JP7415097B1 - Control device, laser processing system, and laser processing method - Google Patents

Abstract

The control device (3) includes a learned model holding unit (26) that holds a plurality of learned models that are the results of learning the relationship between the feature amount indicating the state of processing by the processing machine and the state of processing by the processing machine; A machining state evaluation unit (23) that evaluates the state of machining by the processing machine by inputting feature quantities to each of the plurality of trained models or by inputting feature quantities to a combination of the plurality of learned models. and a monitoring model used to monitor the machining state by the processing machine based on the results of evaluating the machining state by observing the workpiece machined by the processing machine and the evaluation results by the machining state evaluation unit (23). A monitoring model determination unit (24) that determines a trained model or a combination of a plurality of trained models; and a processing control unit (24) that controls the processing machine based on the result of monitoring the state of processing by the processing machine using the monitoring model. 28) and.

Description

The present disclosure relates to a control device that controls a laser processing machine, a laser processing system, and a laser processing method.

A control device that controls a laser beam machine reads out machining conditions set in advance, and controls the laser beam machine in accordance with the read machining conditions. The control device is preset with multiple machining conditions that are associated with the workpiece material, workpiece thickness, type of machining gas, etc., and the machining conditions suitable for the machining are selected from the multiple machining conditions. Ru.

The processing conditions that enable good processing with a laser processing machine may vary from the preset processing conditions due to individual differences in workpieces depending on the workpiece manufacturer or workpiece lots. be. Therefore, the control device may monitor the state of machining by the laser beam machine, and adjust machining parameters used for machining based on the monitoring results. A laser processing machine can perform good processing by appropriately adjusting processing parameters.

Patent Document 1 discloses a numerical control device that detects light emitted from the vicinity of a processing point of a workpiece and evaluates the quality of the processing state of the workpiece based on the detected light intensity distribution. The numerical control device disclosed in Patent Document 1 generates a learned model by learning the relationship between the detected light intensity distribution and the quality of the machining state, and uses the learned model to determine the machining state of the workpiece. Evaluate the quality.

Japanese Patent Application Publication No. 2022-118774

The state of light emitted from the vicinity of the processing point of the workpiece when the processing state is good may change depending on the state of the laser processing machine or variations in the components contained in the workpiece. While the state of light may change in this way, in the case of the conventional technology disclosed in Patent Document 1, a learned model generated in advance is used to evaluate the quality of the machining state. It may not be possible to accurately assess the quality of the condition. According to the conventional technique disclosed in Patent Document 1, there is a problem in that it may not be possible to accurately evaluate the quality of the machining state, and therefore it may not be possible to continue good machining using a laser beam machine.

The present disclosure has been made in view of the above, and an object of the present disclosure is to obtain a control device that can continue good processing by a laser processing machine.

In order to solve the above-mentioned problems and achieve the purpose, a control device according to the present disclosure provides a learned control device that is a result of learning the relationship between a feature value indicating the state of processing by a processing machine and the state of processing by the processing machine. A trained model holding unit that holds multiple models, and processing by a processing machine by inputting feature quantities to each of the multiple trained models, or by inputting feature quantities to a combination of multiple trained models. A machining state evaluation unit that evaluates the state of the workpiece, and a machining state evaluation unit that evaluates the machining state by the processing machine based on the results of evaluating the machining state by observing the workpiece machined by the processing machine and the evaluation results by the machining state evaluation unit. A monitoring model determination unit that determines a trained model or a combination of multiple trained models as the monitoring model to be used, and a processing unit that controls the processing machine based on the result of monitoring the processing state of the processing machine using the monitoring model. A control unit.

The control device according to the present disclosure has the effect of allowing the laser processing machine to continue good processing.

A diagram showing a configuration example of a laser processing system according to Embodiment 1. A diagram showing a configuration example of a control device included in the laser processing system according to Embodiment 1. A diagram showing a configuration example of a trained model in Embodiment 1 Flowchart showing a first example of processing procedures executed by the laser processing system according to the first embodiment Flowchart showing a second example of processing procedures executed by the laser processing system according to the first embodiment Flowchart showing a third example of processing procedures executed by the laser processing system according to the first embodiment A diagram illustrating an example configuration of a monitoring model determination unit included in a control device according to a second embodiment. A diagram showing an example of the configuration of a control device included in the laser processing system according to Embodiment 3. A diagram illustrating a configuration example of a monitoring model determination unit included in a control device according to Embodiment 3. A diagram showing a configuration example of a control circuit according to Embodiments 1 to 3. A diagram showing a configuration example of a dedicated hardware circuit according to Embodiments 1 to 3.

Below, a control device, a laser processing system, and a laser processing method according to an embodiment will be described in detail based on the drawings.

Embodiment 1.
FIG. 1 is a diagram showing a configuration example of a laser processing system 1 according to the first embodiment. The laser processing system 1 includes a laser processing machine 2 and a control device 3. The laser processing machine 2 is a processing machine that processes the workpiece 10 using a laser beam. The laser processing machine 2 processes the work 10 by locally melting the work 10 by irradiating the work with laser light. The laser processing machine 2 performs laser processing such as cutting, welding, lamination processing, or heat treatment. In the example shown in FIG. 1, the workpiece 10 is a metal plate. Further, it is assumed that the laser processing machine 2 performs laser processing to cut the workpiece 10. The control device 3 controls the laser processing machine 2.

The laser processing machine 2 includes a laser oscillator 4 that is a light source that outputs laser light, a processing head 6, a cable 5 that is a transmission path for the laser light from the laser oscillator 4 to the processing head 6, and a table on which a workpiece 10 is placed. Equipped with. Illustration of the table is omitted. The cable 5 includes, for example, an optical fiber. The output end of the cable 5 is placed inside the processing head 6.

FIG. 1 schematically shows the internal configuration of the processing head 6. As shown in FIG. Inside the processing head 6, a collimating optical system 11, an imaging optical system 12, and a protective glass 13 are provided. Laser light diverging from the output end of the cable 5 enters the collimating optical system 11 . The collimating optical system 11 emits parallel light. The imaging optical system 12 forms an image of the output end of the cable 5. The protective glass 13 is a flat glass plate. The protective glass 13 is an optical component for protecting the internal structure of the processing head 6. Note that FIG. 1 shows a collimating optical system 11 that is a single lens and an imaging optical system 12 that is a single lens. At least one of the collimating optical system 11 and the imaging optical system 12 may be composed of a plurality of lenses.

The processing head 6 has a processing nozzle 7 through which laser light to be irradiated onto the workpiece 10 and processing gas to be injected to the workpiece 10 pass. The laser beam that has passed through the collimating optical system 11, the imaging optical system 12, and the protective glass 13 passes through the processing nozzle 7. Processing gas is supplied into the processing head 6 from a gas supply source outside the processing head 6 . The laser processing machine 2 injects processing gas from a processing head 6 to a workpiece 10 via a processing nozzle 7. Illustration of the gas supply source is omitted.

The laser processing machine 2 includes a processing head drive section 14 that drives the processing head 6. In the laser processing machine 2, the processing head drive unit 14 moves the processing head 6 with respect to the table, thereby moving the laser beam and the workpiece 10 relative to each other. The laser processing machine 2 relatively moves the laser beam and the workpiece 10 by controlling the incident position of the laser beam on the workpiece 10. Note that the laser processing machine 2 may be one in which the laser beam and the workpiece 10 are moved relative to each other by moving the table relative to the processing head 6 without moving the processing head 6.

Inside the processing head 6, a moving mechanism 15 for moving the collimating optical system 11 and a moving mechanism 16 for moving the imaging optical system 12 are provided. The moving mechanism 16 changes the imaging position by moving the imaging optical system 12 in the optical axis direction. By changing the imaging position, the laser processing machine 2 changes the positional relationship between the imaging position and the workpiece 10 without changing the positional relationship between the processing nozzle 7 and the workpiece 10. The moving mechanism 15 adjusts the position of the collimating optical system 11 by moving the collimating optical system 11 in the optical axis direction as the imaging optical system 12 moves. Alternatively, the moving mechanism 15 may adjust the divergence angle of the laser light incident on the imaging optical system 12 by moving the collimating optical system 11 in the optical axis direction. Note that the optical axis direction in the above description is the direction of each optical axis of the collimating optical system 11 and the imaging optical system 12. The optical axis of the collimating optical system 11 and the optical axis of the imaging optical system 12 coincide with each other. These optical axes coincide with the central axis of the laser beam incident on the workpiece 10.

