Method and apparatus for operating a laser material processing machine

Abstract

Computer-implemented method and apparatus for operating a laser material processing machine (1, 2), wherein an estimate (y sim ) is determined from predefined process parameters (x), which estimate characterizes how good an actual result (y exp ) of the laser material processing is, and wherein the process parameters (x) are changed using a data-based model by means of Bayesian optimization until the actual result (y exp ) of the laser material processing is good enough.

Description

Technical Field

[0001] The present invention relates to a method, test bench, computer program, and machine-readable storage medium for operating a laser material processing machine. Background Technology

[0002] Laser radiation drilling is a manufacturing method for creating holes in very different materials. Here, for example, a pulsed and focused laser beam is applied to the workpiece. Due to the very high intensity, the absorbed laser energy causes the workpiece material to be heated very rapidly in a pulsed manner, which leads to melt formation and partial evaporation in a very short time and very locally.

[0003] Molten material is expelled from the borehole by the vapor pressure generated by the process-related explosion and the resulting large pressure gradient, or by an externally supplied gas flow. At particularly high intensities, achieved, for example, by using laser radiation with ultrashort laser pulses, a larger evaporation fraction can be achieved, and more precise drilling can be accomplished.

[0004] With longer pulse durations and lower intensity, hole formation is largely dominated by melt ejection, leading to reduced accuracy but significantly increased productivity. In many cases, a large number of laser pulses are required per hole to create the desired borehole. To improve drilling accuracy, the laser beam can typically be guided in a circular or helical path at the drilling location using appropriate equipment.

[0005] In this manufacturing method of laser drilling, process development is typically experimental because it is currently impossible to model the numerous highly dynamic and interacting physical effects with sufficient accuracy. This also includes the fact that workpiece parameters for relevant pressures and temperatures are often unknown. At most, highly simplified models can be used, which can make certain predictions about the achieved borehole shape under given process parameters and within specific parameter ranges. These models cannot currently reliably predict quality characteristics such as hardened melt deposits in the form of burrs inside or at the borehole entrance, damage to the borehole edges, or the roundness of the borehole.

[0006] Laser welding is a well-established manufacturing method for joining workpieces made of different materials. A focused laser beam is applied to the workpieces to be joined. Due to its very high intensity, the absorbed laser energy causes very rapid, localized heating of the workpiece material, resulting in the formation of a common molten pool in a short time and a very localized location. After the molten pool hardens, a weld-like connection is formed between the workpieces.

[0007] To meet the requirements for joint strength (and fatigue strength), it may be desirable that the weld geometry be no less than the minimum permissible weld depth and minimum permissible weld width. To achieve the desired weld shape, process parameters can be selected such that rapid and localized heating of the material by laser radiation causes evaporation in the weld pool. The molten material is then expelled from the weld pool by the vapor pressure generated by the process-related explosion and the resulting large pressure gradient, or by an externally supplied gas flow. The resulting metal spatter (so-called weld spatter) can lead to degraded part quality and / or require production interruptions to clean the laser welding facility, resulting in a significant increase in manufacturing costs.

[0008] Similar to the case of laser drilling, process development (process optimization aimed at minimizing weld spatter) in the case of laser welding is also strongly experimental because it is impossible to model the large number of highly dynamic and interacting physical effects with sufficient accuracy.

[0009] The challenge in modeling here is that the workpiece parameters for the relevant pressures and temperatures are often unknown. Manufacturing tolerances and material variations in individual workpieces can also have a very strong impact on the formation of weld spatter. Although highly simplified models can be used, which can make some predictions about the achieved weld shape under given process parameters and within specific parameter ranges, these models cannot reliably predict quality characteristics such as hardened weld spatter.

[0010] Because there are many configurable process parameters (which are typically time- and location-dependent), such as laser power, focal diameter, focal position, welding speed, laser beam tilt, circular orbit frequency, and shielding gas, optimizing these parameters is a lengthy process requiring numerous experiments. Since these experiments require many workpieces or components, and evaluation (manufacturing cross-sections for measuring weld geometry) is also laborious, the number of trials must be minimized.

