Estimation model generation device and machining state estimation device

Abstract

The estimation model generation apparatus is an apparatus for generating an estimation model for estimating the processing state of laser processing, comprising: an information acquisition unit that, during laser processing based on a first apparatus, observes first thermal radiation, first visible light, first reflected light, and first laser light from a workpiece and acquires first waveform data including a first waveform and a second waveform for at least two of the first thermal radiation, first visible light, first reflected light, and first laser light; an estimation model generation unit that uses teaching data established by using the first waveform data as an explanatory variable and the processing state as a target variable to perform machine learning to generate a first estimation model estimating the processing state of the first apparatus; and a storage unit that stores the first estimation model.

Description

Technical Field

[0001] This disclosure relates to a method for generating a laser-based model for estimating processing state and a processing state estimation device. Background Technology

[0002] When judging the processing status in laser processing, it is known to detect visible light, reflected light, or thermal radiation emitted from the workpiece during laser processing.

[0003] For example, the welding state determination device described in Patent Document 1 detects the intensity of plasma light and reflected light emitted from the workpiece during laser welding, and uses the extreme values ​​of feature values ​​extracted based on the intensity of the detected light in a given interval to determine the state of the workpiece.

[0004] [Prior Technology Documents]

[0005] [Patent Literature]

[0006] [Patent Document 1] Japanese Patent Application Publication No. 2000-153379 Summary of the Invention

[0007] The estimating model generation apparatus disclosed herein is an apparatus for generating an estimating model for estimating the processing state of laser processing.

[0008] The presumption model generation device includes:

[0009] The information acquisition unit observes first thermal radiation, first visible light, first reflected light and first laser from the workpiece during laser processing based on the first device, and acquires first waveform data including a first waveform and a second waveform for at least two of the first thermal radiation, first visible light, first reflected light and first laser.

[0010] The estimation model generation unit uses teaching data that establishes a corresponding model by using the first waveform data as an explanatory variable and the processing state as a target variable, performs machine learning, and generates a first estimation model that estimates the processing state of the first device; and

[0011] The storage unit stores the first presumed model.

[0012] The estimation model generation apparatus involved in other aspects of this disclosure is an apparatus for generating estimation models for estimating the processing state of laser processing in the first and second apparatuses.

[0013] The presumption model generation device includes:

[0014] The information acquisition unit observes third thermal radiation, third visible light, third reflected light, and third laser light from the workpiece during laser processing based on the first device, and observes fourth thermal radiation, fourth visible light, fourth reflected light, and fourth laser light from the workpiece during laser processing based on the second device, and acquires third waveform data including a fifth waveform and a sixth waveform for at least two of the third thermal radiation, third visible light, third reflected light, and third laser light, and fourth waveform data including a seventh waveform and an eighth waveform for at least two of the fourth thermal radiation, fourth visible light, fourth reflected light, and fourth laser light;

[0015] The information transformation unit multiplies the third coefficient, calculated based on the relationship between the fifth and seventh waveforms, with the fifth waveform, and multiplies the fourth coefficient, calculated based on the relationship between the sixth and eighth waveforms, with the sixth waveform to generate second transformed data.

[0016] The estimation model generation unit generates a second estimation model estimating the processing state of the first device and a third estimation model estimating the processing state of the second device; and

[0017] Storage section, storage second presumption model and third presumption model,

[0018] Presumed model generation unit,

[0019] By using the third waveform data as the explanatory variable and the processing state as the target variable to establish corresponding teaching data, machine learning is performed to generate a second inference model.

[0020] Using the second transformation data as the explanatory variable and the processing state as the target variable, corresponding teaching data is established, and machine learning is performed to generate a third inference model. Attached Figure Description

[0021] Figure 1A This is a block diagram illustrating the presumption model generation apparatus according to Embodiment 1.

[0022] Figure 1B This is a block diagram illustrating the processing state estimation device according to Embodiment 1.

[0023] Figure 1C This is a block diagram representing the first device.

[0024] Figure 1D This is a block diagram representing the second device.

[0025] Figure 2A It is a graph representing the first waveform data obtained by the information acquisition unit.