Optical systems other than the collimating optical system 11 and the imaging optical system 12 may be provided inside the processing head 6. A zoom optical system that changes the size of an image by moving in the optical axis direction may be provided inside the processing head 6. The laser processing machine 2 moves at least one of a plurality of optical systems provided inside the processing head 6 in the optical axis direction. The imaging position may be a position on the beam waist or a position shifted from the beam waist. A condensing optical system may be provided inside the processing head 6 instead of the imaging optical system 12. In the following description, the optical systems provided inside the processing head 6 may be collectively referred to as a processing optical system.

An example of the laser oscillator 4 is a fiber laser oscillator. The laser oscillator 4 may be a direct diode laser, a carbon dioxide laser, a copper vapor laser, or various ion lasers. Alternatively, the laser oscillator 4 may be a solid laser having a YAG (Yttrium Aluminum Garnet) crystal or the like as an excitation medium. The laser processing machine 2 may include a wavelength conversion section that converts the wavelength of the laser beam output from the laser oscillator 4.

The control device 3 outputs control signals to each of the laser oscillator 4, the processing head drive unit 14, the movement mechanism 15, and the movement mechanism 16, thereby controlling the movement of the laser oscillator 4, the processing head drive unit 14, and the movement mechanism 16. Each of the mechanism 15 and the moving mechanism 16 is controlled.

The laser processing system 1 includes a sound sensor 8 and an optical sensor 9. Each of the sound sensor 8 and the optical sensor 9 is a sensor that detects the state of processing by the laser processing machine 2. The sound sensor 8 detects the sound generated near the processing point and outputs a signal indicating the magnitude of the detected sound to the control device 3. The optical sensor 9 detects light emitted from the vicinity of the processing point and outputs a signal indicating the intensity of the detected light to the control device 3. The processing point is a position on the workpiece 10 where processing is being performed, and is a position where the laser beam is incident.

In the following description, data obtained by each of the sound sensor 8 and the optical sensor 9 will be referred to as sensor data. The sensor data obtained by the sound sensor 8 is data indicating the volume of sound generated during processing. The sensor data obtained by the optical sensor 9 is data indicating the intensity of light emitted during processing. Each of the sensor data obtained by the sound sensor 8 and the sensor data obtained by the optical sensor 9 is data indicating the state of processing by the laser processing machine 2.

Note that the laser processing system 1 may include sensors other than the sound sensor 8 and the optical sensor 9. The laser processing system 1 may include, for example, a vibration sensor that detects vibrations generated near the processing point, or a camera that photographs the vicinity of the processing point. Sensor data obtained by the vibration sensor is data indicating the magnitude of vibrations generated during processing. The control device 3 acquires data indicating the dimensions of the processed portion, data indicating the molten state of the work 10, etc. by processing the image taken by the camera. The sensor data obtained by the camera is data obtained by processing an image photographed near the processing point. Laser processing system 1 includes at least one of a sound sensor 8, a light sensor 9, a vibration sensor, and a camera. The laser processing system 1 may include two or more of a sound sensor 8, a light sensor 9, a vibration sensor, and a camera. The sensor included in the laser processing system 1 may be any sensor that can detect the state of processing by the laser processing machine 2, and is not limited to the sensor described in the first embodiment.

In the example shown in FIG. 1, the optical sensor 9 is arranged outside the processing head 6. The optical sensor 9 may be placed inside the processing head 6. Since the optical sensor 9 has a short response time, the laser processing system 1 can detect the processing state at high speed by providing the optical sensor 9. Since the sound sensor 8 can detect sound over a wide range, it has the advantage that the installation position can be easily determined.

Next, the configuration of the control device 3 will be explained. FIG. 2 is a diagram showing a configuration example of the control device 3 included in the laser processing system 1 according to the first embodiment. The control device 3 includes a processing monitoring unit 20 that monitors the state of processing by the laser processing machine 2 and a processing control unit 28 that controls the laser processing machine 2.

When the laser processing machine 2 performs continuous processing, processing defects may occur due to heat accumulation in the parts constituting the processing head 6 or heat accumulation in the workpiece 10. The processing monitoring unit 20 monitors processing defects by evaluating the processing state of the laser beam machine 2.

In the laser processing system 1 shown in FIG. 1, the control device 3 is a device external to the laser processing machine 2. The laser processing machine 2 and the control device 3 are communicably connected to each other, and information is transmitted and received between the laser processing machine 2 and the control device 3. The control device 3 may be a device included in the laser processing machine 2. Alternatively, the processing control section 28 of the control device 3 may be realized by a device included in the laser beam machine 2, and the processing monitoring section 20 of the control device 3 may be realized by a device external to the laser beam machine 2. When the processing monitoring section 20 is implemented by a device external to the laser processing machine 2, the device implementing the processing monitoring section 20 and the device implementing the processing control section 28 are connected to each other so as to be able to communicate with each other.

The machining monitoring unit 20 includes a data acquisition unit 21, a feature value calculation unit 22, a machining state evaluation unit 23, a monitoring model determination unit 24, a machining parameter correction unit 25, a learned model holding unit 26, and a monitoring model A holding section 27 is provided.

The data acquisition unit 21 acquires sensor data output from each of the sound sensor 8 and the optical sensor 9. That is, the data acquisition unit 21 acquires sensor data that is a detection result of light generated during processing by the laser processing machine 2 and sensor data that is a detection result of sound generated during processing by the laser processing machine 2. do. By continuously inputting sensor data from each of the sound sensor 8 and the optical sensor 9, the data acquisition unit 21 acquires sensor data that is time-series data. The data acquisition unit 21 outputs the acquired sensor data to the feature amount calculation unit 22.

The feature amount calculation unit 22 calculates a feature amount from the sensor data acquired by the data acquisition unit 21. In the first embodiment, the feature amount indicates the state of processing by the laser processing machine 2. The feature amount calculation unit 22 outputs the calculated feature amount to the machining state evaluation unit 23.

The feature quantity is a value obtained by processing sensor data, and is, for example, a statistical quantity such as an average value or a standard deviation. The feature amount may be a value obtained by processing such as frequency analysis, filter bank, or wavelet transform. The feature quantity calculation unit 22 may calculate a combination of a plurality of values as a feature quantity.

Note that the feature amounts calculated by the feature amount calculation unit 22 are not limited to those exemplified here. The feature amount may be any feature amount that can be calculated by any method that is a general analysis method for time-series data. Further, the feature amount calculation unit 22 may output the value of the sensor data as it is as the feature amount. The feature amount calculation unit 22 may store the position in the feature space of a feature vector indicating the feature amount at the time of starting machining, and may output the amount of change in the position of the feature vector after the start of machining as the feature amount. When the data acquisition unit 21 acquires sensor data of a plurality of sensors, the feature quantity calculation unit 22 may calculate a feature quantity that reflects the output of each sensor, or a feature quantity that reflects a combination of the outputs of the plurality of sensors. Alternatively, the feature amount may be calculated.

The trained model holding unit 26 holds a plurality of trained models. Each of the plurality of learned models is a result of learning the relationship between the feature amount indicating the state of machining by the laser beam machine 2 and the state of machining by the laser beam machine 2 . Each of the plurality of trained models held in the trained model holding unit 26 is a plurality of cases in which at least one of the thickness of the plate-shaped workpiece 10 and the type of processing gas during processing by the laser processing machine 2 are different from each other. This is the result of learning the relationship between the feature amount and the machining state for each. Thereby, the control device 3 holds learned models for a plurality of cases in which the thickness of the workpiece 10 or the type of processing gas differs from each other in the learned model holding unit 26.

The machining state evaluation section 23 reads out a plurality of learned models from the learned model holding section 26. The machining state evaluation unit 23 evaluates the machining state of the laser beam machine 2 by inputting the feature quantities calculated by the feature quantity calculation unit 22 to each of the plurality of learned models. The machining state evaluation unit 23 evaluates the machining state by determining whether the machining state is good or bad. That is, the machining state evaluation unit 23 determines whether the machining state by the laser beam machine 2 is good or bad. The machining state evaluation section 23 outputs the evaluation result of the machining state to the monitoring model determination section 24 .

