 for machining a workpiece by means of a laser beam 10 according to embodiments of the present disclosure. The laser machining system 100 comprises a laser machining head 101, in particular a laser cutting, laser soldering or laser welding head, and a system 300 for monitoring the laser machining process including a sensor unit or sensor assembly 310 as an example of a laser machining head according to one embodiment. The system 300 for monitoring the laser machining process further comprises a computing unit or processor 320.

[0101] The laser beam 10 is supplied, for example, via an optical fiber 104 from a laser source (not shown). The beam path of the (machining) laser beam 10 extends via collimation optics 122 and focusing optics 124 onto a workpiece 18 being machined.

[0102] According to embodiments, the laser machining system 100 or parts thereof, such as the machining head 101, may be movable relative to the workpiece 18 along a machining direction 20. The machining direction 20 may be a cutting, soldering or welding direction and/or a direction of movement of the laser machining system 100, such as the machining head 101, with respect to the workpiece 18. In particular, the machining direction 20 may be a horizontal direction.

[0103] The laser machining system 100 is controlled by a controller 140 configured to control the machining head 101.

[0104] Process emissions 13 occur during machining, for example thermal radiation, plasma radiation and/or laser radiation reflected from a surface of a workpiece. A portion of the process emissions 13 is guided at least partially coaxially or collinearly to the beam path of the machining beam 10 from the surface of the workpiece via a beam splitter 115 and collimator optics 312 to the sensor unit 310.

[0105] The computing unit 320 includes a deep neural network in the form of a deep autoencoder 400 and receives from the sensor unit 310 process signals 11 which are input into the neural network 400 in the form of process signal data sets 30.

[0106] FIG. 2 shows a schematic diagram of the deep autoencoder 400 for processing process signal data sets 31 of the process signals 11 of the laser machining process. The autoencoder 400 includes an input layer 1, an encoder 2, a coding layer 3, a decoder 4 and an output layer 5. The coding layer 3 is a common layer of the encoder 2 and the decoder 4.

[0107] The autoencoder 400 performs the following processing: At least one process signal data set 31 of a process signal 11 is input into the input layer 1. The encoder 2, consisting of a deep neural network, extracts features of incoming signals 11 into an encoding in a coding layer 3. This encoding also represents a dimension reduction of the incoming process signals 11. This enables the process signals 11 to be described in a few dimensions. The decoder 4, consisting of a deep neural network, reconstructs the process signals. A reconstructed process signal data set 51 of the reconstructed signal is output at the output layer 5.

[0108] FIG. 3 schematically shows a process signal data set 31 and an associated reconstructed process signal data set 51, for example as a course of a signal intensity I over time t. Except for signal noise, fluctuations in the process signal 11 that also occurred in the signals learned by the autoencoder 400 are almost perfectly reconstructed.

[0109] FIG. 4 schematically shows a process signal data set 31A and an associated reconstructed process signal data set 51A in case an anomaly 40 occurs in the laser machining process. When anomalies 40 occur in the form of gross differences in the signal curve compared to the previously trained signal curves, these are not correctly reconstructed by the autoencoder 400.

[0110] The computing unit 320 determines a reconstruction error as the mean absolute or squared deviation of the input signal 11 or process signal data set 31/31A from the reconstructed signal or reconstructed process signal data set 51/51A.

[0111] When the reconstruction error exceeds a threshold value, the computing unit 320 detects the presence of an anomaly 40. When the method is used directly for quality assessment in the laser machining process, the anomalies 40 indicate a quality defect or faulty process, and the product is classified by the computing unit 320 as “not OK” when an anomaly 40 was detected, i.e. when the reconstruction error exceeds a threshold value.

[0112] Some of the process signals of the process signal data set 31 may also be transmitted by the controller 140. Process signals transmitted by the controller 140 may include, for example, control signals, e.g. a specified laser power, a target focal position, a machining speed, etc. Corresponding reconstructed process signals are then part of the reconstructed process signals of the reconstructed process signal data set 51.

[0113] FIGS. 5 and 6 show schematic diagrams of the deep autoencoder 400 together with a regressor 600 or classifier 600 for predicting a quality feature x of the laser machining process. In the quality monitoring of the laser machining process, the quality feature x of the laser machining process is predicted, the quality feature x is assessed as valid when no anomaly 40 is detected, and the quality feature x is assessed as invalid when an anomaly 40 is detected.

[0114] The autoencoder 400 is trained together with the regressor 600 or classifier 600 with known data. For this purpose, the data are labeled with a discrete or continuous value that represents the quality feature x. The regressor 600 or classifier 600 is implemented as a deep neural network 6 with an output layer 7 in the computing unit 320.

[0115] FIG. 5 shows a schematic diagram of a deep autoencoder 400 in parallel with a regressor 600 or classifier 600.

[0116] At least one process data set 31 is input into the common input layer 1 of the autoencoder 400 and the regressor 600 or classifier 600 for the quality monitoring of laser machining processes. Therefrom, the quality feature x is predicted by means of the deep neural network 6 of the regressor 600 or classifier 600, and at the same time the process signals are reconstructed as a process signal data set 51 via the autoencoder 400.

[0117] FIG. 6 shows a schematic diagram of an extended deep autoencoder 400 that includes a regressor or classifier 600.

[0118] The extended autoencoder 400 creates a common encoding, which represents recurring features and patterns, in the coding layer 3. For quality monitoring of laser machining processes, the quality feature x is predicted from the encoding of the coding layer 3 by means of the deep neural network 6 of the regressor 600 or classifier 600, and at the same time the process signals are reconstructed from the encoding of the coding layer 3 as a process signal data set 51 via the decoder 4.

[0119] FIG. 7 shows a schematic diagram of an exemplary method for monitoring a laser machining process, such as may be carried out by the system 300 according to FIG. 4 or 5.

[0120] The method includes the following steps:

[0121] Step 510: measuring at least some of the process signals of the process signal data set 31 and/or transmitting at least some of the process signals of the process signal data set 31 from the control 140;

[0122] Step S20: inputting the process signal data set 31 of the laser machining process into the autoencoder 400;

[0123] Step S30: generating an encoding by means of the encoder 2;

[0124] Step S40: generating of a reconstructed process signal data set 51 by means of the autoencoder 400;

[0125] Step S50: determining a reconstruction error by means of the computing unit 320;

[0126] Step S60: detecting an anomaly of the laser machining process by means of the computing unit 320 based on the determined reconstruction error;

[0127] Step S70: determining a value of a quality feature of the laser machining process, for example by means of the regressor 600 or classifier 600;

[0128] Step S80: evaluating the quality feature as valid when no anomaly is detected in step S60 (case N); and

[0129] Step S90: evaluating the quality feature as not valid when an anomaly is detected in step S60 (case A).

[0130] As described for the exemplary embodiments, a plausibility check and anomaly detection in combination with deep neural networks for regression and classification in laser material machining is made possible by an extended autoencoder. Based on the autoencoder, the signals may be reconstructed and a reconstruction error may be calculated. The reconstruction error may be used both to evaluate the quality of the laser machining process and to check the validity of the prediction of the regressor or classifier.

[0131] The initial effort when generating error examples in laser material machining for training quality monitoring systems based on a regressor or classifier can thus be significantly reduced since their availability space can be limited according to the known error examples. The user can monitor the quality with a few error examples and a few good examples. The prediction accuracy for the quality of the process and the product can be increased during the ongoing production process by the worker analyzing anomalies, providing them with a quality feature and the proposed algorithm or method, i.e. the autoencoder and, if necessary, the regressor and/or classifier, re-learning them.