Machine learning device, machine learning method, and laser processing system

Abstract

The present invention provides a machine learning device, a machine learning method, and a laser processing system. The machine learning device learns the focus position deviation of a laser processing machine. A data acquisition unit acquires data of a focus position instruction of a focusing optical system provided to the laser processing machine, and detection data of a physical quantity of light detected when a laser is output from a laser oscillator according to a processing instruction including the focus position instruction, as a learning data set. The learning unit uses the learning data set to generate a learning model that represents the correlation between the physical quantity of the detected light and the following positional relationship, which is the positional relationship between a workpiece and an effective focusing position of the focusing optical system. When executing a processing procedure, the physical quantity of light is detected in the same manner as during learning, thereby being able to infer the positional relationship between the workpiece and the effective focusing position when executing the processing procedure based on the detection value and the learning model, while also considering differences in processing conditions and the workpiece.

Description

Technical Field

[0001] The present invention relates to a machine learning device and a machine learning method for learning a focus position deviation of a laser processing machine, and also to a laser processing system for correcting the focus position deviation. Background Art

[0002] In a laser processing machine having a focusing optical system including lenses, reflectors, protective windows, etc., the focal distance of the focusing optical system changes due to interference such as heat associated with laser irradiation, and the position of the focusing point of the laser light that is actually focused (referred to as the effective focusing position of the focusing optical system in this application) sometimes deviates from the designed focal position obtained by calculation (referred to as the focal position of the focusing optical system in this application). If the effective focusing position of the focusing optical system deviates from the focal position during the processing process, the positional relationship between the workpiece and the effective focusing position deviates from the target positional relationship determined by the processing program, which may affect the quality of laser processing.

[0003] Conventionally, there is known a technique for correcting a deviation of an effective focusing position of a laser processing machine from a focal position (referred to as focal position deviation in the present application).

[0004] For example, Japanese Patent Publication No. 2002-239768 (JP2002-239768A) states the following (claim 1, paragraph 0003): "The optical path or optical area of ​​the laser (3) on the processed surface (4a) is determined. When the optical path or optical area deviates from a predetermined reference value, the focal position of the laser (3) deviates from the set position serving as the reference, and the laser head (2) and the workpiece (4) are relatively moved in accordance with the deviation amount." "The laser 13 irradiated on the processed surface 14a is photographed by a camera 16, and the deviation between the working distance and the focal distance, that is, the deviation between the laser irradiation position and the focal position, is detected based on the position where the photographed laser 13 appears on the camera screen, and the working distance is controlled based on the deviation."

[0005] Japanese Patent Publication No. 2005-334928 (JP2005-334928A) records the following (abstract): "The laser beam output from the laser oscillator 550 of the laser processing machine is reflected by the output mirror 560 to the reflection mirror 570 and introduced into the laser processing tool 60. The laser beam focused by the processing lens 62 is reflected by the reflection plate M1 and detected by the reflected light detection device 580. Automatic adjustment of the focus is achieved by checking the change in the focal position caused by the output of the reflected light and the contamination of the processing lens 62.".

[0006] Japanese Patent Publication No. 2016-002580 (JP2016-002580A) states the following (claim 1): "For a measuring reference surface having a small opening, a processing head is positioned so that the laser optical axis passes through the small opening, and the laser is irradiated with the focus on the measuring reference surface or the front or rear side thereof, and the level of the radiated light emitted from at least a portion of the periphery of the small opening of the measuring reference surface is measured, and the focus offset is calculated based on the measured value, and the focus position of the laser is corrected." Summary of the invention

[0007] The positional relationship between the workpiece and the effective focusing position of the focusing optical system when the workpiece is processed by a laser processing machine with a focusing optical system can be estimated while also taking into account processing conditions and differences in the workpiece, thereby improving the correction accuracy of the focus position deviation of the laser processing machine.

[0008] One method of the present disclosure provides a machine learning device that learns the positional relationship between a workpiece and an effective focusing position of a focusing optical system when the workpiece is processed by a laser processing machine having a focusing optical system, and comprises: a data acquisition unit that acquires data of a focus position instruction provided to the focusing optical system of the laser processing machine, and detection data of a physical quantity of light detected when a laser is output from a laser oscillator in accordance with a processing instruction including the focus position instruction, as a learning data set; and a learning unit that uses the learning data set to generate a learning model that represents the correlation between the physical quantity of light and the positional relationship.

[0009] Another embodiment of the present invention provides a laser processing system, which comprises: a laser processing machine, which has a laser oscillator, a focusing optical system for focusing laser light output by the laser oscillator, and a focus position control unit for controlling the movement of the focus position of the focusing optical system; a machine learning device of the above-mentioned embodiment; a light detector, which detects the physical quantity of light sensed when laser light is output from the laser oscillator in accordance with a processing instruction including a focus position instruction; a position relationship estimating unit, which uses a learning model generated by the learning unit and the physical quantity of light detected by the light detector to estimate the positional relationship between a workpiece and an effective focusing position of the focusing optical system during processing of the workpiece; and a position instruction correction unit, which corrects the focus position instruction provided to the laser processing machine based on the positional relationship estimated by the position relationship estimating unit.

[0010] Another embodiment of the present invention provides a machine learning method for learning the positional relationship between a workpiece and an effective focusing position of a focusing optical system when the workpiece is processed by a laser processing machine having a focusing optical system. A processor obtains data of a focus position instruction of the focusing optical system provided to the laser processing machine, and detection data of a physical quantity of light detected when a laser is output from a laser oscillator in accordance with a processing instruction including the focus position instruction, as a learning data set, and uses the learning data set to generate a learning model representing the correlation between the physical quantity of light and the positional relationship.

[0011] According to a machine learning device of one embodiment, when a laser processing machine having a focusing optical system processes a workpiece, the positional relationship between the workpiece and the effective focusing position of the focusing optical system can be associated with the physical quantity of light detected according to the processing conditions and the difference of the workpiece, and the positional relationship can be learned. When executing the processing process, the physical quantity of light is detected in the same way as when learning, so that the positional relationship between the workpiece and the effective focusing position when light is detected (i.e., when executing the processing process) can be estimated based on the detection value and the learning model generated by the learning unit, taking into account the processing conditions and the difference of the workpiece.

[0012] According to another embodiment of the laser processing system, when executing the processing step, the position instruction correction unit corrects the focus position instruction according to the position relationship estimated by the position relationship estimation unit, thereby complementing the above-mentioned effect of the machine learning device and improving the correction accuracy of the focus position deviation of the laser processing machine. As a result, the quality of the laser processing performed by the laser processing machine is prevented from deteriorating, and the life of the structural components of the focusing optical system is extended.