The evaluation results by the machining state evaluation section 23 are input to the monitoring model determination section 24 . In addition to the evaluation results by the machining state evaluation section 23 , evaluation results obtained by observing the workpiece 10 processed by the laser processing machine 2 are input to the monitoring model determination section 24 . The monitoring model determination unit 24 determines a learned model, which is a monitoring model, based on the results of evaluating the machining state by observing the workpiece 10 processed by the laser processing machine 2 and the evaluation results by the machining state evaluation unit 23. decide. In the example described here, the user observes the workpiece 10 processed by the laser processing machine 2 to evaluate the processing state. The monitoring model determination unit 24 selects a trained model whose evaluation result outputted by the machining state evaluation unit 23 matches the evaluation result by the user, from among the plurality of trained models held in the learned model storage unit 26. is determined as the monitoring model.

The trained model determined as the monitoring model may be selected by the user or may be selected by the monitoring model determination unit 24. When the user selects a trained model determined as a monitoring model, the user checks the evaluation results output by the machining state evaluation unit 23 for each of the plurality of trained models, and One trained model whose evaluation result outputted in , matches the evaluation result by the user is selected. Alternatively, the user selects one trained model whose evaluation result outputted by the machining state evaluation unit 23 is closest to the evaluation result by the user.

The user inputs information indicating the selected trained model to the monitoring model determining unit 24 by operating an input device provided in the control device 3, for example. In FIG. 2, illustration of input devices is omitted. The monitoring model determining unit 24 determines the learned model selected by the user as the monitoring model. Note that the laser processing system 1 may be provided with a display device that displays the evaluation results by the processing state evaluation section 23. The user can select a trained model by checking the evaluation results displayed on the display device. In FIG. 2, illustration of the display device is omitted.

When the trained model determined as the monitoring model is selected by the monitoring model determining unit 24, the user inputs the result of evaluating the state of machining by the laser beam machine 2 to the monitoring model determining unit 24. The user inputs the results of evaluating the machining state to the monitoring model determining unit 24, for example, by operating an input device included in the control device 3. The monitoring model determining unit 24 selects one learned model whose evaluation result outputted by the machining state evaluation unit 23 matches the evaluation result by the user from among the plurality of learned models. Alternatively, the monitoring model determining unit 24 selects one learned model whose evaluation result outputted by the machining state evaluation unit 23 is closest to the evaluation result by the user from among the plurality of learned models. The monitoring model determining unit 24 determines the learned model selected by the monitoring model determining unit 24 as the monitoring model.

The monitoring model determining unit 24 stores the determined monitoring model in the monitoring model holding unit 27. The monitoring model holding unit 27 holds a monitoring model. The control device 3 determines the monitoring model by comparing the evaluation result by the user with the evaluation result output by the machining state evaluation unit 23, thereby obtaining a monitoring model that can perform the same evaluation as the evaluation by the user. can.

The processing monitoring section 20 monitors the state of processing by the laser processing machine 2 using the monitoring model determined by the monitoring model determining section 24. When the processing monitoring unit 20 monitors the state of processing by the laser processing machine 2, the data acquisition unit 21 acquires sensor data output from each of the sound sensor 8 and the optical sensor 9. The feature amount calculation unit 22 calculates a feature amount from the sensor data acquired by the data acquisition unit 21.

When the processing monitoring section 20 monitors the state of processing by the laser processing machine 2, the processing state evaluation section 23 reads out the monitoring model from the monitoring model holding section 27. The machining state evaluation section 23 evaluates the machining state of the laser beam machine 2 by inputting the feature amount calculated by the feature amount calculation section 22 into the monitoring model. The machining state evaluation section 23 outputs the evaluation result of the machining state to the machining parameter correction section 25.

The machining parameter correction section 25 calculates the amount of correction of the machining parameters based on the evaluation result by the machining state evaluation section 23. In the first embodiment, the processing parameters are parameters used for processing by the laser beam machine 2. The machining parameter correction unit 25 outputs information on the calculated correction amount to the machining control unit 28.

The processing control unit 28 adjusts processing parameters according to processing conditions. Upon acquiring the information on the correction amount, the processing control unit 28 corrects the processing parameters based on the correction amount. The processing control unit 28 controls each of the laser oscillator 4, the processing head drive unit 14, the moving mechanism 15, and the moving mechanism 16 according to the corrected processing parameters.

In the above description, the machining condition evaluation unit 23 determines whether the machining condition by the laser beam machine 2 is good or bad. In this case, the machining state evaluation unit 23 outputs information indicating whether the machining state is good or bad. Whether the item is good or bad is expressed by a binary value. The machining state evaluation unit 23 outputs a value indicating the determination result. Note that the evaluation by the machining state evaluation unit 23 is not limited to determining whether the machining state is good or bad. The machining state evaluation unit 23 may evaluate the machining state by calculating an evaluation value indicating the degree to which the machining state is good. An example of the evaluation value is a value within the range from 0% to 100%. For example, the evaluation value is assumed to be higher as the processing condition is better. The evaluation value may be a value within the range of 0 to 10. The processing state evaluation unit 23 outputs the calculated evaluation value.

The machining state evaluation unit 23 may set a plurality of items regarding the machining state and output evaluation results for each item. For example, when determining whether the machining state is good or bad, the machining state evaluation unit 23 determines whether the machining state is good or bad for each item. The machining state evaluation unit 23 outputs determination results for each item.

Examples of items regarding processing conditions include generation of dross or roughness of cut surfaces. Dross is a molten substance that adheres to the workpiece 10 when the workpiece 10 is cut. Dross adheres to the vicinity of the lower end of the cut surface of the workpiece 10. It can be said that the less dross there is, the better the processing condition is. The roughness of the cut surface is periodic unevenness that occurs on the upper part of the cut surface. When roughness occurs on the cut surface, the depth of the striations generated on the cut surface becomes deeper than when no roughness occurs. The less rough the cut surface is, the better the processing condition is.

When the processing gas used in laser processing is oxygen, an oxide film is formed on the cut surface. When the processing gas used in laser processing is oxygen, peeling of the oxide film that occurs on the cut surface may be included in the item regarding the processing state. The less peeling of the oxide film that occurs on the cut surface, the better the processing condition.

Items regarding processing conditions are not limited to the above items. Items regarding the state of machining may include items such as discoloration of the workpiece 10 or the presence or absence of vibrations on the surface of the workpiece 10. The machining state evaluation unit 23 can output evaluation results for any one or more items among the plurality of items described in the first embodiment. The machining state evaluation unit 23 may change the items to be evaluated depending on the machining conditions and the like. For example, the machining state evaluation unit 23 may change the items to be evaluated depending on the combination of laser output, machining speed, and thickness of the workpiece 10. The thickness of the workpiece 10 is defined as the thickness in the direction of the central axis of the laser beam incident on the workpiece 10.

Alternatively, the machining state evaluation unit 23 may change the items to be evaluated depending on the type of machining gas. For example, when the processing gas used in laser processing is oxygen, the processing state evaluation unit 23 includes peeling of the oxide film as an item to be evaluated. On the other hand, when the processing gas used in laser processing is nitrogen, no oxide film is formed on the cut surface, so the processing state evaluation unit 23 excludes the peeling of the oxide film from the items to be evaluated. Regarding the laser processing machine 2 that performs welding processing, items regarding the processing state may include items such as occurrence of spatter. Regarding the laser processing machine 2 that performs lamination processing, the items regarding the processing state may include items such as the height of the object or excessive melting of the object.

The machining state evaluation unit 23 may determine the machining state for two or more items, and output the overall result of the determination for each item as the machining state determination result. For example, when the machining state is determined to be poor for more than a preset number of items, the machining state evaluation unit 23 may output a determination result indicating that the machining state is poor. Alternatively, the machining state evaluation unit 23 may analyze the quality of each item when it is determined that the machining state is poor.

As described above, the laser processing system 1 may include a display device that displays the evaluation results by the processing state evaluation section 23. The display device may be provided in the laser processing machine 2 or may be provided in a device external to the laser processing machine 2. For example, the display device displays information indicating whether the processing status is good or bad. The display device may display the evaluation result by the machining state evaluation section 23 only when the machining state evaluation section 23 determines that the machining state is poor. In this case, the display device displays information indicating that the processing condition is poor, but does not display information indicating that the processing condition is good.