[0011] Therefore, for example, some process parameters are set to empirical values, and only a relatively small number of parameters are changed. In this case, it is usually impossible to find a practically achievable optimal value.

[0012] Advantages of the present invention

[0013] It has been recognized that the accuracy and productivity achievable in laser material processing depend very strongly on the process parameters set, the workpiece material used, and, in part, also very strongly on the geometry of the workpiece.

[0014] There are many quality standards for the drilling process. For example, borehole size (e.g., diameter distribution depending on depth), borehole roundness, borehole wall shape, possible melt deposits, droplet ejection during drilling, and rounding of borehole edges are all important. Productivity is typically defined by the number of boreholes that can be produced per unit of time. Furthermore, the cost of the necessary production equipment is, of course, decisive in practice, and this cost naturally increases with the flexibility of variable parameters.

[0015] Because there are many configurable process parameters (e.g., pulse duration, focal point position, focal point size, pulse repetition frequency, circular track diameter, circular track frequency, positioning angle, drilling duration, pulse energy, wavelength, process gas type, and process gas pressure), and these parameters can often vary over time, optimizing these process parameters is a lengthy process requiring numerous experiments. Since these experiments require many workpieces or components, and evaluation (especially of internal borehole shapes) is also very time-consuming, the number of trials required must be minimized.

[0016] Therefore, some process parameters can be set to empirical values, and only a relatively small number of parameters can be changed. Consequently, it is often difficult to find the practically achievable optimal values. As a planning method for such attempts, methods can be employed where experts pre-define a series of attempts and / or a statistical attempt plan is executed.

[0017] In laser welding, the optimization of process parameters is a lengthy process, requiring numerous experiments, due to the many configurable parameters (which are typically time- and location-dependent), such as laser power, focal diameter, focal position, welding speed, laser beam tilt, circular orbit frequency, and shielding gas. Since these experiments require many workpieces or components, and the evaluation (preparing cross-sections for measuring weld geometry) is also laborious, it is desirable to minimize the number of trials required.

[0018] Conversely, the subject matter characterized by the present invention has the advantage that process parameters for laser material processing machines that ensure high-quality laser material processing can be found with only a small number of experiments.

[0019] Other aspects of the invention are described below. Summary of the Invention

[0020] This invention relates to a method for effectively and specifically optimizing process parameters. For this purpose, a Bayesian optimization method is used. This method allows the finding of optimal values ​​within an unknown function. The optimal value is characterized by one or more quality characteristics (features) q. i The target value q specified by the user i,ZeilTo obtain a unique function to optimize, multiple quality characteristics can be computed in what is called the cost function K. This cost function must also be pre-defined by the user. An example is the sum of scaling deviations from the corresponding target values:

[0021] .

[0022] Here, parameter s i These are pre-defined scaling parameters. To find the optimal value of the cost function, a set of process parameters (hereinafter referred to as the parameter set) can be proposed for the next experiment by applying Bayesian optimization. After the experiment is performed, the quality standard value obtained can be determined, and the current cost function value can be determined from it, and provided to the optimization method as a data point along with the set of parameters.

[0023] Bayesian optimization methods are suitable for finding the set of parameters that leads to the optimal function value for a function that maps a multidimensional parameter space to scalar values. Depending on the optimization objective, the optimal value is defined here as the maximum possible value that the function value can take, or alternatively, the minimum achievable value. In the sense of process optimization, this means, for example, that the parameter set is given by a specific set of process parameters; the function value belonging to this specific set of process parameters can be determined by the cost function described above.

[0024] Because determining the value of the cost function requires conducting and evaluating experiments, the function, in principle, only provides a table of values ​​with data, which still contains experimental "noise." Since these experiments are very laborious, it is generally impossible to suppress this noise by repeatedly performing numerous experiments with the same parameter set and then averaging the results. Therefore, it is advantageous to perform the optimization using a method that achieves good global optimization results with fewer evaluation attempts, without needing to compute the gradient of the cost function. It has been recognized that the Bayesian optimization satisfies these properties.