[0013] According to another aspect of the machine learning method, an effect equivalent to the above-mentioned effect of the machine learning device is produced. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The purpose, features and advantages of the present disclosure will become more apparent through the following description of the embodiments in conjunction with the accompanying drawings.

[0015] Figure 1 This is a functional block diagram showing a machine learning device according to one embodiment.

[0016] Figure 2 The overall structure of the laser processing system is schematically shown.

[0017] Figure 3 is a flowchart representing a machine learning method according to one embodiment.

[0018] Figure 4 This is a functional block diagram showing a machine learning device according to another embodiment.

[0019] Figure 5A An example of light detection data is shown.

[0020] Figure 5B Another example of light detection data is shown.

[0021] Figure 5C Still another example of light detection data is shown.

[0022] Figure 5D Still another example of light detection data is shown.

[0023] Figure 5E Still another example of light detection data is shown.

[0024] Fig. 6A Still another example of light detection data is shown.

[0025] Figure 6B Still another example of light detection data is shown.

[0026] Figure 6C Still another example of light detection data is shown.

[0027] Fig.6D Still another example of light detection data is shown.

[0028] Fig. 6E Still another example of light detection data is shown.

[0029] Fig. 7A Illustration of a neuron.

[0030] Figure 7B Describe the neural network.

[0031] Figure 8 This is a functional block diagram showing a laser processing system according to one embodiment.

[0032] Fig. 9 This is a functional block diagram showing a laser processing system according to another embodiment.

[0033] Fig.10 This is a functional block diagram showing a laser processing system according to another embodiment. DETAILED DESCRIPTION

[0034] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. In all the drawings, corresponding components are denoted by common reference numerals.

[0035] Figure 1 The machine learning device 10 according to one embodiment is represented by functional blocks. Figure 2 The overall structure of the laser processing system 12 having the machine learning device 10 is schematically shown. Figure 2 The basic structure of the laser processing system 12 will be described.

[0036] The laser processing system 12 includes: a machine learning device 10, a laser processing machine 14, and a photodetector 16. The laser processing machine 14 includes: a laser oscillator 18; a focusing optical system 20 that focuses the laser light L output by the laser oscillator 18; a control device 22 that controls the oscillation of the laser oscillator 18 and controls the focal position of the focusing optical system 20 with respect to the workpiece W; and a driving mechanism 24 that moves the focusing optical system 20 and the workpiece W relative to each other according to a command from the control device 22. The laser processing machine 14 may also include a processing head that accommodates the focusing optical system 20, a gas supply mechanism that supplies auxiliary gas to the processing head, a workbench that carries the workpiece W, and other structural elements not shown in the figure.

[0037] The laser oscillator 18 has, for example, a resonance unit (not shown) of a fiber laser method. The laser light L amplified by the resonance unit is introduced into the processing head via, for example, a transmission unit 26 composed of an optical fiber. In addition, the laser oscillator 18 can adopt various oscillation methods such as CO2 laser, and the transmission unit 26 can have various structures using a light guide, a reflector, etc.

[0038] The focusing optical system 20 includes a lens 28 built into the processing head and a protective window 30. When the transmission unit 26 is composed of an optical fiber, the focusing optical system 20 further includes an optical component 32 such as a collimator installed at the end of the optical fiber. The reflector and the like as the transmission unit 26 can also be regarded as a part of the focusing optical system 20.

[0039] The drive mechanism 24 can, for example, relatively move the processing head and the worktable in a direction along the surface of the workpiece W, or selectively move in a direction of approaching and leaving each other. In addition, the drive mechanism 24 can also move, for example, the lens 28 of the focusing optical system 20 in the processing head. The drive mechanism 24 can, for example, have at least three control axes (X axis, Y axis, Z axis) in an orthogonal three-axis coordinate system. In this case, each control axis can be set to one or both of the focusing optical system 20 (or processing head) and the workpiece W (or worktable).

[0040] The control device 22 has, for example, a structure of a numerical control device. The control device 22 interprets the given processing program P to send instructions to the control objects such as the laser oscillator 18 and the drive mechanism 24, so that the laser oscillator 18 oscillates the laser at the desired oscillation conditions (output, frequency, pulse width, etc.), and the focusing optical system 20 operates at the desired operation conditions (focal position, focusing diameter, processing speed, auxiliary gas pressure, etc.). The control device 22, for example, operates at least three control axes (X axis, Y axis, Z axis) of the drive mechanism 24 respectively through the focus position instruction of the orthogonal three-axis coordinate system, thereby enabling the focus of the focusing optical system 20 to move relative to the surface of the workpiece W in three dimensions. In addition, the focus position instruction can be an instruction value of a relative position relative to the workpiece W, or can be an instruction value of an absolute position defining an origin at an arbitrary position.

[0041] When the laser light L output from the laser oscillator 18 is irradiated to the workpiece W through the focusing optical system 20, the photodetector 16 detects the light sensed in the processing environment including the laser processing machine 14 and the workpiece W. The light to be detected by the photodetector 16 can be the reflected light of the laser light L irradiated to the workpiece W, and can also be the generated light (such as plasma, thermal radiation, etc.) generated by irradiating the laser light L to the workpiece W. The physical quantity Q of light measured by the photodetector 26 can be the radiation flux (W), radiation energy (J), light quantity (Im·s), illuminance (W / m 2 ) etc., and can also be a spectrum (light intensity distribution) as a function of light frequency. In a structure in which the light detector 16 detects the reflected light of the laser L, the light detector 16 can use, for example, a reflected light sensor installed in the laser processing machine 14 for the purpose of protecting the laser oscillator 18.

[0042] In the laser processing machine 14, the focal distance of the focusing optical system 20 may change due to interference such as heat associated with laser irradiation, and the actual effective focusing position F of the focusing optical system 20 may deviate from the designed focal position. If the effective focusing position of the focusing optical system 20 deviates from the focal position during the processing, the positional relationship R between the workpiece W and the effective focusing position F deviates from the target positional relationship specified by the processing program P, which may affect the quality of laser processing.

[0043] The machine learning device 10 self-learns the positional relationship R between the workpiece W and the effective focusing position F of the focusing optical system 20 when the workpiece W is processed by a laser processing machine 14 having a focusing optical system 20 through so-called machine learning. The machine learning device 10 has hardware such as a processor of an electronic computer (such as a central processing unit (CPU)) for executing software that defines various learning algorithms. The positional relationship R learned by the machine learning device 10 is equivalent to an indicator representing the current state of the focus position deviation of the laser processing machine 14. The machine learning device 10 generates a learning model M that associates the positional relationship R with the physical quantity Q of light, thereby being able to accurately estimate and correct the focus position deviation based on the learning model M when executing the processing step. Below, refer to Figure 1 as well as Figure 2 The structure of the machine learning device 10 will be described.