In the above description, the machining state evaluation unit 23 evaluates the machining state by the laser processing machine 2 by inputting the feature quantities calculated by the feature quantity calculation unit 22 to each of the plurality of learned models. . The machining state evaluation unit 23 may evaluate the machining state by the laser beam machine 2 by inputting information other than the feature amount calculated by the feature amount calculation unit 22 to each of the plurality of learned models. Examples of information input to each of the plurality of trained models include values of processing parameters at the time of evaluation, temperature of the processing optical system at the time of evaluation, temperature change of the processing optical system, thickness of the workpiece 10, or This is information such as materials. In this case, each of the plurality of learned models is the result of learning the relationship between such information, feature amounts, and processing state. Further, the machining state evaluation unit 23 evaluates the machining state by the laser processing machine 2 by inputting such information and feature amounts to the monitoring model.

The machining parameter correction section 25 calculates the amount of correction of the machining parameters based on the evaluation result by the machining state evaluation section 23. For example, when a determination result indicating that the state of machining by the laser beam machine 2 is poor is input to the machining parameter correction unit 25, the machining parameter correction unit 25 calculates the amount of correction of the machining parameter. The processing control section 28 corrects the processing parameters using the correction amount calculated by the processing parameter correction section 25. The processing monitoring unit 20 repeats calculation of the correction amount by the processing parameter correction unit 25 and correction of the processing parameters by the processing control unit 28 until a determination result indicating that the processing state is good is obtained.

The machining parameter correction unit 25 may calculate the correction amount based on the value of the machining parameter set in the machining control unit 28 and the determination result by the machining state evaluation unit 23. In this case, the machining parameter correction section 25 acquires the value of the machining parameter set in the machining control section 28 from the machining control section 28 .

The machining parameter correction unit 25 may determine whether there is a sign of a machining defect based on the evaluation result by the machining state evaluation unit 23, and calculate a correction amount when a sign of a machining defect is confirmed. The processing monitoring section 20 may repeat the calculation of the correction amount by the processing parameter correction section 25 and the correction of the processing parameters by the processing control section 28 until no sign of processing defects is confirmed.

The processing parameters corrected by the correction amount calculated by the processing parameter correction unit 25 include, for example, laser output, beam quality of laser light, pressure of processing gas, processing speed, focal length of the imaging optical system 12, and imaging optical system. The diameter of the image formed by the system 12, the pulse frequency of the laser oscillator 4, the duty ratio of the pulses of the laser oscillator 4, the magnification of the imaging optical system 12, the nozzle diameter, the distance between the workpiece 10 and the processing nozzle 7, or the laser beam This is a parameter that indicates the type of mode. The nozzle diameter is the diameter of the hole in the processing nozzle 7 through which the laser beam passes. The processing parameter corrected by the correction amount calculated by the processing parameter correction unit 25 may be a parameter indicating the positional relationship between the center position of the hole in the processing nozzle 7 through which the laser beam passes and the central axis of the laser beam.

When the determination results for each of a plurality of items are input to the machining parameter correction unit 25, the machining parameter correction unit 25 determines the machining parameters to be corrected and the correction target based on the combination of the determination results for each item. The correction amount of the target machining parameter may also be determined.

Here, information indicating that the processing state is good is set to "1", and information indicating that the processing state is poor is set to "0". Further, it is assumed that the machining state evaluation unit 23 outputs the determination results regarding the occurrence of dross, the roughness of the cut surface, and the peeling of the oxide film generated on the cut surface. When “1” is input for the generation of dross, “0” is for the roughness of the cut surface, and “0” is for the peeling of the oxide film generated on the cut surface, the machining parameter correction is performed. The unit 25 determines a processing parameter to be corrected and a correction amount based on the input value combination "1, 0, 0". Based on this combination, the processing parameter correction unit 25 determines, for example, each parameter of laser output and processing gas pressure as processing parameters to be corrected. Furthermore, the processing parameter correction unit 25 determines correction amounts for each of the laser output and the pressure of the processing gas based on this combination. For example, the processing parameter correction unit 25 calculates a correction amount for increasing the laser output and a correction amount for reducing the pressure of the processing gas.

In the machining parameter correction unit 25, a relationship between a machining parameter to be corrected and a correction amount is predetermined for each combination of a plurality of combinations of determination results for each item. The machining parameter correction unit 25 can determine the machining parameters to be corrected and the amount of correction based on a predetermined relationship and a combination of determination results for each item.

In the above description, the machining parameter correction unit 25 calculates the correction amount based on the evaluation result by the machining state evaluation unit 23, but the correction amount may be calculated based on the feature amount. The processing parameter correction section 25 may calculate the correction amount by inputting the feature amount calculated by the feature amount calculation section 22 to the monitoring model. Each of the plurality of trained models held by the trained model holding unit 26 is a trained model that outputs at least one of the evaluation result of the machining state and the correction amount by inputting the feature amount. good. When the learned model outputs a correction amount, the correction amount is calculated after the learned model evaluates the machining state.

In the above description, the machining monitoring unit 20 repeats calculation of the correction amount by the machining parameter correction unit 25 and correction of the machining parameters by the machining control unit 28 until a determination result indicating that the machining state is good is obtained. I decided to do so. The laser processing system 1 may stop the laser processing by the laser processing machine 2 if a situation continues in which a determination result indicating that the processing state is good is not obtained.

As described above, the monitoring model determining unit 24 determines a monitoring model based on the result of the user's evaluation of the processing state by the laser beam machine 2 and the evaluation result by the processing state evaluation unit 23. The user may judge the quality of processing by, for example, visually observing the cut surface of the workpiece 10. The result of the user's evaluation of the processing state may be the result of measuring the roughness of the cut surface using a measuring device such as a three-dimensional measuring device or a roughness measuring device. The monitoring model determination unit 24 may receive information indicating the result of the user's evaluation of the machining state. The user inputs the results of evaluating the machining state to the monitoring model determining unit 24, for example, by operating an input device included in the control device 3.

In the above description, the evaluation result by the user is input to the monitoring model determining unit 24 as a result of evaluating the machining state by observing the workpiece 10 machined by the laser beam machine 2. The evaluation results input to the monitoring model determination unit 24 are not limited to the evaluation results by the user. That is, the results of evaluating the state of machining by observing the workpiece 10 are not limited to the results of evaluation by the user. The monitoring model determining unit 24 may receive a result of determining the quality of machining from a determination device that determines the machining state by observing the workpiece 10. The determination device automatically determines the state of processing based on, for example, an image photographed by a camera. In this way, the monitoring model determining unit 24 may receive either the result of the user's evaluation of the machining state or the result of the determination device's evaluation of the machining state. The monitoring model determination unit 24 selects one learned model among the plurality of learned models whose evaluation result outputted by the machining state evaluation unit 23 is closest to the result of evaluating the machining state by observing the workpiece 10. You can choose. Note that illustration of the determination device is omitted.

When the evaluation result by the user or the determination device is input to the monitoring model determination unit 24, when the machining status evaluation unit 23 determines whether the machining status is good or poor, the machining status is good. The result of the determination made by the user or the determination device as to whether the product falls under the category of "defective" or "defective" is input to the monitoring model determination unit 24. When the determination result of the machining state evaluation unit 23 and the determination result of the user or the determination device match, the monitoring model determination unit 24 determines the trained model that outputs the determination result of the machining state evaluation unit 23 as the monitoring model. do.

When the evaluation results by the user or the determination device are input to the monitoring model determination section 24, and when the machining state evaluation section 23 outputs the evaluation results for each of a plurality of items, the machining state is determined by the user for each of the plurality of items. Alternatively, the result determined by the determination device is input to the monitoring model determination unit 24. When the determination result for each item by the machining state evaluation unit 23 and the determination result for each item by the user or the determination device match, the monitoring model determination unit 24 uses the learning model that outputs the determination result of the machining state evaluation unit 23. The monitored model is determined as the monitored model.

When the evaluation result by the user or the determination device is input to the monitoring model determination unit 24, and the machining state evaluation unit 23 outputs an evaluation value regarding the machining state, the evaluation result obtained by the user or the determination device regarding the machining state The value is input to the monitoring model determination unit 24. When the evaluation value output by the machining state evaluation unit 23 is included in the tolerance range that includes the evaluation value determined by the user or the determination device, the monitoring model determination unit 24 outputs the evaluation value of the machining state evaluation unit 23. The learned model is determined as the monitoring model. The permissible range is a range based on the evaluation value obtained by the user or the determination device, and is a range of a preset numerical range.

