METHOD AND DEVICE FOR OPERATING A LASER MATERIAL PROCESSING MACHINE

Abstract

A computer-implemented method for operating a laser material processing machine. Process parameters are varied with the aid of Bayesian optimization until a result of the manufacturing, in particular the laser material processing, is sufficiently good. The Bayesian optimization is carried out with the aid of a data-based process model in a first phase, the data-based process model being trained as a function of estimated results. In a second phase, the data-based process model is trained as a function of the ascertained result resulting upon activation of the laser material processing machine.

Claims

1-17. (canceled)

18. A computer-implemented method for operating a manufacturing machine, the method comprising the following steps: varying process parameters using Bayesian optimization until a result of a manufacturing with the manufacturing machine is sufficiently good, the Bayesian optimization being carried out using a data-based process model, wherein in a first phase, the data-based process model is trained as a function of estimated results, and in a second phase, the data-based process model is trained as a function of an ascertained result resulting upon activation of the manufacturing machine.

19. The method as recited in claim 18, wherein the manufacturing machine is a lase material processing machine, and the manufacturing is laser material processing.

20. The method as recited in claim 18, wherein the data-based process model to be trained is given as a sum of a first regression model and a second regression model, and wherein in the first phase, the first regression model is trained, and in the second phase, the second regression model is trained.

21. The method as recited in claim 20, wherein the second regression model is trained with a difference of the ascertained result and a prediction of the first regression model in the case of associated process parameters.

22. The method as recited in claim 21, wherein the second regression model is a Gaussian process model.

23. The method as recited in claim 22, wherein the first regression model is also a Gaussian process model.

24. The method as recited in claim 18, wherein it is taken into consideration in the variation of the process parameters how probable it is that a variable, which characterizes a quality of the ascertained result is within predefinable limits.

25. The method as recited in claim 24, wherein in an acquisition function, as a function of which the variation of the process parameters is ascertained, a probability is taken into consideration which characterizes how probable it is that the ascertained result is within predefinable limits.

26. The method as recited in claim 25, wherein the probability is ascertained based on a data-based quality model.

27. The method as recited in claim 26, wherein the data-based quality model is configured to output parameters which characterize a statistical prognosis of a result to be expected.

28. The method as recited in claim 27, wherein it is taken into consideration in the variation of the process parameters how probable it is that further variables, which each characterize further qualities of the ascertained result, are within respective predefinable limits, and the particular probabilities are ascertained using a particular further data-based quality model

29. The method as recited in claim 24, wherein the data-based process model and/or the data-based quality model is trained as a function of the ascertained result resulting upon activation of the laser material processing machine and/or the variable resulting upon activation of the laser material processing machine.

30. The method as recited in claim 26, wherein the data-based quality model is trained as a function of the ascertained result and the estimated result.

31. The method as recited in claim 30, wherein in the first phase, the data-based quality model is trained as a function of the estimated results, and in the second phase, the data-based quality model is trained as a function of the ascertained result resulting upon activation of the laser material processing machine and/or the variable resulting upon activation of the laser material processing machine.

32. The method as recited in claim 18, wherein following a setting of the process parameters, the laser material processing machine is operated using the process parameters set.

33. A test stand for a laser material processing machine, the test stand configured to operate the laser material processing machine, the test stand configured to: varying process parameters using Bayesian optimization until a result of a laser machine processing with the laser material processing machine is sufficiently good, the Bayesian optimization being carried out using a data-based process model, wherein in a first phase, the data-based process model is trained as a function of estimated results, and in a second phase, the data-based process model is trained as a function of an ascertained result resulting upon activation of the laser material processing machine.