[0026] Figure 2B It is a graph representing the second waveform data obtained by the information acquisition unit.

[0027] Figure 3A It is a chart after the first waveform data has been standardized.

[0028] Figure 3B It is a chart after the data from the first transformation has been standardized.

[0029] Figure 4A It is a graph showing the relationship between the average signal strength and the amount of defocus in the first waveform data and the first transformation data.

[0030] Figure 4B It is a graph showing the relationship between the correct value and the estimated value when the processing state is estimated based on the first estimation model in the processing state estimation device, with the first transformation data and the second waveform data as input data.

[0031] Figure 5A This is a block diagram illustrating the presumption model generation apparatus involved in Embodiment 2.

[0032] Figure 5B This is a block diagram illustrating the processing state estimation device involved in Embodiment 2. Detailed Implementation

[0033] (The process by which this disclosure was derived)

[0034] When monitoring the processing status during laser processing, sensors are used to detect the thermal radiation, reflected light, visible light, and laser emitted from the laser processing unit, as described in Patent Document 1, to determine the welding status.

[0035] For example, the following method is also known: using teaching data with the aforementioned sensing data as explanatory variables and processing status as target variables, machine learning is performed to generate an inference model, and the processing status is determined based on the sensing data during processing using the inference model.

[0036] Because laser processing is a non-contact process, the reproducibility of the processing phenomena is high. In other words, under the same processing conditions, the intensity of thermal radiation, reflected light, visible light, and laser light will be at the same level. However, due to deviations in light transmission efficiency or amplifier gain, machine differences will occur in the sensing data obtained by the sensor, depending on the processing device. Therefore, if the estimation model generated using sensing data from a given processing device is applied to other processing devices, it becomes difficult to estimate the correct processing condition. In other words, in the welding condition determination device described in Patent Document 1, there is a problem that, due to machine differences in the sensors that detect light intensity, even under the same processing conditions, different intensities of light will be detected.

[0037] Furthermore, it is difficult to use methods that involve changing parameters during learning, i.e., obtaining sensing data under conditions other than standard processing conditions, in the equipment used on production lines. Therefore, sometimes experimental equipment is used to obtain sensing data for learning and to create an inference model. For the reasons mentioned above, such inference models are difficult to use for estimating the processing state of equipment used on production lines.

[0038] Therefore, the inventors of this application have studied a method for using a presumed model generated using sensing data from a given processing device, which can be used even in other processing devices, and have arrived at the following invention. This disclosure provides a presumed model generation apparatus and a processing state estimation apparatus that can use a presumed model generated with a given processing device even in other processing devices.

[0039] The estimating model generation apparatus disclosed herein is an apparatus for generating an estimating model for estimating the processing state of laser processing.

[0040] The presumption model generation device includes:

[0041] The information acquisition unit observes first thermal radiation, first visible light, first reflected light and first laser from the workpiece during laser processing based on the first device, and acquires first waveform data including a first waveform and a second waveform for at least two of the first thermal radiation, first visible light, first reflected light and first laser.

[0042] The estimation model generation unit performs machine learning on corresponding teaching data established using the first waveform data as an explanatory variable and the processing state as a target variable to generate a first estimation model for estimating the processing state based on the first device; and

[0043] The storage unit stores the first presumed model.

[0044] The processing state estimation device disclosed herein is a device for estimating the processing state of laser processing in a second device that is different from the first device.

[0045] The processing state estimation device includes:

[0046] The information acquisition unit, during laser processing based on the second device, observes second thermal radiation, second visible light, second reflected light and second laser from the workpiece, and acquires second waveform data including a third waveform and a fourth waveform for at least two of the second thermal radiation, second visible light, second reflected light and second laser.

[0047] The information transformation unit multiplies the third waveform by a first coefficient calculated based on the relationship between the third waveform and the first waveform, and multiplies the fourth waveform by a second coefficient calculated based on the relationship between the fourth waveform and the second waveform, thereby generating first transformed data.

[0048] The storage unit stores the first presumption model generated by the aforementioned presumption model generation apparatus; and

[0049] The estimation unit estimates the processing state based on the first estimation model and the first transformation data.