[0044] like Figure 1 As shown in the functional blocks in , the machine learning device 10 includes: a data acquisition unit 34 that acquires data D1 of a focus position instruction provided to the focusing optical system 20 of the laser processing machine 14, and detection data D2 of the physical quantity Q of light detected when the laser oscillator 18 outputs the laser light L according to the processing instruction including the focus position instruction, as a learning data set DS; and a learning unit 36 ​​that uses the learning data set DS to generate a learning model M representing the correlation between the physical quantity Q of the detected light and the positional relationship R. The machine learning device 10 may have a structure of an electronic computer such as a personal computer. Alternatively, the machine learning device 10 may be installed in the above-mentioned control device 22.

[0045] The data acquisition unit 34 may be configured as, for example, a processor of an electronic computer (one function thereof), or may be configured as, for example, a program (a part thereof) for causing the processor of an electronic computer to function. The data D1 of the focus position instruction in the learning data set DS acquired by the data acquisition unit 34 (hereinafter also referred to as the focus position instruction data D1) is acquired from the processing program P provided to the control device 22. The focus position instruction data D1 is equivalent to the instruction value of the focus position (relative position or absolute position) instructed to the following laser processing machine 14, which has the focusing optical system 20 assumed to have no focus position deviation.

[0046] In addition, the detection data D2 of the physical quantity Q of light in the learning data set DS (hereinafter, also referred to as light detection data D2) is obtained from the light detector 16 as the actual measured value of the light sensed in the processing environment when the laser light L is irradiated to the workpiece W under the processing instruction of the focus position instruction including the focus position instruction data D1. The light detection data D2 is obtained from the light detector 16 as various different values ​​(physical quantity Q) corresponding to the differences in processing conditions such as the oscillation conditions (output, frequency, pulse width, etc.) of the laser oscillator 18 instructed by the processing program P, the operating conditions (focal position, focusing diameter, processing speed, assist gas pressure, etc.) of the focusing optical system 20, and the differences in the types of the workpiece W (shape, surface state, material, temperature, etc.).

[0047] The learning unit 36 ​​may be configured as, for example, a processor of an electronic computer (one function thereof), or may be configured as, for example, a program (a part thereof) for causing the processor of an electronic computer to function. The learning unit 36 ​​performs a computational process for learning the positional relationship R between the workpiece W and the effective focusing position F of the focusing optical system 20 according to an arbitrary learning algorithm generally referred to as machine learning.

[0048] As described above, the light detection data D2 acquired by the data acquisition unit 34 represents the physical quantity Q of light corresponding to the processing conditions including the focus position and the type of the workpiece W. In particular, it is considered that the focus position instruction data D1 has a linear correlation with the light detection data D2. In addition, it is assumed that the processing conditions other than the focus position and the type of the workpiece W also have a certain correlation with the light detection data D2. The learning unit 36 ​​learns the correlation between the two data D1 and D2 based on the focus position instruction data D1 and the light detection data D2 detected in response to the focus position instruction. By repeatedly executing this learning cycle for a variety of workpieces W under a variety of processing conditions, the learning unit 36 ​​can generate a learning model M that represents the correlation between the physical quantity Q of light and the positional relationship R.

[0049] For example, even when the laser is irradiated under the same processing conditions with respect to the workpiece W of the same material and size, if the shape, state, and temperature of the irradiated surface of the workpiece W are different, the physical quantity Q of the detected light is also different. In addition, when the laser L is actually irradiated under various processing conditions with respect to the same type of workpiece W, various physical quantities of light Q are detected even if the focal position is the same. Therefore, by using various learning data sets DS obtained by irradiating the laser L with respect to various workpieces W of different shapes, surface states, materials, temperatures, etc., under a plurality of processing conditions with different oscillation conditions of the laser oscillator 18 and operation conditions of the focusing optical system 14, a highly reliable learning model M can be generated.

[0050] Even if it is assumed that the correlation between the physical quantity Q of light and the positional relationship R is essentially unknown at the beginning of learning, the learning unit 36 ​​gradually explains the correlation as the learning proceeds. If the correlation between the physical quantity Q of light and the positional relationship R is explained to a level that can be trusted to a certain extent, the learning model M generated by the learning unit 36 ​​can be used to select the behavior of how much the current state (i.e., before correction) of the command value of the focus position commanded by the processing program P should be corrected for the laser processing machine 14 with the focusing optical system 20 having the current state (i.e., the focus distance may have changed due to interference such as heat). That is, as the learning algorithm proceeds, the learning unit 16 can make the learning model M representing the correlation between the physical quantity Q of light and the positional relationship R gradually approach the optimal solution. In addition, regardless of whether the focus position command data D1 is a command value of a relative position or a command value of an absolute position, the positional relationship R learned by the learning unit 36 ​​can be a positional relationship expressed by the relative position of the workpiece W and the effective focusing position F, or a positional relationship expressed by the absolute position of the effective focusing position F.

[0051] Figure 3 A machine learning method of one embodiment performed by the machine learning device 10 is represented by a flowchart. In the illustrated learning process, in step S1, a workpiece W set as a learning object is selected from a plurality of workpieces W having different current states such as shapes, surface states, materials, and temperatures. The selected workpiece W may be a workpiece W that is actually laser processed by executing a processing procedure, or may be a workpiece W prepared separately for learning. In step S2, a processing instruction for laser processing the workpiece W of the learning object is selected from various processing instructions that meet the processing conditions that the laser processing machine 14 can cope with (the oscillation conditions of the laser oscillator 18, the operating conditions of the focusing optical system 20). The selected processing instruction may be a processing instruction that actually performs laser processing on the workpiece W by executing a processing procedure, or may be a processing instruction prepared separately for learning. The selection behavior of steps S1 and S2 can be performed arbitrarily by an operator who performs learning through the machine learning device 10 in the laser processing system 12.

[0052] According to the selected processing instruction, in step S3, the laser oscillator 18 is made to oscillate the laser light L, and the laser light L is irradiated to the workpiece W of the learning object through the focusing optical system 20. The operator of the laser processing system 12 places the workpiece W of the learning object on the workbench, and provides the processing program P recording the selected processing instruction to the control device 22, thereby executing step S3. In step S4, as a learning data set DS, the data D1 of the focus position instruction provided to the focusing optical system 20 of the laser processing machine 14 and the detection data D2 of the physical quantity Q of the light detected when the laser oscillator 18 outputs the laser light according to the processing instruction including the focus position instruction are obtained. In step S4, the data acquisition unit 34 of the machine learning device 10 obtains the focus position instruction data D1 from the processing program P, and obtains the light detection data D2 from the light detector 16.