[0025] The Bayesian optimization consists of a mathematical method of Gaussian processes and algorithmic formula rules. The mathematical method of Gaussian processes is used to obtain predictions of the most likely function values ​​(including their variances) for each parameter set based on a given value table. The rules indicate for which parameter sets other function evaluations based on the predictions of the Gaussian processes should be performed (i.e., experiments in this paper).

[0026] Specifically, in parameter set x N+1 In this case, the prediction of the result of the function evaluation is given by the most probable value (“mean”) of the Gaussian process.

[0027]

[0028] Where variance is

[0029]

[0030] Here C N The covariance matrix is ​​given by the following formula:

[0031] , where n, m=1..N, (4)

[0032] Where x n and x m It is a parameter set that has already undergone function evaluation. Variables δ represents the variance of a normal distribution, which describes the dispersion of experiments given the same set of parameters. nm It is the symbol for tensor product. A scalar c is usually represented by... Given a vector t with respect to the parameter set x that has been evaluated by the function. n (n=1..N) contains the corresponding results. Kernel function Describes the parameter set x n In the case that the result of the function evaluation is still valid for the parameter set x m The value indicates the extent of the impact of the function evaluation result. Here, a large value represents a significant impact, while a value of zero indicates no further impact.

[0033] Therefore, in order to predict the mean and variance in the above formula, for all parameter sets x n (n=1..N) and the parameter set x to be predicted N+1 Calculate vector k, where There are different options for the kernel function to be used in specific situations. The quadratic exponential kernel function is a simple one:

[0034] (5)

[0035] It has optional hyperparameters Θ0 and Θ d (d=1..D), where D is the dimension of the parameter space. In this kernel function, Θ0 describes the proportion of change in the function value, while Θ d Describes the "distance" in the parameter space for the parameter set x n and x m The impact of the correlation between the values ​​of the two functions under certain circumstances. Other kernel functions are also possible.

[0036] Based on the predictions of the mean and variance calculated using the formula above, select the next set of parameters to be used for the trial. Different strategies may be employed here; for example, "expected improvement".

[0037] Here, we select the following set of parameters for the next experiment, under which we find a known function that is larger (or smaller, depending on the optimization objective) than the one obtained from the previous N iterations. The expected value of a function that is larger (or smaller, depending on the optimization objective), i.e.

[0038] (7)

[0039] Maximum. The maximum possible function values ​​at point x. Here, the distribution is normal, with a mean according to formula (2) and a variance according to formula (3), each having x. N+1 =x. This type of function to be optimized is also called an acquisition function. Other acquisition functions are also possible, such as knowledge gradients or entropy search.

[0040] Here, the "+" operator means to use only positive values, while setting negative values ​​to zero. In the case of Bayesian optimization, iteratively...

[0041] - Determine new trial points (i.e., parameter sets).

[0042] -Execute an attempt,

[0043] Update the Gaussian process with the new function values.

[0044] Until the optimization is terminated.

[0045] Optimizing the Gaussian process with new trial points and new function values ​​adds new pairs of trial points and function values ​​to the already recorded trial data from the pairs of trial points and function values, and adapts the hyperparameters to maximize the probability (e.g., likelihood) of the trial data.

[0046] Combination Figure 4 The process is illustrated.

[0047] Through the iterative process described above (performing experiments, evaluating quality criteria and determining cost function values, updating the Gaussian process and suggesting the next set of parameters), a process model (mapped by the Gaussian process) can be progressively built. Then, the optimal set of parameters evaluated or tried from all evaluated functions is used as the best optimization result.

[0048] Advantages in performing the optimization are gained by incorporating existing process knowledge. One or more process models P can be derived through the process described below. 1...nKnowledge in the form of simulations is introduced into the optimization process by replacing real experiments with simulations under specific conditions. Here, the degree of uncertainty with which the model maps the process and the number of quality standards it describes are irrelevant.