[0050] According to this structure, an estimation model generation apparatus and a processing state estimation apparatus can be provided that allow the estimation model generated by a given processing apparatus to be used in other processing apparatuses. The estimation model generation apparatus generates a first estimation model using first waveform data from a first apparatus. By multiplying given coefficients by waveform data from a second apparatus and transforming the waveform data from the second apparatus in conjunction with the waveform data from the first apparatus, the same first estimation model can be used in different apparatuses.

[0051] Alternatively, the first coefficient and the second coefficient can be determined so that the first transformed data is generated while keeping the ratio of the average value of the first waveform to the average value of the second waveform constant.

[0052] Based on the structure described above, a model generation device and a processing state estimation device with higher estimation accuracy can be provided.

[0053] The estimation model generation apparatus involved in other aspects of this disclosure is an apparatus for generating estimation models for estimating the processing state of laser processing in the first and second apparatuses.

[0054] The presumption model generation device includes:

[0055] The information acquisition unit observes third thermal radiation, third visible light, third reflected light, and third laser light from the workpiece during laser processing based on the first device, and observes fourth thermal radiation, fourth visible light, fourth reflected light, and fourth laser light from the workpiece during laser processing based on the second device, and acquires third waveform data including a fifth waveform and a sixth waveform for at least two of the third thermal radiation, the third visible light, the third reflected light, and the third laser light, and fourth waveform data including a seventh waveform and an eighth waveform for at least two of the fourth thermal radiation, the fourth visible light, the fourth reflected light, and the fourth laser light;

[0056] The information transformation unit multiplies the fifth waveform by a third coefficient calculated based on the relationship between the fifth waveform and the seventh waveform, and multiplies the sixth waveform by a fourth coefficient calculated based on the relationship between the sixth waveform and the eighth waveform, thereby generating second transformed data.

[0057] The estimation model generation unit generates a second estimation model that estimates the processing state based on the first device and a third estimation model that estimates the processing state based on the second device; and

[0058] The storage unit stores the second presumption model and the third presumption model.

[0059] The presumption model generation unit,

[0060] The second inference model is generated by using machine learning to establish corresponding teaching data with the third waveform data as an explanatory variable and the processing state as the target variable.

[0061] The third inference model is generated by using machine learning to establish corresponding teaching data with the second transformation data as explanatory variables and the processing state as target variables.

[0062] According to this structure, by transforming the waveform data in the first device in conjunction with the second device, the waveform data of the first device can be used to generate a third estimation model for the second device.

[0063] Alternatively, the third and fourth coefficients can be determined such that the second transformation data is generated while keeping the ratio of the average values ​​of the seventh waveform and the eighth waveform constant.

[0064] Based on this structure, it is possible to provide a estimation model generation device and a processing state estimation device with higher estimation accuracy.

[0065] The processing state estimation device involved in other aspects of this disclosure is a device for estimating the processing state of laser processing, and includes:

[0066] The storage unit stores the third presumption model generated by the presumption model generation device described in technical solution 4 or 5;

[0067] The information acquisition unit acquires the fourth waveform data; and

[0068] The estimation unit estimates the processing state based on the third estimation model and the fourth waveform data.

[0069] Based on this structure, a presumption model can be generated for each processing unit.

[0070] The embodiments disclosed herein will now be described in detail with appropriate reference to the accompanying drawings. Sometimes, unnecessary details are omitted. For example, detailed descriptions of already known matters or repetitive descriptions of substantially the same structures are sometimes omitted. This is to avoid unnecessarily redundancy in the following description and to facilitate understanding by those skilled in the art. It should be noted that the inventors have provided the drawings and the following description to enable those skilled in the art to fully understand this disclosure, and not to limit the subject matter of the claims.

[0071] (Implementation Method 1)

[0072] [Overall Structure]

[0073] Figure 1A This is a block diagram illustrating the presumption model generation apparatus 100 according to Embodiment 1. Figure 1B This is a block diagram illustrating the processing state estimation device 200 according to Embodiment 1. Figure 1C A block diagram representing the first device 300. Figure 1D This is a block diagram showing the second device 400. They can also be installed separately in the same factory or in two or more locations. The estimation model generation device 100 and the processing state estimation device 200 can also be integrated.