[0053] Using the acquired learning data set DS, in step S5, a learning model M representing the correlation between the physical quantity Q of light and the positional relationship R is generated. In step S5, the learning unit 36 ​​of the machine learning device 10 learns the correlation between the focus position instruction data D1 and the light detection data D2 according to the learning algorithm, and generates the learning model M as a result. In step S6, the operator determines whether the learning data set DS used to generate the learning model M is sufficient, that is, whether the learning model M reaches a reliable level. When it is determined that the learning data set DS is sufficient, the learning process ends.

[0054] When it is determined that the learning data set DS is insufficient, in step S7, the operator determines whether to change the processing instruction. When it is determined that the processing instruction is to be changed, the process returns to step S2, a processing instruction different from the previous one is selected, and the subsequent learning process is continued. When it is determined that the processing instruction is not to be changed, in step S8, the operator determines whether to change the workpiece W. When it is determined that the workpiece W is to be changed, the process returns to step S1, a workpiece W different from the previous one is selected, and the subsequent learning process is continued. When it is determined that the workpiece W is not to be changed, the learning process is terminated in a state where the learning model M does not reach a reliable level.

[0055] The above-mentioned learning process is performed using at least a plurality of different learning data sets DS including a plurality of different focus position command data D1. The number of different focus position command data D1 or the spacing between positions is set according to the reliability (precision and accuracy) required of the learning model M. The number of light detection data D2 combined with each focus position command data D1 (the number corresponding to the difference in processing conditions other than the focus position and the workpiece W) is determined in consideration of the time and cost allowed for learning in addition to the reliability required of the learning model M.

[0056] As described above, according to the machine learning device 10, the positional relationship R between the workpiece W and the effective focusing position F of the focusing optical system 20 when the workpiece W is processed by the laser processing machine 14 having the focusing optical system 20 can be associated with the physical quantity Q of light detected according to the processing conditions and the difference of the workpiece W, and the positional relationship R can be learned. When the processing process is executed, the physical quantity Q of light is detected in the same way as when learning, and thus the positional relationship R between the workpiece W and the effective focusing position F at the time of light detection (that is, when the processing process is executed) can be estimated based on the detected value and the learning model M generated by the learning unit 36, taking into account the difference of the processing conditions and the workpiece W. Therefore, when the focus position is deviated in the focusing optical system 20 due to interference such as heat associated with laser irradiation, the focus position instruction in the processing process can be corrected based on the positional relationship R estimated from the learning model M to offset the focus position deviation. If the focus position instruction can be appropriately corrected, the quality of laser processing by the laser processing machine 14 can be prevented from being deteriorated. Furthermore, by eliminating focus position deviation and continuously performing laser processing at the effective focusing position F as the target, it is possible to reduce contamination and overheating of the structural components (especially lenses and protective windows) of the focusing optical system 20 and extend the life of the structural components.

[0057] When executing a processing step, when processing the same workpiece W as the workpiece W for learning by the processing instruction (excluding the focus position instruction) that was used when the learning model M was generated (i.e., for learning) and processing is performed, the positional relationship R when executing the processing step can be accurately estimated by using the learning model M. Even if the processing instruction or the workpiece W in the actual processing step is different from the processing instruction or the workpiece W for learning, the reliability of the learning model M is improved by repeatedly executing the above-mentioned learning process, thereby making it possible to estimate the positional relationship R when executing the processing step.

[0058] As described above, one strategy for improving the reliability of the learning model M is to use a variety of workpieces W with different shapes, surface conditions, materials, temperatures, etc. for learning. As another strategy, the surface of the workpiece W can be processed in advance by the laser processing machine 14 before obtaining the learning data set DS, and the light detection data D2 when the processed surface is irradiated with the laser L can be used to generate the learning model M. According to this strategy, the surface state of the workpiece W is uniformed by surface processing in advance, thereby reducing the noise in the light detection data D2 and improving the reliability of the learning model M.

[0059] As another strategy for improving the reliability of the learning model M, a variety of processing conditions and workpieces W are grouped into processing conditions and workpieces W that are similar to a certain extent in advance to set a plurality of condition groups, and by performing learning with each condition group, a plurality of learning models M corresponding to each other can be generated. When executing a processing step, the learning model M corresponding to the condition group to which the processing conditions and workpieces W in the processing step belong is selected from the plurality of learning models M, thereby accurately estimating the positional relationship R when executing the processing step.

[0060] In the machine learning device 10 having the above structure, the learning method performed by the learning unit 36 ​​is not particularly limited, and for example, supervised learning, unsupervised learning, reinforcement learning, and other learning methods known as machine learning can be used. Supervised learning is a learning method in which a known learning data set is given in advance with a large amount of input data and output data corresponding to the input data (generally referred to as training data, labels, etc.), and features that imply the correlation between the input data and the output data are identified from these known learning data sets, thereby generating a correlation model (learning model M in the machine learning device 10) for estimating the desired output relative to the new input. In contrast, unsupervised learning and reinforcement learning start learning from a state where the relationship between the input and the output is unknown.

[0061] Figure 4 The machine learning device 40 of another embodiment of the present disclosure that performs supervised learning is represented by a functional block. Figure 4 as well as Figure 2 The structure of the machine learning device 40 will be described. The machine learning device 40 has the same structure as the machine learning device 10 described above except that supervised learning is performed as a learning method, and the description of the corresponding structure will be omitted as appropriate.

[0062] The machine learning device 40 includes: a data acquisition unit 42 that acquires data D1 of a focus position command of the focusing optical system 20 provided to the laser processing machine 14 and detection data D2 of a physical quantity Q of light detected when the laser oscillator 18 outputs the laser light L according to the processing command including the focus position command, as a learning data set DS; and a learning unit 44 that generates a learning model M representing the correlation between the physical quantity Q of the detected light and the positional relationship R using the learning data set DS. The learning unit 44 learns the correlation between the light detection data D2 and the training data D1 from a known learning data set DS using the focus position command data D1 as the training data D1. The learning model M is generated by repeatedly executing this learning cycle using a variety of known learning data sets DS.