[0049] By utilizing process models that perfectly map real experiments, each real experiment can be replaced by a simulation. If the evaluation duration is shorter than the actual execution, time can be saved in addition to cost. However, the predictive accuracy of such process models is generally limited. Typically, these models are only valid in a sub-region of the parameter space and / or describe only a subset of process results, and do not consider all physical effects, thus producing results only within the uncertainty range. Therefore, process models can usually only partially replace physical experiments, not completely.

[0050] In the sense of the invention described herein, in each iteration of the optimization step, a process simulation model capable of predicting a subset of relevant features with known accuracy is first invoked. If, based on the predicted process results, it can be ruled out with sufficient certainty that the process results will approach the target value, within the range of prediction accuracy, then a real experiment is not performed. Instead, the results calculated using the process model are used instead as experimental results, and the optimization process continues.

[0051] If multiple process simulation models with different prediction accuracies can be used for different regions in the parameter space, then the process simulation model with the best prediction accuracy can be used respectively.

[0052] Therefore, in a first aspect, the present invention relates to a computer-implemented method for operating a laser material processing machine, wherein an estimated result of laser material processing is determined in a simulated manner based on pre-given process parameters, particularly without operating the laser material processing machine, the estimated result characterizing how good the actual result of the laser material processing is under the process parameters, and wherein the process parameters are modified using a data-based model by means of Bayesian optimization until the actual result of the laser material processing is sufficiently good, the model being configured to estimate the result of the laser material processing based on the process parameters.

[0053] This can be accomplished by determining the value of the cost function based on estimated or actual variables, where the estimated variables characterize the estimated outcome of laser material processing and the actual variables characterize the actual outcome of laser material processing, and then determining whether this value of the cost function is below a pre-defined threshold. The variables characterizing the estimated or actual outcome of laser material processing can here characterize the events and / or processes generated by the laser material processing.

[0054] Here, the value of the cost function can be determined based on how much the estimated or actual variables differ from the target variable characterizing the target result of the laser material processing.

[0055] Bayesian optimization allows for the rapid determination of optimal values ​​within a predefined parameter range, without the need to determine the gradient. Determining the gradient requires numerous practical steps in laser material processing, and due to unavoidable experimental noise, it can only be reliably determined through difference quotients. To minimize this noise, a large number of trials are necessary, which can be avoided by using Bayesian optimization. Furthermore, Bayesian optimization enables the determination of the global optimum.

[0056] To minimize the number of actual laser material processing steps required, the process parameters can be continuously varied until the estimated results are sufficiently good before testing the actual laser material processing results against those process parameters. In other words, the actual experiment used to determine the actual results is only performed if simulation experiments indicate that good actual (i.e., experimental) results can be expected.

[0057] Then, the data-based model can be trained based on the actual results, that is, based on the actual variables that characterize the actual results.

[0058] In particular, the data-based model can be trained and updated based on the estimation results, i.e., based on the estimated variables that characterize the estimation results.

[0059] Although the estimation results are inadequate, it may be advantageous to use them to train the data-based model, thereby reducing the actual number of laser material processing steps required.

[0060] To suppress potential erroneous training of the data-based model, it can be stipulated that the estimation result (y) sim If the model is good enough, that is, close enough to the optimization objective, then the data-based model is trained not based on the estimated results, but only on the actual results.

[0061] As mentioned at the beginning, the data-based model can advantageously be a Gaussian process model. This allows for particularly targeted modification of the process parameters because, in addition to the estimated results, the uncertainties of the estimated results and the uncertainties of the actual results, especially those due to noise, can be identified and accounted for.

[0062] Alternatively or additionally, the estimation result may be determined by means of a physical model of laser material processing, wherein it may be specified that if the evaluation of the physical model will be performed with parameters outside a pre-given range, the estimation result is determined by means of the data-based model. This allows for the compensation in a particularly simple manner for any known deficiencies of the physical model.

[0063] It should be understood that the estimation results may include multiple variables. In this case, it can be specified that the data-based model is a multidimensional model, or multiple one-dimensional data-based models corresponding to multiple variables are used, or a mixture of one-dimensional and multidimensional models is used.