In the above description, the monitoring model determination unit 24 selects a trained model whose evaluation result outputted by the machining state evaluation unit 23 matches the result of evaluating the machining state by observing the workpiece 10 among the plurality of trained models. We decided to use a monitoring model as the model. When each of the plurality of trained models is a trained model that outputs a correction amount, the monitoring model determining unit 24 determines which learning model to use as the monitoring model based on the correction amount output from each of the plurality of trained models. You may also decide on an existing model. For example, the monitoring model determining unit 24 compares the machining parameters corrected based on the correction amounts output from the learned model and the machining parameters based on the machining conditions adjusted by the user. The monitoring model determining unit 24 may determine a trained model to be used as a monitoring model through such comparison.

The learned model holding unit 26 holds a plurality of learned models generated by the manufacturer of the laser processing machine 2, for example. The plurality of trained models held in the trained model holding unit 26 may include trained models generated by a user as well as trained models generated by a manufacturer.

Here, an example of a learned model in the first embodiment will be explained. As a learning algorithm used by the laser processing system 1, a known algorithm such as supervised learning, unsupervised learning, or reinforcement learning can be used. As an example, a case where a neural network is applied will be explained.

The learned model in the first embodiment is the result of learning the relationship between the feature amount and the processing state by supervised learning. Here, supervised learning is a method in which a set of input and result data is given to a learning device to learn features in the learning data and infer results from the input. The learning data includes an input and a label that is a result corresponding to the input. The feature amount corresponds to the input. The processing state is training data and corresponds to a label. A neural network is composed of an input layer consisting of a plurality of neurons, a hidden layer which is an intermediate layer consisting of a plurality of neurons, and an output layer consisting of a plurality of neurons. The intermediate layer may be one layer or two or more layers.

FIG. 3 is a diagram illustrating a configuration example of a learned model in the first embodiment. FIG. 3 shows an example of the configuration of a neural network. The neural network shown in FIG. 3 is a three-layer neural network. The input layer includes neurons X1, X2, and X3. The middle layer includes neurons Y1 and Y2. The output layer includes neurons Z1, Z2, Z3. Note that the number of neurons in each layer is arbitrary. The plurality of values input to the input layer are multiplied by weights W1, w11, w12, w13, w14, w15, and w16, and then input to the intermediate layer. The plurality of values input to the intermediate layer are multiplied by weights W2, w21, w22, w23, w24, w25, and w26, and output from the output layer. The output result output from the output layer changes according to the values of weights W1 and W2. The neural network is generated by inputting feature amounts into an input layer and adjusting weights W1 and W2 so that the result output from the output layer approaches the processed state.

Next, the procedure of processing executed by the laser processing system 1 will be explained. Here, three examples of processing procedures when determining a monitoring model will be described.

FIG. 4 is a flowchart showing a first example of a processing procedure executed by the laser processing system 1 according to the first embodiment. In the first example, the laser processing system 1 determines a monitoring model by causing the laser processing machine 2 to perform trial processing. In the first example, the monitoring model determining unit 24 determines the monitoring model based on the result of the user's evaluation of the machining state and the evaluation result by the machining state evaluation unit 23.

In step S1, the laser processing machine 2 starts trial processing. When trial machining is started, each of the sound sensor 8 and the optical sensor 9 detects the machining state and outputs sensor data. In step S2, the data acquisition unit 21 acquires sensor data output from each of the sound sensor 8 and the optical sensor 9. The data acquisition unit 21 outputs the acquired sensor data to the feature amount calculation unit 22.

In step S3, the feature amount calculation unit 22 calculates a feature amount based on the sensor data. The feature amount calculation unit 22 outputs the calculated feature amount to the machining state evaluation unit 23. In step S4, the machining state evaluation unit 23 evaluates the machining state based on the learned model. The machining state evaluation unit 23 obtains an evaluation result of the machining state by inputting a feature amount to one of the plurality of learned models held in the learned model holding unit 26. The machining state evaluation section 23 outputs the obtained evaluation results to the monitoring model determination section 24.

In step S5, the monitoring model determining unit 24 determines whether the evaluation result of the machining state in step S4 matches the evaluation result by the user. In the first example shown in FIG. 4, it is assumed that the evaluation result by the user is input to the monitoring model determination unit 24. The monitoring model determining unit 24 compares the evaluation result of the machining state in step S4 with the evaluation result by the user, thereby determining whether or not an evaluation result that matches the evaluation result by the user has been obtained.

If the evaluation result of the machining state in step S4 matches the evaluation result by the user (step S5, Yes), the monitoring model determining unit 24 determines a monitoring model in step S6. The monitoring model determination unit 24 determines the learned model used for evaluating the machining state in step S4 as the monitoring model.

On the other hand, if the evaluation result of the machining state in step S4 does not match the evaluation result by the user (step S5, No), in step S7, the machining state evaluation unit 23 selects a learned model to be used for evaluating the machining state. do. The machining state evaluation unit 23 selects one of the learned models held in the learned model holding unit 26 other than the learned model used for evaluating the machining state in step S4. After that, the laser processing system 1 returns the procedure to step S4 and performs the processes of steps S4 and S5 for the trained model selected in step S7.

The laser processing system 1 ends the processing according to the procedure shown in FIG. 4 by completing step S6. The laser processing system 1 uses a monitoring model to monitor the processing state when the laser processing machine 2 is processing a product. Note that when the determination device determines the machining state by observing the workpiece 10, the monitoring model determination unit 24 combines the result of the determination device’s evaluation of the machining state with the evaluation result by the machining state evaluation unit 23 in step S5. Determine the monitoring model based on.

The user may determine whether the evaluation result of the machining state by the machining state evaluation unit 23 matches the evaluation result by the user. The user checks the evaluation results output by the machining state evaluation unit 23 for each of the plurality of learned models, and selects one whose evaluation result output by the machining state evaluation unit 23 matches the evaluation result by the user. Select a trained model. In this case, the monitoring model determining unit 24 determines the trained model selected by the user as the monitoring model. In the second example, the monitoring model determination unit 24 determines the monitoring model based on the result of the user's evaluation of the machining state and the evaluation result by the machining state evaluation unit 23.

FIG. 5 is a flowchart illustrating a second example of the processing procedure executed by the laser processing system 1 according to the first embodiment. In the first example described above, the laser processing system 1 obtained evaluation results from each of the plurality of trained models, and compared the obtained evaluation results with the evaluation results by the user. In the second example, the laser processing system 1 obtains evaluation results from all of the plurality of trained models, and compares each obtained evaluation result with the evaluation result by the user. Also in the second example, the laser processing system 1 determines the monitoring model by causing the laser processing machine 2 to perform trial processing.

In step S11, the laser processing machine 2 starts trial processing. In step S12, the data acquisition unit 21 acquires sensor data. In step S13, the feature amount calculation unit 22 calculates a feature amount based on the sensor data. Steps S11 to S13 are similar to steps S1 to S3 shown in FIG. 4.

In step S14, the machining state evaluation unit 23 evaluates the machining state based on each of the plurality of learned models. The machining state evaluation section 23 obtains evaluation results for each of the plurality of learned models by inputting feature amounts to all of the plurality of learned models held in the learned model holding section 26. The machining state evaluation section 23 outputs the obtained evaluation results to the monitoring model determination section 24.

In step S15, the monitoring model determining unit 24 selects a learned model whose processing state evaluation result is closest to the evaluation result by the user from among the plurality of learned models. In the second example shown in FIG. 5, it is assumed that the evaluation result by the user is input to the monitoring model determination unit 24. The monitoring model determining unit 24 compares the evaluation results of each of the plurality of trained models with the evaluation results of the user, and selects one trained model that has obtained the evaluation result closest to the evaluation result of the user.

In step S16, the monitoring model determining unit 24 determines a monitoring model. The monitoring model determining unit 24 determines the trained model selected in step S15 as the monitoring model. The laser processing system 1 ends the processing according to the procedure shown in FIG. 5 by completing step S16. The laser processing system 1 uses a monitoring model to monitor the processing state when the laser processing machine 2 is processing a product.