[0074] Reference Figures 1A to 1D The estimation model generation apparatus 100 and the processing state estimation apparatus 200 involved in this embodiment will be described. The estimation model generation apparatus 100, the processing state estimation apparatus 200, and the apparatuses 300 and 400 are interconnected via wired or wireless means. Communication can be performed using public lines such as the Internet and / or dedicated lines.

[0075] Figure 1A The estimated model generation device 100 shown is used to generate models for use in practice. Figure 1C The apparatus shown is a first estimation model that uses sensing data from the first device 300 to estimate the processing state. The estimation model generation apparatus 100 can be constructed, for example, using a computer system such as a PC or a workbench. The estimation model generation apparatus 100 includes an information acquisition unit 11, an estimation model generation unit 12, and a storage unit 13. The internal structure of the estimation model generation apparatus 100 will be described later.

[0076] Figure 1B The processing state estimation device 200 shown is based on Figure 1AThe processing state estimation device 200 is a device that uses a first estimation model generated in the estimation model generation device 100 to estimate the processing state of a first device 300 or a second device 400 different from the first device 300. The processing state estimation device 200 can be configured, for example, by a microcomputer, CPU, MPU, GPU, DSP, FPGA, or ASIC. The functionality of the processing state estimation device 200 can be implemented solely by hardware or by a combination of hardware and software. The processing state estimation device 200 includes an information acquisition unit 21, an information transformation unit 22, an estimation unit 23, and a storage unit 24. The internal structure of the processing state estimation device 200 will be described later.

[0077] Figure 1C and Figure 1D The first device 300 and the second device 400 shown are laser processing apparatuses that perform welding or cutting using a laser. By irradiating a metal plate, which is the object of processing, with a laser, welding or cutting of the metal plate can be performed. The first device 300 includes a laser welding unit 31 and a first sensor 32. The second device 400 includes a laser welding unit 41 and a second sensor 42.

[0078] The first device 300 is, for example, an experimental device or a laser processing device used to generate a first presupposition model. The second device 400 is, for example, a device used in a production line.

[0079] In both the first device 300 and the second device 400, thermal radiation, visible light, and reflected light are generated when a laser is irradiated onto the workpiece. The thermal radiation, visible light, and reflected light generated during laser processing in the first device 300 are detected as first thermal radiation, first visible light, and first reflected light by a first sensor 32. Furthermore, in the first device 300, the laser irradiated during processing is detected as a first laser by the first sensor 32. Similarly, during laser processing in the second device 400, second thermal radiation, second visible light, second reflected light, and second laser are detected by a second sensor 42. The first sensor 32 and the second sensor 42 are, for example, photodetectors. It should be noted that the first sensor 32 and the second sensor 42 may also include different sensors for thermal radiation, visible light, reflected light, and laser light.

[0080] Thermal radiation is generated when a laser irradiates a workpiece, causing the temperature of the irradiated portion to rise. Visible light is plasma light generated when a laser irradiates a workpiece, causing the melted workpiece to absorb the laser as described above. Reflected light is light that reflects the laser irradiating the workpiece.

[0081] <Presumed Model Generation Device>

[0082] The estimation model generation apparatus 100 is an apparatus for generating a first estimation model for estimating the processing state of laser processing in the first apparatus 300. As described above, the estimation model generation apparatus 100 includes an information acquisition unit 11, an estimation model generation unit 12, and a storage unit 13.

[0083] The information acquisition unit 11 acquires first waveform data, including a first waveform and a second waveform of at least two of the first thermal radiation, first visible light, first reflected light, and first laser light detected from the workpiece during laser processing in the first device 300. The first waveform data is acquired based on the detection values ​​detected by the sensor 32 of the first device 300.