[0064] Since physical simulation models can sometimes only predict a subset of features relevant to optimization, the value of the estimated result can still be determined using heuristic methods. Therefore, in another aspect, it is stipulated that the estimated result is determined by means of a physical model evaluated under pre-given process parameters and by means of actual results determined under other process parameters. Attached Figure Description

[0065] Embodiments of the present invention will now be explained in more detail with reference to the accompanying drawings. In the drawings:

[0066] Figure 1 The structure of a laser drilling machine is schematically shown.

[0067] Figure 2 The structure of a laser welding machine is schematically shown;

[0068] Figure 3 The structure of the test bench is schematically shown;

[0069] Figure 4 A flowchart illustrates the implementation method for running the test bench;

[0070] Figure 5 A flowchart illustrates an implementation method for running the test bench. Detailed Implementation

[0071] Figure 1 The structure of the laser drilling machine 1 is schematically shown. Control signal A is provided by control logic 40 to control the laser 10a. The laser beam is incident on a material block 12, and the laser beam creates a hole 11 in the material block.

[0072] Figure 2 The structure of the laser welding machine 2 is schematically shown. Control signals A are also provided by control logic 40 to control the laser 10b. The laser beam is incident on two material blocks 13 and 14, where a weld 15 is produced.

[0073] A laser cutting machine (not shown) can also be conceived in a similar way.

[0074] Figure 3 The structure of the test bench 3 for determining the optimal process parameters x is schematically shown. The current process parameters x are provided by a parameter memory P via an output interface 4 of a laser material processing machine (e.g., a laser drilling machine 1 or a laser welding machine 2). The laser material processing machine performs laser material processing based on these provided process parameters x. Sensor 30 determines sensor variables S characterizing the laser material processing results. These sensor variables S are transmitted via an input interface 50 as quality characteristics y. exp Provided to machine learning block 60.

[0075] In this embodiment, machine learning block 60 includes a Gaussian process model, such as Figure 4 and Figure 5 As shown, based on the provided quality characteristic y exp The Gaussian process model is then trained. Based on the Gaussian process model, varying process parameters x' can be provided, and these process parameters are stored in the parameter memory P.

[0076] Alternatively, or additionally provided via output interface 4, the process parameter x can also be provided to the estimation model 5, which provides the estimated quality characteristic y to the machine learning block 60. sim rather than the actual quality characteristic y exp .

[0077] In this embodiment, the test bench includes a processor 45 configured to execute a computer program stored on a computer-readable storage medium 46. The computer program includes instructions that, when executed, cause the processor 45 to perform... Figure 4 or Figure 5 The method shown is described above. This computer program can be implemented in software, hardware, or a combination of both.

[0078] Figure 4 An exemplary method for running test bench 3 is illustrated in a flowchart. This is achieved by initializing the process parameter x... init The method begins 100 by providing process parameters x and initializing the recorded trial data to date as an empty set. Optionally, process parameters x are pre-defined using experimental design methods, and these process parameters x are used to control laser material processing machines 1 and 2, as described in more detail below, to determine the variable y. exp The Gaussian process is trained using the trial data determined in this way.

[0079] In the case of laser drilling, in one embodiment these process parameters x include pulse duration and / or time-resolved focal position and / or focal size and / or pulse repetition frequency and / or time-resolved circular orbit diameter and / or circular orbit frequency and / or time-resolved positioning angle and / or drilling duration and / or time-resolved pulse energy and / or wavelength and / or parameters characterizing the process shielding gas, such as process gas type or process gas pressure. The circular orbit mentioned herein is a feature known in many drilling methods, such as in auger drilling or open-hole drilling.

[0080] In the case of laser welding, these process parameters x include laser power and / or focal diameter and / or focal position and / or welding speed and / or laser beam tilt and / or the circular orbit frequency of the laser oscillation, which are time-dependent and / or location-dependent, and / or parameters characterizing the process shielding gas, through a family of characteristic curves.