The learned model whose processing state evaluation result is closest to the evaluation result by the user may be selected by the user. The user checks the evaluation results of each of the plurality of trained models, and selects the one trained model that has obtained the evaluation result closest to the user's evaluation result. In this case, the monitoring model determining unit 24 determines the trained model selected by the user as the monitoring model. Note that when the determination device determines the machining state by observing the workpiece 10, the monitoring model determination unit 24 determines, in step S15, that the evaluation result of the machining state is the evaluation result by the determination device from among the plurality of learned models. Select the closest trained model.

FIG. 6 is a flowchart showing a third example of the processing procedure executed by the laser processing system 1 according to the first embodiment. In the third example, the laser processing system 1 determines the monitoring model while the laser processing machine 2 is processing a product. The laser processing system 1 may determine the monitoring model through trial processing as in the first or second example, or may determine the monitoring model while product processing is being performed as in the third example. You may do so. In the third example, the monitoring model determination unit 24 determines the monitoring model based on the result of the user's evaluation of the machining state and the evaluation result by the machining state evaluation unit 23.

In step S21, the laser processing machine 2 starts processing the product. In step S22, the data acquisition unit 21 acquires sensor data. In step S23, the feature amount calculation unit 22 calculates a feature amount based on the sensor data. In step S24, the machining state evaluation unit 23 evaluates the machining state based on each of the plurality of learned models. In step S25, the monitoring model determining unit 24 selects a learned model whose processing state evaluation result is closest to the user's evaluation result from among the plurality of learned models. In step S26, the monitoring model determining unit 24 determines a monitoring model. Steps S22 to S26 are similar to steps S12 to S16 shown in FIG. 5.

The laser processing system 1 ends the processing according to the procedure shown in FIG. 6 by completing step S26. The laser processing system 1 continues the product processing by the laser processing machine 2 and monitors the processing state using the monitoring model. The laser processing system 1 may determine the monitoring model while the product is being processed, as in the third example, if processing defects are tolerable to some extent in the product processing. Note that when the determination device determines the machining state by observing the workpiece 10, the monitoring model determination unit 24 determines, in step S25, that the evaluation result of the machining state is the evaluation result by the determination device from among the plurality of learned models. Select the closest trained model.

It is assumed that the timing of executing the processing for determining the monitoring model is arbitrary when the product is being processed. For example, the laser processing system 1 executes processing for determining a monitoring model when terminating the operation of the laser processing machine 2 for product processing. The laser processing system 1 may execute processing for determining a monitoring model after finishing processing a certain number of products. The laser processing system 1 may execute processing for determining a monitoring model every time one product is processed. The laser processing system 1 may execute processing for determining a monitoring model when a processing defect occurs. The laser processing system 1 may execute processing for determining a monitoring model when starting the operation of the laser processing machine 2 for product processing.

By determining a monitoring model while product processing is being performed, the laser processing system 1 can determine a trained model to be used as a monitoring model based on a large amount of data. The laser processing system 1 can evaluate the processing state using the monitoring model that reflects aging of the laser processing machine 2 by determining a monitoring model at any time while product processing is being performed.

According to the first embodiment, the control device 3 controls the laser processing machine 2 based on the result of evaluating the processing state by observing the workpiece 10 processed by the laser processing machine 2 and the evaluation result by the processing state evaluation unit 23. Determine a trained model that is a monitoring model used to monitor the state of machining. The control device 3 obtains a monitoring model that enables accurate evaluation of the machining state even when the machining state measured by the sensor changes due to the state of the laser processing machine 2 or variations in the contained components of the workpiece 10. Can be done. By being able to accurately evaluate the quality of the machining state, the control device 3 can allow the laser beam machine 2 to continue good machining. As described above, the control device 3 has the effect of allowing the laser processing machine 2 to continue good processing.

Embodiment 2.
In the first embodiment, the monitoring model determination section 24 monitors a learned model whose evaluation result outputted by the machining state evaluation section 23 matches the evaluation result obtained by observing the workpiece 10, among the plurality of learned models. I decided to go with the model. In Embodiment 2, an example will be described in which a combination of a plurality of trained models is determined as a monitoring model. In Embodiment 2, the same components as in Embodiment 1 are given the same reference numerals, and configurations that are different from Embodiment 1 will be mainly explained.

FIG. 7 is a diagram showing a configuration example of the monitoring model determination unit 30 included in the control device 3 according to the second embodiment. The control device 3 according to the second embodiment includes a monitoring model determining section 30 shown in FIG. 7 instead of the monitoring model determining section 24 shown in FIG. 2. The monitoring model determining unit 30 adjusts the weight for the output value of each trained model for the combination of the plurality of trained models held in the trained model holding unit 26, and adjusts the weight for each trained model. A combination of trained models is determined as a monitoring model.

The monitoring model determining section 30 includes a weight determining section 31. The weight determination unit 31 adjusts the weight for each trained model and determines the weight for each trained model.

The monitoring model determining unit 30 reads out a plurality of trained models from the trained model holding unit 26. In the example shown in FIG. 7, it is assumed that the learned model holding unit 26 holds three learned models, and the monitoring model determination unit 30 reads out the three learned models.

v1, v2, and v3 shown in FIG. 7 represent weights for the output values of each learned model. The weight determining unit 31 adjusts each of v1, v2, and v3. The dashed arrows shown in FIG. 7 indicate that the weight determination unit 31 adjusts each of v1, v2, and v3. Each trained model is, for example, a neural network shown in FIG. 3.

The machining state evaluation unit 23 evaluates the machining state of the laser beam machine 2 based on a combination of a plurality of learned models with adjusted weights for each learned model. The monitoring model determining unit 30 adjusts each of v1, v2, and v3 so that the difference between the evaluation result output by the machining state evaluation unit 23 and the evaluation result obtained by observing the workpiece 10 becomes small. Thereby, the monitoring model determining unit 30 adjusts each of v1, v2, and v3 so that the evaluation result output by the machining state evaluation unit 23 matches the evaluation result obtained by observing the workpiece 10. Note that the evaluation result obtained by observing the workpiece 10 is the result of evaluating the state of processing by observing the workpiece 10 processed by the laser processing machine 2. The evaluation result based on observation of the workpiece 10 is the result of the user evaluating the machining state by observing the machined workpiece 10, or the evaluation result by a determination device that determines the machining state by observing the workpiece 10. It is.

The monitoring model determining unit 30 determines, as the monitoring model, a combination of three learned models whose weights have been adjusted so that the evaluation results output by the machining state evaluation unit 23 match the evaluation results obtained by observing the workpiece 10. do. Alternatively, the monitoring model determination unit 30 monitors a combination of three learned models whose weights have been adjusted so that the evaluation result output by the machining state evaluation unit 23 is closest to the evaluation result obtained by observing the workpiece 10. Decide on the model. In this way, the monitoring model determination unit 30 determines, as the monitoring model, a combination of a plurality of trained models in which the weights of each trained model are adjusted. The monitoring model determining unit 30 stores the determined monitoring model in the monitoring model holding unit 27. The monitoring model holding unit 27 holds a monitoring model.

The processing monitoring section 20 monitors the state of processing by the laser processing machine 2 using the monitoring model determined by the monitoring model determining section 30. The monitoring model outputs the sum of weighted output values of each trained model by inputting the feature amount. The monitoring model may output a weighted average of the output values of each trained model by inputting the feature amount. The monitoring model determining unit 30 may determine the weight of each learned model by learning the relationship between the weight, the feature amount, and the processing state.

When each learned model is a neural network, the monitoring model determination unit 30 does not change the weights set in the neural network, that is, w11-w16 and w21-w26 shown in FIG. 3. The monitoring model determining unit 30 determines a monitoring model by adjusting the weight for the output value of each learned model without changing the weight set in the neural network. The processing monitoring section 20 can use a combination of a plurality of learned models as a monitoring model by adjusting the weight for the output value of each learned model by the monitoring model determining section 30.

The control device 3 obtains a monitoring model for evaluating the state of machining by the laser beam machine 2 by adjusting the weight for each learned model. Each of the plurality of learned models held in the learned model holding unit 26 can be a model that is the basis of the monitoring model. The model that is the basis of the monitoring model is, for example, a model set for each manufacturer of the work 10 or a model set for each material of the work 10. The control device 3 can reduce the number of trained models set in advance, compared to the case where a trained model to be a monitoring model is selected from among a plurality of trained models. The control device 3 can determine a monitoring model based on relatively little data.