[0084] Figure 2A This is a graph representing the first waveform data acquired by the information acquisition unit 11. For example... Figure 2A As shown, the first waveform data includes the waveforms of the first thermal radiation, the first visible light, the first reflected light, and the first laser beam during laser processing. Figure 2A In the example, the first waveform data includes four waveforms: thermal radiation, visible light, reflected light, and laser light. For example, the laser waveform can be used as the first waveform, and the visible light waveform as the second waveform. The first waveform data contains at least two waveforms: the first waveform and the second waveform. It should be noted that the first and second waveforms are not limited to the laser waveform and the visible light waveform; any waveform that can be detected by the sensor 32 is acceptable.

[0085] The first waveform data only needs to include at least two waveforms. Since the laser waveform is the laser output from the laser welding unit 31 detected by sensor 32, the output is relatively stable. Therefore, the first waveform data can include the waveform of the first laser. Furthermore, the included waveforms are not limited to two; they can also be three or more.

[0086] The first waveform data contains more than two waveforms, which enables the generation of a estimation model with higher estimation accuracy.

[0087] The estimation model generation unit 12 uses teaching data established by using the first waveform data as an explanatory variable and the processing state as a target variable to perform machine learning, and generates a first estimation model that estimates the processing state based on the first device 300. The processing state includes, for example, the state during processing such as deviation of the laser focus position, deviation of the laser irradiation position, or the presence or absence of an opening.

[0088] In the first apparatus 300, laser processing can be performed under both standard processing conditions used in the production line and processing conditions other than standard processing conditions, and first waveform data is acquired by the information acquisition unit 11 based on the respective sensing data. In this way, data on various processing conditions are acquired, thereby enabling the generation of a first estimation model with higher estimation accuracy.

[0089] Storage Department 13 Storage First Presumed Model.

[0090] <Processing Status Estimation Device>

[0091] The processing state estimation device 200 is a device for estimating the processing state of laser processing in the second device 400. As described above, the processing state estimation device 200 includes an information acquisition unit 21, an information conversion unit 22, an estimation unit 23, and a storage unit 24.

[0092] The information acquisition unit 21 acquires second waveform data, including a third waveform and a fourth waveform of at least two of the second thermal radiation, second visible light, second reflected light and second laser detected during laser processing in the second device.

[0093] The second waveform data contains the same type of waveform as the first waveform data. Figure 2B This is a graph representing the second waveform data acquired by the information acquisition unit 21. The second waveform data includes... Figure 2A The first waveform data shown is the same waveform data, namely, the four waveforms of thermal radiation, visible light, reflected light, and laser light. For example, the laser waveform is taken as the third waveform, and the visible light waveform is taken as the fourth waveform. That is, the first waveform of the first waveform data and the third waveform of the second waveform data represent the same laser light, and the second waveform of the first waveform data and the fourth waveform of the second waveform data represent the same visible light light.

[0094] Generally speaking, laser processing, being a non-contact process, offers high reproducibility. Therefore, under the same processing conditions, the tendencies of the aforementioned thermal radiation, visible light, reflected light, and laser waveform will be identical in both the first device 300 and the second device 400. However, as... Figure 2A and Figure 2B As shown, due to deviations in light transmission efficiency or amplifier gain, the detection values ​​of thermal radiation, visible light, reflected light, and laser light in the first device 300 and the second device 400 may differ.

[0095] Therefore, when the processing state is estimated based on the processing data obtained during processing in the second device 400, the estimation result may not be correct, as the first estimation model generated by machine learning using the first waveform data is used to estimate the processing state.

[0096] Therefore, in this embodiment, the processing state estimation device 200 generates first transformed data, which is generated by transforming second waveform data obtained based on detection values ​​from the second device 400 according to a given coefficient and in conjunction with first waveform data. The first transformed data is data transformed in conjunction with first waveform data based on detection values ​​from the first device 300.

[0097] By transforming the second waveform data into the first transformed data, the processing state based on the first estimation model can be estimated for laser processing in the second device 400.