[0081] Using the current process parameter x, control laser material processing machines 1 and 2 (110), and determine the variable y (120) characterizing the actual result of laser material processing. exp .

[0082] In the case of laser drilling, these variables y exp In one embodiment, variables characterizing the size and / or roundness and / or wall shape of the borehole 11 and / or the presence of melt deposits and / or large amounts of droplet ejection during the drilling process and / or edge rounding and / or productivity of the borehole 11 are included.

[0083] In the case of laser welding, in another embodiment, these variables y exp Variables include those characterizing the minimum weld depth and / or minimum weld width and / or productivity along weld 15 and / or a certain number of weld spatter and / or a certain number of porosity and / or weld distortion and / or weld stress and / or weld cracks.

[0084] The cost function K is evaluated based on these variables, which can be given, for example, by equation (1), where the variable y is used to evaluate the cost function K. exp As a feature q i And the corresponding target values ​​q of these variables i,Ziel supply.

[0085] Alternatively, a cost function K can be considered that penalizes deviations from the target value, particularly when the deviation exceeds a pre-defined tolerance, and rewards high productivity. The "penalty" can be achieved, for example, by using a high value for the cost function K, while the "reward" can be correspondingly achieved by using a low value.

[0086] Then it is determined whether the cost function K indicates that the current process parameter x is good enough; in cases where the penalty implies a high value and the reward implies a low value, the way to do this is to check whether the cost function K is below the maximum pre-given cost value of 140. If this is the case ("yes"), the method ends with the current process parameter x of 150.

[0087] If this is not the case ("No"), then the process parameter x and the associated variable y characterizing the result will be determined in this way. exp The data points (x, y) constitute exp Add 160 to the determined trial data, and then retrain the Gaussian process, i.e., adapt the hyperparameters Θ0, Θ of the Gaussian process. d This maximizes the probability of obtaining the trial data from the Gaussian process.

[0088] The 170 acquisition function is then evaluated, as exemplarily shown in Equation (7), and the new process parameter x' is determined thereby. The branch then returns to step 110.

[0089] Figure 5 Another exemplary method for running test bench 3 is illustrated in a flowchart. Steps 100 to 170 are related to... Figure 4 The same applies as shown, so a separate description is omitted.

[0090] However, after the new process parameters x' have been determined, the 180 simulation model is invoked using these new process parameters x' to determine the estimated variable y. sim Instead of the actual variable y exp .

[0091] In the case of laser drilling, this can be done as follows: For the radius r of borehole 11 along the depth coordinate z, r(z) is numerically determined as a solution to the following equation:

[0092]

[0093] in

[0094]

[0095] .

[0096] Here:

[0097] The refractive index of material block 12 is a predefined complex number, where the refractive index is n and the extinction coefficient is k.

[0098] The pre-defined removal threshold flux for material block 12.

[0099] Q is the pre-defined pulse energy of the 10a laser.

[0100] d Fok It is the pre-defined focal diameter of laser 10a.

[0101] l Rayleigh The Rayleigh length of the 10a laser can be predefined.

[0102] R is the reflectivity of the determined material block 12.

[0103] α is the angle of the determined local beam propagation direction.

[0104] θ is a predefined relative angle between the incident laser beam and the surface normal of the material block 12.

[0105] F0 is the determined radiative flux of the laser 10a.

[0106] w(z) is the determined local beam radius.

[0107] Using this model, some features cannot be predicted, such as the presence of melt deposits and / or large amounts of droplet ejection during the drilling process. To determine these features, an empirical model can be pre-defined, or the results can be determined from values ​​experimentally determined up to that point in time, such as the average of all these values, or the experimentally determined values ​​can be weighted according to the distance between the current process parameters and those process parameters that have determined their corresponding experimentally determined actual values. In particular, predictions from a Gaussian process trained on the actual variables can be used as estimates.

[0108] Alternatively or additionally, it may be impossible to reliably calculate at least some characteristics for all process parameters x. It is possible to check whether the current process parameter x is within a pre-given range, and if not, to determine these characteristics using one of the methods described above.