According to the second embodiment, the control device 3 determines, as the monitoring model, a combination of a plurality of trained models in which the weights of each trained model are adjusted. The control device 3 can obtain a monitoring model that enables accurate evaluation of the machining state, and can allow the laser beam machine 2 to continue good machining. As described above, the control device 3 has the effect of allowing the laser processing machine 2 to continue good processing. The laser processing system 1 can improve productivity by being able to determine a monitoring model based on relatively little data.

Embodiment 3.
In the first and second embodiments, an example has been described in which a plurality of trained models are set in advance. In Embodiment 3, an example will be described in which a learned model is added through learning in the laser processing system 1. In Embodiment 3, the same components as in Embodiment 1 or 2 are given the same reference numerals, and configurations that are different from Embodiment 1 or 2 will be mainly explained.

FIG. 8 is a diagram showing a configuration example of the control device 40 included in the laser processing system 1 according to the third embodiment. A laser processing system 1 according to the third embodiment includes a laser processing machine 2 and a control device 40. The control device 40 includes a processing control section 28 and a processing monitoring section 41. The control device 40 may be a device external to the laser processing machine 2 or may be a device included in the laser processing machine 2. Alternatively, the processing control section 28 of the control device 40 may be realized by a device included in the laser beam machine 2, and the processing monitoring section 41 of the control device 40 may be realized by a device external to the laser beam machine 2.

The machining monitoring unit 41 includes a data acquisition unit 21, a feature value calculation unit 22, a machining state evaluation unit 23, a machining parameter correction unit 25, a learned model holding unit 26, a monitoring model holding unit 27, and a monitoring model It includes a determining section 42 and a learning section 43.

The learning section 43 acquires the feature amount calculated by the feature amount calculation section 22 from the feature amount calculation section 22 . The learning unit 43 acquires information indicating the state of processing by the laser processing machine 2. The learning unit 43 generates a learned model by learning the relationship between the feature amount and the state of machining by the laser processing machine 2, and adds the generated learned model to the learned model holding unit 26.

The monitoring model determining unit 42 adjusts the weight for the output value of each trained model for the combination of the plurality of trained models held in the trained model holding unit 26, and adjusts the weight for each trained model. A combination of trained models is determined as a monitoring model. The plurality of learned models held in the learned model holding unit 26 include the learned models added by the learning unit 43.

Each of the plurality of trained models is, for example, a neural network shown in FIG. 3. The learning unit 43 learns the relationship between the feature amount and the processing state by supervised learning, and generates a learned model.

FIG. 9 is a diagram illustrating a configuration example of the monitoring model determination unit 42 included in the control device 40 according to the third embodiment. The monitoring model determining section 42 includes a weight determining section 44 . The weight determination unit 44 adjusts the weight for each trained model and determines the weight for each trained model.

The monitoring model determining unit 42 reads out a plurality of trained models from the trained model holding unit 26. In the example shown in FIG. 9, it is assumed that the learned model holding unit 26 holds five learned models, and the monitoring model determination unit 42 reads out the five learned models. Two of the five learned models are learned models added by the learning unit 43.

v1, v2, v3, v4, and v5 shown in FIG. 9 represent weights for the output values of each trained model. The weight determining unit 44 adjusts each of v1, v2, v3, v4, and v5. The broken line arrows shown in FIG. 9 indicate that the weight determination unit 44 adjusts each of v1, v2, v3, v4, and v5.

The machining state evaluation unit 23 evaluates the machining state of the laser beam machine 2 based on a combination of a plurality of learned models with adjusted weights for each learned model. The monitoring model determining unit 42 adjusts each of v1, v2, v3, v4, and v5 so that the difference between the evaluation result output by the machining state evaluation unit 23 and the evaluation result obtained by observing the workpiece 10 becomes small. do. Thereby, the monitoring model determining unit 42 adjusts each of v1, v2, v3, v4, and v5 so that the evaluation result output by the machining state evaluation unit 23 matches the evaluation result obtained by observing the workpiece 10. .

The monitoring model determination unit 42 determines, as the monitoring model, a combination of five learned models whose weights have been adjusted so that the evaluation results output by the machining state evaluation unit 23 match the evaluation results obtained by observing the workpiece 10. do. Alternatively, the monitoring model determining unit 42 monitors a combination of five learned models whose weights have been adjusted so that the evaluation result output by the machining state evaluation unit 23 is closest to the evaluation result obtained by observing the workpiece 10. Decide on the model. In this way, the monitoring model determination unit 42 determines, as the monitoring model, a combination of a plurality of trained models in which the weights of each trained model are adjusted. The monitoring model determining unit 42 stores the determined monitoring model in the monitoring model holding unit 27. The monitoring model holding unit 27 holds a monitoring model.

The processing monitoring section 20 monitors the state of processing by the laser processing machine 2 using the monitoring model determined by the monitoring model determining section 42. The monitoring model outputs the sum of weighted output values of each trained model by inputting the feature amount. The monitoring model may output a weighted average of the output values of each trained model by inputting the feature amount. The monitoring model determination unit 42 may determine the weight of each learned model by learning the relationship between the weight, the feature amount, and the processing state.

The learning unit 43 generates a learned model using feature amounts calculated based on sensor data actually obtained from the laser processing machine 2. By creating a monitoring model based on the learned model, the control device 40 can obtain a highly accurate monitoring model based on the actual state of the laser processing machine 2.

The learning unit 43 generates a new trained model using the feature amounts and information indicating the state of processing. Alternatively, the learning unit 43 may perform additional learning of a trained model generated in the past. When the learned model is added to the learned model holding unit 26 by the learning unit 43, the added learned model may be accumulated in the learned model holding unit 26. The trained model added to the trained model holding unit 26 may be deleted from the trained model holding unit 26 as appropriate.

In the control device 40, when the material of the workpiece 10 processed by the laser processing machine 2 is changed, the learned model generated by the learning section 43 is added to the learned model holding section 26, or the learned model is added to the learned model holding section 26, or The trained model generated by the unit 43 and added to the trained model holding unit 26 may be deleted from the trained model holding unit 26. The control device 40 obtains a highly accurate monitoring model according to the material of the work 10 by updating the learned model held in the learned model holding unit 26 when the material of the work 10 is changed. be able to.

In the control device 40, when the state of processing by the laser processing machine 2 changes, the learned model generated by the learning section 43 is added to the learned model holding section 26, or the learned model generated by the learning section 43 is added to the learned model holding section 26, or the learned model generated by the learning section 43 is The trained model added to the trained model holding unit 26 may be deleted from the trained model holding unit 26. A change in the machining state is, for example, a change in the machining state from a good state to a poor state. The control device 40 can obtain a highly accurate monitoring model according to the machining state by updating the learned model held in the learned model holding unit 26 when the machining state changes. .

The learned model that is the learning result by the learning unit 43 may be any learned model that outputs at least one of the processing state evaluation result and the correction amount by inputting the feature amount. The processing parameter correction section 25 may calculate the correction amount by inputting the feature amount calculated by the feature amount calculation section 22 to the monitoring model.

The machining state evaluation unit 23 may evaluate the machining state by the laser beam machine 2 by inputting information other than the feature amount calculated by the feature amount calculation unit 22 to each of the plurality of learned models. Examples of information input to each of the plurality of trained models include values of processing parameters at the time of evaluation, temperature of the processing optical system at the time of evaluation, temperature change of the processing optical system, thickness of the workpiece 10, or This is information such as materials. In this case, the learning unit 43 generates a learned model by learning the relationship between the information, the feature amount, and the processing state. Further, the machining state evaluation unit 23 evaluates the machining state by the laser processing machine 2 by inputting such information and feature amounts to the monitoring model.

In the above description, the learning section 43 is provided inside the control device 40. The learning unit 43 may be realized by a device external to the control device 40. The learning device that implements the learning section 43 may be a device that can be connected to the control device 40 via a network. The learning device may be a device existing on a cloud server.

According to the third embodiment, the control device 40 determines, as the monitoring model, a combination of a plurality of trained models in which the weights of each trained model are adjusted. The control device 40 can obtain a monitoring model that enables accurate evaluation of the machining state, and can allow the laser beam machine 2 to continue good machining. As described above, the control device 40 has the effect of allowing the laser processing machine 2 to continue good processing. The laser processing system 1 can improve productivity by being able to determine a monitoring model based on relatively little data. The control device 40 can obtain a highly accurate monitoring model by generating a learned model using feature amounts calculated based on sensor data actually obtained from the laser processing machine 2.