[0098] The information conversion unit 22 generates first conversion data by multiplying the waveforms of the second waveform data based on the detection values ​​in the second device 400 by a given coefficient. The given coefficient is calculated based on the relationship between waveforms of the same type. Specifically, the first coefficient is calculated based on the relationship between the third waveform (second laser) and the first waveform (first laser). Furthermore, the second coefficient is calculated based on the relationship between the fourth waveform (second visible light) and the second waveform (first visible light). When the first and second waveform data include three or more waveforms, coefficients are calculated for each waveform of the same type, and the waveforms included in the second waveform data are multiplied by the calculated coefficients.

[0099] Figure 3A It is a chart after the first waveform data has been standardized. Figure 3B This is a chart after standardizing the data from the first transformation. It should be noted that... Figure 3A The first waveform data shown is... Figure 2A The chart was standardized so that the laser intensity was represented by 1V data. That is, Figure 3A The chart is to Figure 2A The graph is a standardized graph obtained by multiplying the signal strength of each waveform by 0.34.

[0100] same, Figure 3B The first transformation data shown is to Figure 2B The graph and coefficients are multiplied and then standardized to ensure that the laser intensity is a 1V value. That is, Figure 3B The chart is to Figure 2A The graph is obtained by multiplying the laser intensity by 0.15 (first coefficient), the reflected light intensity by 0.15, the visible light intensity by 0.10 (second coefficient), and the thermal radiation intensity by 0.13. Each coefficient can be calculated, for example, based on the average signal intensity of each waveform of the first waveform data during the 1ms to 4ms period and the average signal intensity of each waveform of the second waveform data during the 1ms to 4ms period.

[0101] That is, a first coefficient is calculated to transform the average signal intensity of the second laser (third waveform) of the second waveform data over 1ms to 4ms into the average signal intensity of the first laser (first waveform) of the first waveform data over 1ms to 4ms. For example, the first coefficient is the ratio of the average signal intensity of the first laser (first waveform) to the average signal intensity of the second laser (third waveform). Similarly, a second coefficient is calculated to transform the average signal intensity of the second visible light (fourth waveform) of the second waveform data into the average signal intensity of the first visible light (second waveform) of the first waveform data. For example, the second coefficient is the ratio of the average signal intensity of the first visible light (second waveform) to the signal intensity of the second visible light (fourth waveform). By calculating the first and second coefficients in this way, the second waveform data can be transformed into the first transformed data while keeping the ratio of the average signal intensities of each waveform in the first waveform data unchanged.

[0102] It should be noted that data standardization is not necessary in the first waveform data and the first transformed data. The first transformed data can be generated by multiplying each waveform of the second waveform data with the given coefficients.

[0103] like Figure 3A and Figure 3B As shown, by multiplying each waveform of the second waveform data by a given coefficient to obtain the first transformation data, it is possible to obtain a waveform that represents the same trend as the first waveform data.

[0104] The storage presumption model generated in the storage presumption model generation device 100 of the storage unit 24.

[0105] The estimation unit 23 estimates the processing state based on the first estimation model and the first transformation data. If the first transformation data is input, the estimation unit 23 estimates the processing state based on the first estimation model and outputs the estimation result. The first transformation data maintains the ratio of the average values ​​of each waveform of the first waveform data. Therefore, if the first transformation data is set as the input to the estimation unit 23, the processing state can be estimated with high accuracy using the first estimation model.

[0106] <Verification>

[0107] Figure 4A This is a graph showing the relationship between the average signal strength and the defocus amount in the first waveform data and the first transformed data. The changes in the average signal strength of the first waveform data and the first transformed data when the defocus amount is varied were verified.

[0108] In the verification, the laser focal position near the surface of the workpiece was defined as 0mm defocus, and the first waveform data and first transformation data were obtained when the laser focal position left the surface of the workpiece. Figure 4A In, the average signal strength represents Figure 3A The standardized first waveform data shown below and Figure 3B The average signal strength during a time period of 1ms to 4ms in the first transformed data after standardization, as shown.