[0063] The learning unit 44 includes: an error calculation unit 46, which compares the learning model M generated at the current time point with the correlation characteristics between the training data D1 and the light detection data D2 identified from the known learning data set DS to be obtained next, and calculates the error E between the two; and a model updating unit 48, which updates the learning model M to reduce the error E. The learning unit 44 deepens the learning of the positional relationship R by repeatedly updating the learning model M by the model updating unit 48, thereby improving the reliability of the generated learning model M. Figure 3 In the illustrated learning flow, the learning unit 44 updates the learning model M in step S5 .

[0064] According to the machine learning device 40 that performs supervised learning, in addition to the effects produced by the above-mentioned machine learning device 10, the effect of improving learning efficiency by using a known learning data set DS and quickly generating a highly reliable learning model M is obtained. Figure 2 The laser processing system 12 may be provided with a machine learning device 40 instead of the machine learning device 10 .

[0065] The machine learning device 40 can also perform semi-supervised learning instead of supervised learning. In semi-supervised learning, based on the known learning data set DS of the training data D1 and the light detection data D2, the focus position instruction data D1 that is different from the training data D1 is intentionally matched with other light detection data D2 as a new learning data set DS. When the number of known learning data sets DS prepared in advance is small, the learning of the positional relationship R can be deepened. Alternatively, it can be configured that the learning unit 44 performs supervised learning in the initial stage of learning, and at a stage where the learning has progressed to a certain extent, the learning unit 44 performs unsupervised learning or reinforcement learning with the learning model M obtained through supervised learning as the initial value.

[0066] In the machine learning devices 10 and 40, the data acquisition units 34 and 42 can acquire data of instantaneous values ​​of light detected at a predetermined timing as light detection data D2, and can also acquire time series data of light detected within a predetermined time as light detection data D2. By acquiring time series data of light as light detection data D2, learning accuracy can be improved.

[0067] Figure 5A to Figure 5E The light detection data D2 obtained as time series data by the data acquisition units 34 and 42 are shown as examples. In each figure, the horizontal axis is time and the vertical axis is light intensity. Figure 5A to Figure 5EThe detection data D2 represents the light intensity (physical quantity Q of light) actually measured corresponding to the different focus position command data D1 (or training data D1). The processing conditions other than the focus position and the type of the workpiece W are the same. For example, when the focus position command is the command value of the relative position of the focus relative to the irradiated surface of the workpiece W, the focus position command data D1 (or training data D1) corresponding to each light detection data D2 is Figure 5A : +2.0mm, Figure 5B : +1.0mm, Figure 5C : 0.0mm, Figure 5D : -1.0mm, Figure 5E : −2.0 mm. A positive sign indicates that the focal position is located before the irradiated surface of the workpiece W as viewed from the condensing optical system 20 , and a negative sign indicates that the focal position is located far from the irradiated surface of the workpiece W as viewed from the condensing optical system 20 . Figure 5C The detection data D2 when the laser light L is focused on the irradiated surface of the workpiece W is shown.

[0068] As shown in the figure, the focus position instruction data D1 (or training data D1) and the light detection data D2 have a one-to-one linear correlation. Figure 5A to Figure 5E When the instantaneous value of the light detected at a predetermined time is obtained as the light detection data D2, the one-to-one correlation between the two data D1 and D2 may not be clear. On the other hand, if the time series data of the light within a predetermined time is obtained as the light detection data D2, then according to Figure 5A to Figure 5E It can be understood that for each focus position instruction data D1 (or training data D1), the difference in the waveform of the light detection data D2 is clear, so the one-to-one correlation between the two data D1 and D2 is clearly explained. As a result, the learning accuracy of the learning units 36 and 44 is improved, and a learning model M with high reliability can be generated.

[0069] When the machining process is performed, the detection data D2 of the physical quantity Q of light as a result of irradiating the workpiece W with the laser light L is, for example, Figure 5A When the light detection data D2 shown is substantially the same, the positional relationship R between the workpiece W and the effective focusing position F during light detection (i.e., when the processing step is executed) is estimated to be +2.0 mm (relative position). If the focus position command at this time is +1.0 mm (relative position), the effective focusing position F in the processing stroke can be accurately matched with the focus position command by correcting the focus position command by -1.0 mm.

[0070] When the time series data of light is obtained as the light detection data D2, the learning units 36 and 44 may perform any one of normalization, smoothing, Fourier transform, function approximation, and differentiation on the time series data. By performing such preprocessing, the learning units 36 and 44 can generate the learning model M using the light detection data D2 suitable for learning, thereby improving the learning efficiency.

[0071] In the machine learning devices 10 and 40, the data acquisition units 34 and 42 can acquire a plurality of data detected at a plurality of different locations of light sensed in the processing environment when the laser light L is irradiated on the workpiece W, as the light detection data D2. For example, in the laser processing system 12, the light detectors 16 can be arranged at a plurality of different locations, and the reflected light of the laser light L irradiated on the workpiece W and the generated light (plasma, thermal radiation, etc.) generated by irradiating the laser light L on the workpiece W can be simultaneously detected at different locations. By acquiring a plurality of light detection data D2 obtained by simultaneously detecting different locations of the same light, the learning accuracy can be improved.

[0072] Figure 6A to Figure 6E Detection data D2 (detection data 1 and 2 in each figure) obtained by simultaneously detecting the same light at two locations as light detection data D2 acquired by the data acquisition units 34 and 42 are illustrated. In each figure, the horizontal axis is time and the vertical axis is light intensity. Figure 6A to Figure 6E The light detection data D2 represents the light intensity (physical quantity Q of light) actually measured corresponding to the different focus position command data D1 (or training data D1). The processing conditions other than the focus position and the type of the workpiece W are the same. For example, when the focus position command is the command value of the relative position of the focus relative to the irradiated surface of the workpiece W, the focus position command data D1 (or training data D1) corresponding to each light detection data D2 (detection data 1 and 2) is Fig. 6A : -2.0mm, Figure 6B : -1.0mm, Figure 6C : 0.0mm, Fig.6D : +1.0mm, Fig. 6E : +2.0 mm. A positive sign indicates that the focal position is located before the irradiated surface of the workpiece W when viewed from the condensing optical system 20 , and a negative sign indicates that the focal position is located far from the irradiated surface of the workpiece W when viewed from the condensing optical system 20 . Figure 6C The detection data D2 when the laser light L is focused on the irradiated surface of the workpiece W is shown.