[0109] In the case of laser welding, the estimated variable y can be determined, for example, as follows: sim :

[0110]

[0111] in

[0112]

[0113] and parameters

[0114] T0 – Predefined ambient temperature;

[0115] x0 — A predefined offset from the origin of the coordinate system to which the laser 10b beam can move with the laser 10b.

[0116] λ — the predefined thermal conductivity of material blocks 13 and 14;

[0117] a——The predefined thermal diffusivity of material blocks 13 and 14;

[0118] q net —Predictable power for laser 10b;

[0119] q 1net —Predictable power distribution of laser 10b along the depth coordinates of material block 10b

[0120] v — the pre-defined velocity of laser 10b;

[0121] h—the predefined thickness of material blocks 13 and 14; and

[0122] Bessel function And the determined temperature distribution T(x, y, z). The width or depth of the weld can be determined from the temperature distribution (e.g., by determining isotherms at the melting temperatures of the materials in material blocks 13 and 14).

[0123] Then, similar to step 130, the cost function K is determined, where the variable y, estimated in a simulation manner, is used. sim Instead of the variable y determined experimentally exp .

[0124] Then, similar to step 140, the current process parameter x of 200 is checked by means of the cost function K, where a second maximum cost value can be used instead of the pre-given maximum cost value, which is greater than the pre-given maximum cost value.

[0125] If the check indicates that the current process parameter x is good enough, then proceed back to step 110. Otherwise, proceed back to step 160.

Claims

1. A computer-implemented method for operating a laser material processing machine (1, 2), comprising: Laser material processing was performed using the first process parameters, and the actual results characterizing the laser material processing (y) were detected. exp The actual variable; Based on the first process parameters and the detected actual variables, the process parameters are changed by means of Bayesian optimization to determine the second process parameters; Based on the second process parameters determined using Bayesian optimization, an estimate characterizing the laser material processing is determined (y). sim The estimated variables; The cost function is evaluated based on the estimated variables; If the cost function exceeds a threshold, then based on the second process parameters and the estimated variables, Bayesian optimization is further used to change the process parameters to determine the third process parameters; or If the evaluated cost function is below the threshold, the laser material is processed using the second process parameters.

2. The method according to claim 1, wherein, Based on the actual results (y) exp To train a data-based model.

3. The method of claim 2, wherein the estimation result (y) is further used as a further step. sim (This is used to train the data-based model.) 4. The method according to any one of claims 2 to 3, wherein, The data-based model is a Gaussian process model.

5. The method according to any one of claims 2 to 3, wherein, The estimation result (y) is determined using a physical model of the laser material processing. sim ).

6. The method according to claim 5, wherein, If the evaluation of the physical model will be performed with parameters (x) outside a pre-defined range, then the estimation result (y) is determined by means of the data-based model. sim ).

7. The method according to claim 5, wherein, Using a physical model evaluated under pre-given process parameters (x) and actual results (y) determined under other process parameters (x') exp ) to determine the estimation result (y) sim ).

8. The method according to any one of claims 1 to 3, wherein, The laser material processing machine is a laser drilling machine (1).

9. The method according to claim 8, characterized in that, To characterize the estimation results (y) sim ) and / or to characterize the actual results (y exp ), using variables that characterize the geometry of the hole (11) drilled by the laser drilling machine (1).

10. The method according to any one of claims 1 to 3, wherein, The laser material processing machine is a laser welding machine (2).

11. The method according to claim 10, characterized in that, To characterize the estimation results (y) sim ) and / or to characterize the actual results (y exp ), using variables that characterize the geometry of the weld (15) welded by the laser welding machine (2).

12. The method according to any one of claims 1 to 3, wherein, After setting the process parameters (x), the laser material processing machine (1, 2) is run using the process parameters (x) set in this way.

13. A test bench (3) for a laser material processing machine (1, 2), said test bench being configured to perform the method according to any one of claims 1 to 11.

14. A computer program configured to perform the method according to any one of claims 1 to 11.

15. A machine-readable storage medium having a computer program as claimed in claim 14 stored thereon.

Images