[0109] exist Figure 4A In the diagram, the first laser, first reflected light, first visible light, and first thermal radiation represent the average signal intensity of the first waveform data based on the detection values ​​in the first device 300. It should be noted that the first waveform data is normalized so that the average signal intensity of the first laser becomes 1V. Furthermore, the second laser, second reflected light, second visible light, and second thermal radiation represent the average signal intensity of the first transformed data generated by multiplying the second waveform data based on the detection values ​​in the second device 400 by various given coefficients. The first transformed data is also similarly normalized so that the average signal intensity of the second laser becomes 1V.

[0110] like Figure 4A As shown in the chart, it can be seen that the change in average signal strength as the defocusing amount increases represents approximately the same shift in both the first waveform data and the first transformed data. Therefore, the processing state estimated based on the first waveform data using the first estimation model and the processing state estimated based on the first transformed data using the first estimation model are approximately the same results.

[0111] In this embodiment, the information transformation unit 22 of the processing state estimation device 200 transforms the second waveform data, which is based on the detection values ​​in the second device 400, into first transformed data. In the estimation unit 23, the first transformed data can be used as input to estimate the processing state based on a first estimation model. That is, the first estimation model, which is generated using the first waveform data based on the detection values ​​in the first device 300, can be used in the second device 400.

[0112] Figure 4B This is a graph showing the relationship between the correct value and the estimated value when the processing state is estimated based on the first estimation model in the processing state estimation device 200, with the first transformation data and the second waveform data as input data. For example... Figure 4BAs shown in the chart, the estimated values ​​obtained by directly inputting the second waveform data into the estimation unit 23 without transforming it in the information conversion unit 22 are approximately distributed between -2.2mm and -1.6mm. This is a large difference compared to the correct value of -0.6mm to 0.3mm. On the other hand, when the information conversion unit 22 transforms the second waveform data into first transformed data and inputs the first transformed data into the estimation unit 23, the error between the estimated value and the correct value converges to within ±0.15mm. In this embodiment, it is clear that it is sufficient to detect defocus of approximately 0.5mm, thus obtaining a very practical estimation result.

[0113] [Effect]

[0114] According to the above embodiments, an estimation model generation apparatus and a processing state estimation apparatus can be provided that allow the estimation model generated by a given processing apparatus to be used even in other processing apparatuses.

[0115] In a production line, it is difficult to use conditions other than the standard processing conditions typically used in production. In the above embodiment, a first device 300, different from the device used in the production line, acquires sensing data suitable for a wide variety of processing conditions, and uses first waveform data based on this sensing data to generate an estimation model. Therefore, it is possible to generate an estimation model with higher estimation accuracy.

[0116] Furthermore, first transformation data is generated based on second waveform data from sensing data in the second device 400 used in the production line. In the estimation unit 23 of the processing state estimation device 200, by using the first transformation data as input, even when there are machine differences in sensors between devices, a more accurate estimation result can be output using the first estimation model to predict the processing state. That is, the first estimation model based on sensing data from the first device 300 can be used to estimate the processing state of other devices, including the second device 400.

[0117] (Implementation Method 2)

[0118] Reference Figures 5A-5B Embodiment 2 will be described. It should be noted that in Embodiment 2, structures that are the same as or equivalent to those in Embodiment 1 are described using the same reference numerals. Furthermore, in Embodiment 2, descriptions that are repeated in Embodiment 1 are omitted.

[0119] Figure 5A This is a block diagram illustrating the presumption model generation apparatus 110 according to Embodiment 2. Figure 5B This is a block diagram illustrating the processing state estimation device 210 according to Embodiment 2. Embodiment 2 differs from Embodiment 1 in that the estimation model generation device 110 includes an information conversion unit 17, while the processing state estimation device 210 does not.

[0120] In Embodiment 1, the first estimation model generated in the estimation model generation apparatus 100 based on the sensing data of the first device 300 can also be used to estimate the processing state of the second device 400. On the other hand, in Embodiment 2, the estimation model generation apparatus 110 generates a second estimation model for estimating the processing state of the first device 300 and a third estimation model for estimating the processing state of the second device 400.

[0121] The estimation model generation device 110 includes an information acquisition unit 16, an information transformation unit 17, an estimation model generation unit 18, and a storage unit 19.