[0073] As shown in the figure, the focus position instruction data D1 (or training data D1) and the light detection data D2 (detection data 1 and 2) have a one-to-one linear correlation. Figure 6A to Figure 6EWhen comparing the light detection data D2 of the same device with the light detection data D1 (detection data 1 and 2), even if the light detection data D2 (detection data 1 and 2) is obtained as the instantaneous value data of the light detected at two locations at a predetermined time, the one-to-one correlation of the two data D1 and D2 can be explained. Furthermore, if the light detection data D2 (detection data 1 and 2) is obtained as the time series data of the light detected at two locations at a predetermined time, then according to Figure 6A to Figure 6E It can be understood that not only the waveform difference of the light detection data D2 for each focus position instruction data D1 (or training data D1) is clear, but also the waveform difference between the light detection data D2 detected at two locations simultaneously (detection data 1 and 2) is clear. Figure 5A to Figure 5E As a result, the learning accuracy of the learning units 36 and 44 is improved, and a highly reliable learning model M can be generated. As the number of simultaneous detection locations of the same light increases, the learning accuracy is expected to increase.

[0074] When the workpiece W is irradiated with the laser light L during the machining process, the detection data D2 of the physical quantity Q of the light detected at two locations simultaneously as in the learning process is, for example, Fig. 6A When the light detection data D2 shown is substantially the same, the positional relationship R between the workpiece W and the effective focusing position F during light detection (i.e., when the processing step is executed) is estimated to be -2.0 mm (relative position). When the focus position command at this time is assumed to be 0.0 mm (relative position), the subsequent focus position command is corrected by +2.0 mm, thereby enabling the effective focusing position F in the processing stroke to accurately match the focus position command.

[0075] In the machine learning devices 10 and 40, the data acquisition units 34 and 42 can acquire data detected in light of a predetermined wavelength as the light detection data D2. For example, in the laser processing system 12, when the light detector 16 is arranged at a position where both the reflected light of the laser L irradiated to the workpiece W and the generated light (plasma, thermal radiation, etc.) generated by irradiating the laser L to the workpiece W can be detected, an optical filter that transmits only one of the reflected light and the generated light (plasma, thermal radiation, etc.) can be attached to the light detector 16. By limiting the light detection data D2 used for learning to only one of the reflected light and the generated light (plasma, thermal radiation, etc.), the learning efficiency and the reliability of the learning model M can be improved.

[0076] In the machine learning devices 10 and 40, when the learning units 36 and 44 perform learning by supervised learning or the like, known learning algorithms such as decision trees and neural networks can be used. Fig. 7AA model schematically representing a neuron. Figure 7B Schematically indicating that Fig. 7A The model of a three-layer neural network is composed of a combination of neurons shown in FIG. For example, a neural network is composed of a computing device and a storage device that simulate a neuron model.

[0077] Fig. 7A The neuron shown in the figure outputs a result y for a plurality of inputs x (in the figure, inputs x1 to x3 are used as an example). Each input x (x1, x2, x3) is multiplied by a weight w (w1, w2, w3). Figure 7B The three-layer neural network shown in the figure inputs multiple inputs x from the left side (inputs x1 to x3 as examples in the figure), and outputs results y from the right side (results y1 to y3 as examples in the figure). In the example shown in the figure, the inputs x1, x2, and x3 are multiplied by corresponding weights (uniformly represented by W1), and each input x1, x2, and x3 is input to three neurons N11, N12, and N13.

[0078] exist Figure 7B In the example shown in the figure, each output of neurons N11 to N13 is collectively represented by Z1. Z1 can be regarded as a feature vector obtained by extracting the feature quantity of the input vector. In the example shown in the figure, each feature vector Z1 is multiplied by a corresponding weight (all represented by w2), and each feature vector Z1 is input to two neurons N21 and N22. The feature vector Z1 represents the feature between the weight W1 and the weight W2. In addition, Figure 7B In the figure, each output of neurons N21~N22 is collectively represented by Z2. Z2 can be regarded as a feature vector obtained by extracting the feature value of feature vector Z1. In the example shown in the figure, each feature vector Z2 is multiplied by the corresponding weight (all represented by w3), and each feature vector Z2 is input to three neurons N31, N32, and N33. The feature vector Z2 represents the feature between the weight W2 and the weight W3. Finally, neurons N31~N33 output results y1~y3 respectively. In the machine learning devices 10 and 40, the learning data set DS is used as the input x, and the learning units 36 and 44 perform operations according to the multi-layer structure of the above-mentioned neural network, thereby being able to output the learning model M (result y).

[0079] Figure 8 The laser processing system 50 of one embodiment is shown by functional blocks. The laser processing system 50 has the following basic configuration: Figure 2 The laser processing system 12 has the same structure as the laser processing system 12, and the corresponding structural elements are marked with common reference symbols and the description is appropriately omitted.

[0080] The laser processing system 50 includes a machine learning device 10, a laser processing machine 14, and a photodetector 16. The laser processing machine 14 includes a laser oscillator 18, a focusing optical system 20 for focusing laser light L output from the laser oscillator 18, and a focus position control unit 52 for controlling the focus position of the focusing optical system 20 with respect to the workpiece W. The focus position control unit 52 is equivalent to the integration of the control device 22 and the drive mechanism 24 of the laser processing system 12. The photodetector 16 detects the physical quantity Q of light sensed in the processing environment when the laser oscillator 18 outputs laser light according to the processing instruction including the focus position instruction. When the machine learning device 10 performs learning, as described above, the data acquisition unit 34 acquires the data D1 of the focus position instruction from the focus position control unit 52, and acquires the detection data D2 of the physical quantity Q of light from the photodetector 16, and the learning unit 36 ​​uses the learning data set DS including these data D1 and D2 to generate a learning model M representing the correlation between the physical quantity Q of light and the positional relationship R.

[0081] The laser processing system 50 further includes a position relationship estimation unit 54 that estimates an effective focusing position F (φ) between the workpiece W and the focusing optical system 20 during the processing of the workpiece W using the learning model M generated by the learning unit 36 ​​and the physical quantity Q of the light detected by the light detector 16. Figure 2 and a position instruction correction unit 56, which corrects the focus position instruction provided to the laser processing machine 14 based on the position relationship R estimated by the position relationship estimating unit 54.

[0082] The positional relationship estimation unit 54 may be configured as, for example, a processor of an electronic computer (one function thereof), or may be configured as a program (a part thereof) for causing the processor of an electronic computer to function. Figure 2 ) is incorporated into the positional relationship estimating unit 54. When the laser processing machine 14 performs a processing step on the workpiece W according to the processing program, the light detector 16 detects the physical quantity Q of light in the same manner as during learning, and the positional relationship estimating unit 54 compares the physical quantity Q of light detected by the light detector 16 with the learning model M generated by the learning unit 36, and estimates the positional relationship R between the workpiece W and the effective focusing position F during light detection (i.e., when the processing step is performed).