34. A non-transitory machine-readable memory medium on which is stored a computer program for operating a manufacturing machine, the computer program, when executed by a computer, causing the computer to perform the following steps: varying process parameters using Bayesian optimization until a result of a manufacturing with the manufacturing machine is sufficiently good, the Bayesian optimization being carried out using a data-based process model, wherein in a first phase, the data-based process model is trained as a function of estimated results, and in a second phase, the data-based process model is trained as a function of an ascertained result resulting upon activation of the manufacturing machine.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0067] FIG. 1 schematically shows a structure of a laser drilling machine, in accordance with an example embodiment of the present invention.

[0068] FIG. 2 schematically shows a structure of a laser welding machine, in accordance with an example embodiment of the present invention.

[0069] FIG. 3 schematically shows a structure of a test stand, in accordance with an example embodiment of the present invention.

[0070] FIG. 4 shows a specific embodiment for operating the test stand in a flowchart, in accordance with the present invention.

[0071] FIG. 5 shows a specific embodiment for operating the test stand in a flowchart, in accordance with the present invention.

[0072] FIG. 6 shows a specific embodiment of a partial aspect of one of the two above-mentioned methods in a flowchart, in accordance with the present invention.

DETAILED DESCRIPTION OF EXAMPLE EMBODIMENTS

[0073] FIG. 1 schematically shows a structure of a laser drilling machine 1. An activation signal A is provided by an activation logic 40 to activate a laser 10a. The laser beam strikes a material piece 12, where it generates a drilled hole 11.

[0074] FIG. 2 schematically shows a structure of a laser welding machine 2. An activation signal A is also provided here by an activation logic 40 to activate a laser 10b. The laser beam strikes two material pieces 13, 14 and generates a weld seam 15 there.

[0075] A laser cutting machine (not shown) is also similarly possible.

[0076] FIG. 3 schematically shows a structure of a test stand 3 for ascertaining optimum process parameters x. Present process parameters x are provided by a parameter memory P via an output interface 4 of the laser material processing machine, for example, laser drilling machine 1 or laser welding machine 2. This machine carries out the laser material processing as a function of these provided process parameters ϕ. Sensors 30 ascertain sensor variables S, which characterize the result of the laser material processing. These sensor variables S are provided as quality properties y.sub.exp to a machine learning block 60 via an input interface 50.

[0077] In the exemplary embodiment, machine learning block 60 includes a Gaussian process model, which is trained as a function of provided quality properties y.sub.exp, as illustrated in FIG. 4 and FIG. 5. Varied process parameters x′ may be provided as a function of the Gaussian process model, which are stored in parameter memory P.

[0078] Process parameters x may also, alternatively or additionally to the provision via output interface 4, be provided to an estimation model 5, which provides estimated quality properties y.sub.sim to machine learning block 60 instead of actual quality properties y.sub.exp.

[0079] In the exemplary embodiment, the test stand includes a processor 45 which is configured to execute a computer program stored on a computer-readable memory medium 46. This computer program includes instructions which prompt processor 45 to carry out the method illustrated in FIGS. 4 and 5 when the computer program is executed. This computer program may be implemented in software or in hardware or in a mixed form of hardware and software.

[0080] FIG. 4 shows a flowchart of an exemplary method for operating test stand 3. Method begins 100 in that initial process parameters x.sub.init are provided as process parameters x and experimental data recorded up to this point are initialized as an empty set. Optionally, process parameters x are predefined using a design of experiment method and as explained in more detail hereinafter, laser material processing machine 1, 2 is activated using these process parameters x, variables y.sub.exp are ascertained, and Gaussian process model GP is trained using the experimental data thus ascertained.

[0081] In the case of laser drilling, these process parameters x include in one exemplary embodiment a pulse duration, a focus position resolved in a time-dependent manner via a characteristic map and/or a focus size and/or a pulse repetition frequency and/or a circular path diameter (time-dependent) resolved in a time-dependent manner via a characteristic map and/or a circular path frequency and/or an incidence angle resolved in a time-dependent manner via a characteristic map and/or a drilling duration and/or a pulse energy resolved in a time-dependent manner via a characteristic map and/or a wavelength and/or parameters which characterize a process shielding gas, for example, a process gas type or a process gas pressure. The mentioned circular path is a conventional feature here in many drilling methods, for example, in spiral drilling or in trepanning drilling.