Next, hardware that realizes the control devices 3 and 40 according to the first to third embodiments will be explained. The control devices 3 and 40 are realized by processing circuits. The processing circuit may be a circuit on which a processor executes software, or may be a dedicated circuit.

When the processing circuit is implemented by software, the processing circuit is, for example, the control circuit 50 shown in FIG. FIG. 10 is a diagram showing a configuration example of the control circuit 50 according to the first to third embodiments. The control circuit 50 includes an input section 51, a processor 52, a memory 53, and an output section 54. The input unit 51 is an interface circuit that receives data input from outside the control circuit 50 and provides it to the processor 52. The output unit 54 is an interface circuit that sends data from the processor 52 or memory 53 to the outside of the control circuit 50.

When the processing circuit is the control circuit 50 shown in FIG. 10, the control devices 3 and 40 are realized by software, firmware, or a combination of software and firmware. Software or firmware is written as a program and stored in memory 53. The processing circuit realizes each function of the control devices 3 and 40 by having the processor 52 read and execute a program stored in the memory 53. That is, the processing circuit includes a memory 53 for storing a program by which the processing of the control devices 3 and 40 will be executed as a result. It can also be said that these programs cause the computer to execute the procedures and methods of the control devices 3 and 40.

The processor 52 is a CPU (Central Processing Unit). The processor 52 may be a central processing unit, a processing unit, an arithmetic unit, a microprocessor, a microcomputer, a processor, or a DSP (Digital Signal Processor). The memory 53 is a nonvolatile memory such as RAM (Random Access Memory), ROM (Read Only Memory), flash memory, EPROM (Erasable Programmable Read Only Memory), and EEPROM (registered trademark) (Electrically Erasable Programmable Read Only Memory). Alternatively, volatile semiconductor memory, magnetic disk, flexible disk, optical disk, compact disk, mini disk, DVD (Digital Versatile Disc), etc. are applicable.

Although FIG. 10 shows an example of hardware in which the control devices 3 and 40 are realized by a general-purpose processor 52 and a memory 53, the control devices 3 and 40 may also be realized by dedicated hardware circuits. FIG. 11 is a diagram showing a configuration example of the dedicated hardware circuit 55 according to the first to third embodiments.

The dedicated hardware circuit 55 includes an input section 51, an output section 54, and a processing circuit 56. The processing circuit 56 is a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a combination thereof. Each function of the control devices 3 and 40 may be realized by the processing circuit 56 for each function, or each function may be realized by the processing circuit 56 collectively. Note that the control devices 3 and 40 may be realized by combining the control circuit 50 and the hardware circuit 55.

When the learning unit 43 is a device external to the control device 40, the learning device that is such a device is realized by a processing circuit similarly to the control devices 3 and 40. The processing circuit that realizes the learning device is the control circuit 50 shown in FIG. 10 or the dedicated hardware circuit 55 shown in FIG. 11.

The configurations shown in each of the embodiments above are examples of the contents of the present disclosure. The configuration of each embodiment can be combined with other known techniques. The configurations of each embodiment may be combined as appropriate. It is possible to omit or change a part of the configuration of each embodiment without departing from the gist of the present disclosure.

1 Laser processing system, 2 Laser processing machine, 3, 40 Control device, 4 Laser oscillator, 5 Cable, 6 Processing head, 7 Processing nozzle, 8 Sound sensor, 9 Optical sensor, 10 Work, 11 Collimating optical system, 12 Imaging Optical system, 13 Protective glass, 14 Processing head drive unit, 15, 16 Movement mechanism, 20, 41 Processing monitoring unit, 21 Data acquisition unit, 22 Feature value calculation unit, 23 Processing state evaluation unit, 24, 30, 42 Monitoring model Determining unit, 25 Processing parameter correction unit, 26 Learned model holding unit, 27 Monitoring model holding unit, 28 Processing control unit, 31, 44 Weight determining unit, 43 Learning unit, 50 Control circuit, 51 Input unit, 52 Processor, 53 memory, 54 output section, 55 hardware circuit, 56 processing circuit.

Claims (10)

a learned model holding unit that holds a plurality of learned models that are results of learning a relationship between a feature value indicating a state of processing by a processing machine and a state of processing by the processing machine;
a machining state in which the state of machining by the processing machine is evaluated by inputting the feature amounts to each of the plurality of trained models or by inputting the feature amounts to a combination of the plurality of trained models; Evaluation department and
The learning, which is a monitoring model used for monitoring the machining state by the processing machine, based on the result of evaluating the machining state by observing the workpiece machined by the processing machine and the evaluation result by the machining state evaluation unit. a monitoring model determining unit that determines a trained model or a combination of the trained models;
A control device comprising: a processing control unit that controls the processing machine based on a result of monitoring the state of processing by the processing machine using the monitoring model. The control device according to claim 1, wherein the monitoring model determining unit determines one of the plurality of learned models held in the learned model holding unit as the monitoring model. The monitoring model determining unit adjusts the weight for the output value of each trained model for the combination of the plurality of trained models held in the trained model holding unit, and determines the weight for each trained model. The control device according to claim 1, wherein a combination of the plurality of learned models, each of which has been adjusted, is determined as the monitoring model. The processing machine is a laser processing machine that processes the workpiece with a laser beam,
a data acquisition unit that acquires data indicating a detection result of light generated during processing by the laser processing machine and data indicating a detection result of sound generated during processing by the laser processing machine;
The control device according to any one of claims 1 to 3, further comprising: a feature amount calculation unit that calculates the feature amount from data acquired by the data acquisition unit. The processing machine is a laser processing machine that processes the workpiece with a laser beam,
Each of the plurality of learned models held in the learned model holding section is a plurality of learned models that are different from each other in at least one of the thickness of the plate-shaped workpiece and the type of processing gas used during processing by the laser processing machine. The control device according to any one of claims 1 to 3 , wherein the control device is a result of learning the relationship between the feature amount and the machining state in each case. A learning unit that generates the learned model by learning the relationship between the feature amount and the state of machining by the processing machine, and adds the generated learned model to the learned model holding unit. A control device according to any one of claims 1 to 3 . When the material of the workpiece processed by the processing machine is changed, the learned model generated by the learning section is added to the learned model holding section, or the learned model generated by the learning section is added to the learned model holding section. The control device according to claim 6, wherein the learned model added to the learned model holding unit is deleted from the learned model holding unit. When the state of machining by the processing machine changes, the learned model generated by the learning section is added to the learned model holding section, or the learned model generated by the learning section and held is 7. The control device according to claim 6, wherein the learned model added to the learned model storage section is deleted from the learned model holding section. A laser processing machine that processes a workpiece with laser light,
A control device that controls the laser processing machine,
The control device includes:
a learned model holding unit that holds a plurality of learned models that are results of learning a relationship between a feature amount indicating a state of processing by the laser processing machine and a state of processing by the laser processing machine;
Processing that evaluates the state of machining by the laser processing machine by inputting the feature amount to each of the plurality of trained models or by inputting the feature amount to a combination of the plurality of trained models. a condition evaluation section;
A monitoring model used for monitoring the processing state by the laser processing machine based on the result of evaluating the processing state by observing the workpiece processed by the laser processing machine and the evaluation result by the processing state evaluation unit. a monitoring model determining unit that determines the trained model or a combination of the trained models;
A laser processing system comprising: a processing control section that controls the laser processing machine based on a result of monitoring the processing state of the laser processing machine using the monitoring model. a step of acquiring a feature amount indicating a processing state by a laser processing machine that processes a workpiece with a laser beam;
By inputting the acquired feature amount to each of a plurality of learned models that are the results of learning the relationship between the machining state by the laser processing machine and the feature amount, or by inputting the obtained feature amount to each of the plurality of learned models, evaluating the state of machining by the laser processing machine by inputting the acquired feature amounts into a combination;
Used to monitor the processing state by the laser processing machine based on the result of evaluating the processing state by observing the workpiece processed by the laser processing machine and the evaluation result from the step of evaluating the processing state. determining the trained model or a combination of the trained models, which is a monitoring model;
A laser processing method comprising: controlling the laser processing machine based on a result of monitoring the processing state of the laser processing machine using the monitoring model.

Images