[0083] The position instruction correction unit 56 may be configured as, for example, a processor of an electronic computer (one function thereof), or, for example, may be configured as a program (a part thereof) for causing the processor of an electronic computer to function. Figure 2) is incorporated into the position instruction correction unit 56. When the laser processing machine 14 performs a processing step on the workpiece W according to the processing program, the position instruction correction unit 56 corrects the focus position instruction recorded in the processing program based on the position relationship R estimated by the position relationship estimation unit 54, so as to offset the focus position deviation that may be generated in the focusing optical system 20 due to interference such as heat associated with laser irradiation.

[0084] According to the laser processing system 50, when executing the processing step, the position command correction unit 56 corrects the focus position command based on the position relationship R estimated by the position relationship estimation unit 54, so that it complements the above-mentioned effect of the machine learning device 10 and can improve the correction accuracy of the focus position deviation of the laser processing machine 14. If the focus position command can be corrected with good accuracy, the quality of laser processing performed by the laser processing machine 14 can be prevented from being deteriorated. In addition, by eliminating the focus position deviation and continuing the laser processing at the effective focusing position F as the target, it is possible to reduce the contamination and overheating of the structural components of the focusing optical system 20, and the life of the structural components can be extended.

[0085] Furthermore, according to the laser processing system 50, when the laser processing machine 14 performs a processing step on the workpiece W according to the processing program, by comparing the physical quantity Q of light detected by the light detector 16 with the learning model M generated by the learning unit 36, the positional relationship R between the workpiece W and the effective focusing position F when the processing step is performed can be estimated in real time, and the focus position instruction can be corrected immediately as needed. The laser processing system 50 can automatically perform such detection, estimation, and correction without relying on manual labor, and can be operated unmanned for a long time without causing processing defects. In this case, the learning unit 36 ​​uses the physical quantity Q of light detected during the processing step as the light detection data D2, so that the machine learning device 10 can update the learning model M at any time.

[0086] In the laser processing system 50, as part of the setup operation immediately before the processing step, the detection of the physical quantity Q of light, the estimation of the positional relationship R, and the correction of the focus position instruction can also be performed. If the focus position instruction can be corrected by the setup operation, the workpiece W can be processed with high quality without waste. In this case, it is also possible to use a workpiece W dedicated to the setup operation (i.e., the correction of the focus position instruction) differently from the workpiece W actually processed. By laser processing the workpiece W dedicated to the correction under the same conditions as the various conditions during learning, the accuracy of the comparison between the detected physical quantity Q of light and the learning model M is improved, and the estimation accuracy of the positional relationship R of the positional relationship estimation unit 54 can be improved. In addition, the laser processing system 50 can also be equipped with a machine learning device 40 instead of the machine learning device 10.

[0087] Fig. 9Another embodiment of a laser processing system 60 is shown by functional blocks. The laser processing system 60 is Figure 8 The laser processing system 50 has additional functions, and the corresponding components are denoted by common reference numerals and the description thereof is appropriately omitted.

[0088] The laser processing system 60 includes the components of the laser processing system 50 and further includes a notification unit 62. The notification unit 62 generates a signal when the position relationship estimation unit 54 estimates the focus position command and the effective focusing position F( Figure 2 ) exceeds a predetermined threshold.

[0089] The notification unit 62 may be configured as, for example, a processor of an electronic computer (one function thereof), or may be configured as a program (a part thereof) for causing the processor of an electronic computer to function. Figure 2 ) is incorporated in the laser processing system 60. When the laser processing machine 14 performs a processing step on the workpiece W according to the processing program, when the focus position instruction corresponding to the estimated position relationship R deviates too much from the effective focusing position F, the notification unit 62 can, for example, cause the machine learning device 10 or the control device 22 to produce a sound, light, image display, etc. that can be perceived by humans. The threshold value can be set in advance by the designer or operator of the laser processing system 60 based on experience. For example, when the contamination and deterioration of the structural components of the focusing optical system 20 develop to the extent that maintenance, replacement, etc. are required, by setting an appropriate threshold value, the notification action of the notification unit 62 can prompt the operator to prepare or perform such operations, resulting in a shortened system downtime.

[0090] The laser processing system 60 may further include a network 66 for establishing communication between the notification unit 62 and peripheral devices 64 of the laser processing machine 14. Even when an operator of the laser processing system 60 is not near the laser processing machine 14, the operator can perceive the notification operation of the notification unit 62 at a remote location, for example, via the peripheral devices 64 such as a monitor or a personal computer connected to the network 66.

[0091] Fig.10 Another embodiment of a laser processing system 70 is shown by functional blocks. The laser processing system 70 is Figure 8 The laser processing system 50 has additional functions, and the corresponding components are denoted by common reference numerals and the description thereof is appropriately omitted.

[0092] The laser processing system 70 includes, in addition to the components of the laser processing system 50, a command unit 72 that provides, as a processing command, a command value for learning or estimation that has a lower energy than the command value for processing the workpiece W. Here, the command value provided by the command unit 72 to the laser oscillator 18 is a value for determining the energy of the laser light emitted from the laser oscillator 18, and is, for example, a value related to an oscillation condition (output, frequency, pulse width, etc.) that the laser oscillator 18 can cope with.

[0093] The instruction unit 72 may be configured as, for example, a processor of an electronic computer (one function thereof), or may be configured as a program (a part thereof) for causing the processor of an electronic computer to function. Figure 2 ) is incorporated into the instruction unit 72. When the machine learning device 10 performs the above-mentioned learning, the instruction unit 72 emits laser light from the laser oscillator 18 using the instruction value for learning, and the instruction value for learning is used to instruct the emission of laser light with lower energy than the instruction value generally used in the processing step of the workpiece W. If the surface of the workpiece W is intentionally processed over a period of time using the low-energy laser light for learning, the time series data of the light detected by the light detector 16 (for example Figure 5A to Figure 6E ) becomes data whose changes are easy to identify, which promotes the learning unit 36 ​​to interpret the correlation between the physical quantity Q of light and the position relationship R. Alternatively, when the laser processing machine 14 performs a processing step on the workpiece W according to the processing program, the command unit 72 emits laser light from the laser oscillator 18 using an estimated command value, and the estimated command value is used to instruct the emission of laser light with lower energy than the command value generally used in the processing step of the workpiece W. If the surface of the workpiece W is intentionally processed by the estimated low-energy laser light for a period of time, the time series data of the light detected by the light detector 16 (for example Figure 5A to Figure 6E ) becomes data whose change is easy to identify, so that the estimation accuracy of the positional relationship R when the processing step is executed by the positional relationship estimation unit 54 is improved. In addition, unnecessary processing of the workpiece W to be processed as the target of the processing step is avoided.