[0082] In the case of laser welding, these process parameters x include laser power resolved in a time-dependent and/or location dependent manner via characteristic maps and/or a focus diameter and/or a focus position and/or a welding speed and/or a laser beam inclination and/or a circular path frequency of a laser wobble and/or parameters which characterize a process shielding gas.

[0083] Laser material processing machine 1, 2 is activated 110 using present process parameters x and variables y.sub.exp are ascertained 120 which characterize the actual result of the laser material processing.

[0084] In the case of laser drilling, these variables y.sub.exp include, in one exemplary embodiment, variables, which characterize the size of drilled hole 12 and/or the circularity of drilled hole 12 and/or the shape of a wall of drilled hole 12 and/or the presence of melt deposits and/or an amount of droplet ejection during the drilling process and/or a rounding of the edges of drilled hole 12 and/or the productivity.

[0085] In the case of laser welding, these variables y.sub.exp include, in another exemplary embodiment, variables which characterize, along weld seam 15, a minimum weld seam depth and/or minimum weld seam width and/or the productivity and/or a number of weld spatters; and/or a number of pores; and/or a welding distortion; and/or welding residual stress; and/or welding cracks.

[0086] As a function of these variables, a cost function K is evaluated 130, as may be given, for example, by equation 1, variables y.sub.exp being provided as quality properties q.sub.i and corresponding target values of these variables a q.sub.i,Target.

[0087] A cost function K is also possible which punishes deviations of the quality properties from the target values, in particular if they exceed a predefinable tolerance distance, and rewards a high productivity. The “punishment” may be implemented, for example, by a high value of cost function K, the “reward” correspondingly by a low value.

[0088] It is then ascertained whether cost function K indicates that present process parameters x are sufficiently good; in the case in which a punishment means a high value and a reward means a low value in that it is checked whether cost function K falls below 140 a predefinable highest cost value. If this is the case (“yes”), the method ends 150 with present process parameters x.

[0089] If this is not the case (“no”), data point x,y.sub.exp thus ascertained made up of process parameters x and associated variables y.sub.exp characterizing the result is added 160 to ascertained experimental data and hyperparameters Θ.sub.0,Θ.sub.1 of Gaussian process model GP are adapted in such a way that a probability that the experimental data result from Gaussian process model GP is maximized.

[0090] An acquisition function is then evaluated 170, as illustrated by way of example in formula (7), and new process parameters x′ are hereby ascertained. The sequence then branches back to step 110.

[0091] FIG. 5 shows a flowchart of a further exemplary method for operating test stand 3. Steps 100, 110, 130, 140, 150 are illustrated identically as in FIG. 4, a separate description is therefore omitted.

[0092] In step 120b, which replaces step 120 of the method illustrated in FIG. 4, some of variables y.sub.exp determined therein are provided separately in each case as bounded variables , ε, , . . . , which are each to be in a bounded interval: ∈[.sub.0−δ,.sub.0+δ], ε∈[ε.sub.0−ϵ, ε.sub.0+ϵ], ∈[.sub.0−ϕ,.sub.0+ϕ)], . . . .

[0093] In step 160b, which replaces step 160 of the method illustrated in FIG. 4, in addition to the step described in 160, for each of bounded variables , ε, , a data point x,, x,ε, x,, . . . is added in each case to the particular ascertained experimental data, thus associated with each of bounded variables , ε, , and similarly to the training of Gaussian process model GP, for each of bounded variables , ε, , a separate Gaussian process model G,GP.sub.ε,G is trained.

[0094] In step 170b, which replaces step 170 of the method illustrated in FIG. 4, in addition to the evaluation described there of the acquisition function. For this purpose, as stated above, a so-called “expected improvement” function may be evaluated and maximized as illustrated in formula 7. Furthermore, a predefinable probability function (x) is provided, which characterizes a probability of whether selected process parameters x result in a satisfactory result in laser material processing machine 1, 2 or not, i.e., whether bounded variables will actually be in the associated interval, thus ∈[.sub.0−δ,.sub.0+δ]. For further bounded variables ε, , . . . corresponding probability functions p.sub.ε(x), (x) are provided.

[0095] In the exemplary embodiment, probability function (x) is ascertained from a variance σ.sup.2 of Gaussian process model . For this purpose, lower limit .sub.0−δ and upper limit .sub.0+δ variables ascertained for Gaussian process model D are provided and it is ascertained, for example, using numeric integration how great the probability is that the variables ascertained by are between lower limit .sub.0−δ and upper limit .sub.0+δ, thus (x):=p(.sub.0−δ<(x)<.sub.0+δ). A similar procedure may be used for probability functions p.sub.ε(x), (x). The mentioned acquisition function may additionally be multiplied by the product of ascertained probabilities (x).Math.p.sub.ε(x).Math.(x) . . . and new process parameters x′ may hereby be ascertained. The sequence then branches back to step 110.

[0096] FIG. 6 illustrates in a flowchart the sequence of a method as may be used for training the Gaussian process models GP,G,GP.sub.ε,G. It is illustrated in the following by way of example on the basis of Gaussian process model GP and may be transferred accordingly to each of the others.

[0097] The method is made up of first Gaussian process GP.sub.0 and second Gaussian process GP.sub.1, which together additively result in GP, thus

GP(x)=GP.sub.0(x)+GP.sub.1(x).

[0098] The method is made up of a first phase A and a second phase B. The method illustrated in FIG. 4 or in FIG. 5 may be used in each of the phases, estimated variables y.sub.sim ascertained by simulation taking the place of actual variables y.sub.exp in first phase A.

[0099] In first phase A, initially process parameters x are predefined 6000. Estimated variables y.sub.sim are then ascertained by simulation 6010.

[0100] In the case of laser drilling, for example, this may be carried out as follows with the aid of a physical method: For a radius r of drilled hole 11 along a depth coordinate z, r(z) is numerically ascertained as the solution of the equation

[1−R(r,z,α,θ)].Math.cos θ.Math.F.sub.0(r,z)−{tilde over (F)}.sub.th=0  (8)

where

[00001]$\begin{array}{cc}1-R=\frac{1}{2}.Math.\left(\frac{4n\cos \theta }{\left(n^2+k^2\right)+2n\cos \theta +{\cos }^2\theta }+\frac{4n\cos \theta }{\left(n^2+k^2\right){\cos }^2\theta +2n\cos \theta +1}\right)& \left(9\right)\\ F_0\left(r,z\right)=\frac{2Q}{\pi w^2\left(z\right)}.Math.\exp \left(-\frac{2r^2}{w^2\left(z\right)}\right)& \left(10\right)\\ w\left(z\right)=\frac{d_{Fok}}{2}\sqrt{1+{\left(\frac{z}{l_{\mathrm{Rayleigh}}}\right)}^2}& \left(11\right)\\ \tan \alpha =\frac{r}{w\left(z\right)}\frac{dw\left(z\right)}{dz}& \left(12\right)\end{array}$

[0101] In this case: [0102] n=n+ik is a predefinable complex index of refraction of material piece 12, with index of refraction n and extinction coefficient k [0103] {tilde over (F)}.sub.th is a predefinable ablation threshold fluence of material piece 12, [0104] Q is a predefinable pulse energy of laser 10a, [0105] d.sub.Fok is a predefinable focus diameter of laser 10a, [0106] l.sub.Rayleigh is a predefinable Rayleigh length of laser 10a, [0107] R is an ascertained reflectivity of material piece 12, [0108] α is an ascertained angle of the local beam propagation direction, [0109] θ is a predefinable relative angle between incident laser beam and the surface normal of material piece 12, [0110] F.sub.0 is an ascertained irradiated fluence of laser 10a, [0111] w(z) is an ascertained local beam radius.