[0094] In the laser processing system 70, the command unit 72 can provide the command value to the laser oscillator 18 while changing the command value for learning or estimation at any time. By changing the laser energy during learning, the learning data set DS is increased, thereby promoting the learning of the learning unit 36. In addition, by changing the laser energy during estimation, the comparison accuracy and precision of the physical quantity Q of light detected when executing the processing step and the learning model M are improved, and the estimation accuracy of the position relationship estimation unit 54 is improved.

[0095] In the laser processing system 70, the focus position control unit 52 can move the focus position of the focusing optical system 20 at any time while the instruction unit 72 provides the instruction value for learning or estimation to the laser oscillator 18. By moving the focus position during learning, the learning data set DS is increased, which promotes the learning of the learning unit 36. In addition, by moving the focus position during estimation, the accuracy and precision of the comparison between the physical quantity Q of light detected when executing the processing step and the learning model M are improved, and the estimation accuracy of the position relationship estimation unit 54 is improved.

[0096] In the laser processing systems 60 and 70 , basically the same effects as those of the laser processing system 50 are obtained. In addition, the laser processing systems 60 and 70 may include the machine learning device 40 instead of the machine learning device 10 .

[0097] Although the embodiments of the present disclosure are described above, it will be understood by those skilled in the art that various modifications and changes can be made without departing from the scope of the disclosure of the protection scope.

Claims

1. A machine learning device for learning the positional relationship between a workpiece and an effective focusing position of a focusing optical system when the workpiece is processed by a laser processing machine having a focusing optical system, characterized in that: The machine learning device comprises: a data acquisition unit that acquires, as a learning data set, data of a focus position command provided to the laser processing machine and detection data of a physical quantity of light detected when a laser oscillator outputs laser light in accordance with a processing command including the focus position command; as well as a learning unit, which uses the learning data set to generate a learning model representing the correlation between the physical quantity and the positional relationship, wherein the learning unit uses the data of the focus position instruction as training data and generates the learning model by finding the correlation between the detection data and the training data, The data acquisition unit acquires time series data of the light detected within a predetermined time as the detection data every time a plurality of different focus position commands are provided to the laser processing machine. The data acquisition unit acquires the data of each of the focus position instructions and the time series data corresponding to each of the focus position instructions as the learning data set. The learning model is compared with time series data of the physical quantity of light within a predetermined time, and estimates the positional relationship having the correlation with the time series data.

2. The machine learning device according to claim 1, characterized in that The physical quantity is the light intensity of reflected light of the laser beam irradiated to the workpiece, or the light intensity of generated light generated by irradiating the laser beam to the workpiece.

3. The machine learning device according to claim 1 or 2, characterized in that: Preset multiple condition groups that group various processing conditions. The data acquisition unit detects, for each of the plurality of condition groups, the time series data of the light detected when the laser oscillator outputs laser light in accordance with the processing instruction satisfying the processing condition included in the condition group as the detection data, The learning unit generates a plurality of the learning models for each of the plurality of condition groups.

4. The machine learning device according to claim 1, characterized in that The data acquisition unit acquires time series data of the light detected within a predetermined time as the detection data.

5. The machine learning device according to claim 4, characterized in that The learning unit uses data obtained by performing any one of normalization, smoothing, Fourier transformation, function approximation, and differentiation on the time series data as the detection data.

6. The machine learning device according to claim 1, characterized in that The data acquisition unit acquires a plurality of data detected at a plurality of different locations of the light as the detection data.

7. The machine learning device according to claim 1, characterized in that The data acquisition unit acquires data detected in the light of a predetermined wavelength as the detection data.

8. A laser processing system, characterized in that: have: A laser processing machine comprising a laser oscillator, a focusing optical system for focusing laser light outputted from the laser oscillator, and a focus position control unit for controlling movement of a focus position of the focusing optical system; The machine learning device according to any one of claims 1 to 7; a photodetector for detecting a physical quantity of light sensed when the laser light is output from the laser oscillator in accordance with a processing instruction including a focus position instruction; a positional relationship estimating unit that estimates a positional relationship between a workpiece and an effective focusing position of the focusing optical system during processing of the workpiece using the learning model generated by the learning unit and the physical quantity detected by the photodetector; as well as A position command correction section corrects the focus position command provided to the laser processing machine based on the positional relationship estimated by the positional relationship estimation section.

9. The laser processing system according to claim 8, characterized in that: The laser processing system further includes a notification unit configured to perform a notification operation when a difference between the focus position command and the effective focusing position when the positional relationship estimating unit estimates the positional relationship exceeds a predetermined threshold value.

10. The laser processing system according to claim 9, characterized in that: The laser processing system further includes a network for establishing communication between the notification unit and peripheral devices of the laser processing machine.

11. The laser processing system according to any one of claims 8 to 10, characterized in that: The laser processing system further includes a command unit that provides, as the processing command, a command value for learning or estimation having lower energy than a command value for processing a workpiece to the laser oscillator.

12. The laser processing system according to claim 11, characterized in that: The focus position control unit moves the focus position of the condensing optical system while the command unit is supplying the learning or estimation command value to the laser oscillator.

13. A machine learning method for learning the positional relationship between a workpiece and an effective focusing position of a focusing optical system when the workpiece is processed by a laser processing machine having a focusing optical system, characterized in that: The processor obtains, as a learning data set, data of a focus position instruction of the focusing optical system provided to the laser processing machine and detection data of a physical quantity of light detected when a laser is output from a laser oscillator in accordance with a processing instruction including the focus position instruction, and The processor generates a learning model representing the correlation between the physical quantity and the positional relationship using the learning data set, and uses the data of the focus position instruction as training data to generate the learning model by finding the correlation between the detection data and the training data. Whenever a plurality of different focus position instructions are provided to the laser processing machine, the processor acquires time series data of the light detected within a predetermined time as the detection data, The processor obtains the data of each of the focus position instructions and the time series data corresponding to each of the focus position instructions as the learning data set, The learning model is compared with time series data of the physical quantity of light within a predetermined time, and estimates the positional relationship having the correlation with the time series data.

14. The machine learning method according to claim 13, characterized in that: The learning model is generated by processing the surface of the workpiece in advance by the laser processing machine before acquiring the learning data set, and using the detection data when the processed surface is irradiated with the laser.

15. The machine learning method according to claim 13 or 14, characterized in that: The learning model is generated using a plurality of workpieces having different shapes, surface conditions, materials or temperatures.

Images