[0112] The prediction of some quality properties such as a presence of melt deposits and/or an amount of droplet ejection during the drilling process is not possible using this physical model. An empirical model may be predefined, for example, to ascertain these quality properties.

[0113] Alternatively or additionally, it is possible that at least some of the quality properties may not be reliably calculated for all process parameters x. It is possible that it is checked whether present process parameters x are within a predefinable range, and if this is not the case, the quality properties are ascertained with the aid of one of the above-mentioned approaches.

[0114] In the case of laser welding, the ascertainment of estimated variables y.sub.sim may be carried out as follows, for example, using a physical model:

[00002]$\begin{array}{cc}\left(13\right)T\left(x,y,z\right)-T_0=\frac{1}{2\pi \lambda h}\exp -\frac{v\left(x-x_0\right)}{2a}(q_{net}{K}_0\left(\frac{\mathrm{vr}}{2a}\right)+\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}2{.Math.}_{m=1}\cos \frac{m_{pi}z}{h}{K}_0\left(\frac{\mathrm{vr}}{2a}\sqrt{1+{\left(\frac{2m_{}a}{\mathrm{vh}}\right)}^2}\right)I_m)& \\ \mathrm{where}& \\ r=\sqrt{{\left(x-x_0\right)}^2+y^2}& \left(14\right)\\ l_m={\int }_0^h{q}_{1net}\left(z\right)\cos \frac{m_{\pi }z}{h}\mathrm{dz}& \left(15\right)\end{array}$

and the parameters
T.sub.0—a predefinable ambient temperature
x.sub.0—a predefinable offset of laser 10b to the origin of a coordinate system movable with laser 10b
λ—a predefinable thermal conductivity of material pieces 13, 14;
a—a predefinable temperature conductivity of material pieces 13, 14;
q.sub.net—a predefinable power of laser 10b;
q.sub.1net—a predefinable power distribution of laser 10b along a depth coordinate of material pieces 13, 14;
v—a predefinable speed of laser 10b;
h—a predefinable thickness of material pieces 13, 14; and Bessel function

[00003]$K_0\left(\omega \right)=\frac{1}{2}{\int }_{-\infty }^{\infty }\frac{e^{i\omega t}}{\sqrt{t^2+1}}dt$

and an ascertained temperature distribution T(x,y,z). A width or a depth of the weld seam may be ascertained from the temperature distribution (for example via the ascertainment of isotherms at a melting temperature of a material of material pieces 13, 14).

[0115] First Gaussian process GP.sub.0 is then trained 6020 with the aid of estimated results y.sub.sim.

[0116] It is subsequently checked whether an abort criterion is reached 6030. For example, it may be checked whether a cost function K ascertained as a function of estimated result y.sub.sim falls below a predefinable threshold value. If the abort criterion is met, second phase B follows; otherwise the sequence branches back to 6000.

[0117] In second phase B, the method illustrated in FIG. 4 and FIG. 5 is executed 6040, instead of Gaussian process model GP, second Gaussian process GP.sub.1 being trained with the aid of actual results y.sub.exp and associated process parameters x, in fact in that actual results y.sub.exp are ascertained by the difference of actual results y.sub.exp and the prediction of first Gaussian process GP.sub.0(x) at associated process parameters x.sub.exp, thus

y.sub.exp.fwdarw.y.sub.exp−GP.sub.0(x)=y.sub.exp*  (16).

[0118] Second Gaussian process GP.sub.1 is then trained using actual results y.sub.exp* thus transformed and associated process parameters x.sub.exp.

[0119] Instead of first Gaussian process GP.sub.0, another suitable regression model may also be used. For example, it is possible to instead use a suitable polynomial (possibly defined piecewise) or a spline.