[0122] Information acquisition unit 16 acquires third waveform data and fourth waveform data. The third waveform data includes data specific to the first device 300 (see reference). Figure 1C The fifth and sixth waveforms are observed during laser processing of the third thermal radiation, third visible light, third reflected light, and third laser light, representing at least two of these. That is, the third waveform data corresponds to the first waveform data in Embodiment 1. The fourth waveform data includes data for the second device 400 (refer to...). Figure 1D The seventh and eighth waveforms are at least two of the fourth thermal radiation, fourth visible light, fourth reflected light, and fourth laser light observed during laser processing. That is, the fourth waveform data corresponds to the second waveform data of Embodiment 1.

[0123] In this embodiment, the information transformation unit 17 of the estimation model generation device 110 transforms the third waveform data for the first device 300 in a way that matches the fourth waveform data for the second device 400, thereby generating second transformed data.

[0124] For example, if the third laser is taken as the fifth waveform and the third visible light as the sixth waveform, then the fourth laser is the seventh waveform and the fourth visible light is the eighth waveform. In this embodiment, the information conversion unit 17 converts the third waveform data into second converted data by multiplying the fifth waveform and the third coefficient, and multiplying the sixth waveform and the fourth coefficient. Just as the first coefficient in Embodiment 1 is calculated based on the relationship between the first and third waveforms, the third coefficient in Embodiment 2 is calculated based on the relationship between the fifth and seventh waveforms. Similarly, the fourth coefficient in Embodiment 2 is calculated based on the relationship between the sixth and eighth waveforms. The third and fourth coefficients are determined such that the second converted data is generated while keeping the ratio of the average value of the seventh waveform to the average value of the eighth waveform constant.

[0125] The estimation model generation unit 18 generates a second estimation model by performing machine learning on corresponding teaching data established using the third waveform data as an explanatory variable and the processing state as a target variable. The second estimation model is an estimation model for estimating the processing state for the first device 300. Furthermore, the estimation model generation unit 18 generates a third estimation model by performing machine learning on corresponding teaching data established using the second transformation data as an explanatory variable and the processing state as a target variable. The third estimation model is an estimation model for estimating the processing state for the second device 400.

[0126] The processing state estimation device 210 includes an information acquisition unit 26, an estimation unit 27, and a storage unit 28. The storage unit 28 stores a second estimation model and a third estimation model generated by the estimation model generation device 110. The information acquisition unit 26 acquires third waveform data and fourth waveform data. Based on the second estimation model, the estimation unit 28 estimates the processing state of the workpiece in the first device 300 according to the third waveform data. The estimation unit 28 also estimates the processing state of the workpiece in the second device 400 based on the third estimation model and the fourth waveform data.

[0127] In Embodiment 1, an example is described in which, during the estimation by the processing state estimation device 200, the processing state is estimated using a first estimation model for the first device 300 by transforming the second waveform data for the second device 400. In contrast, in this embodiment, the estimation model generation device 110 generates a second estimation model and a third estimation model for the first device 300 and the second device 400, respectively.

[0128] In this embodiment, the processing state estimation device 210 can estimate the processing state as input to the estimation unit 27 without specifically applying a transformation to the fourth waveform data based on the detection value in the second device 400.

[0129] [Effect]

[0130] According to the above-described embodiment, waveform data transformation processing is not performed in the processing state estimation device 210, thus shortening the estimation time required in the estimation unit 27.

[0131] [Industrial Applicability]

[0132] The estimation model generation device and processing state estimation device disclosed herein can be widely used for predicting the processing state in processing devices that perform laser processing.

[0133] [Explanation of reference numerals in the attached figures]

[0134] 100, 110 estimation model generation device

[0135] 11,16 Information Acquisition Department

[0136] 17 Information Transformation Department

[0137] 12, 18 Predicted Model Generation Department

[0138] Storage Units 13 and 19

[0139] 16 Information Acquisition Department

[0140] 200, 210 Processing Status Estimation Device

[0141] Information Acquisition Department, 21st and 26th

[0142] 22 Information Transformation Unit

[0143] Presumptions 23 and 27

[0144] Storage sections 24 and 28.

Claims

1. A processing state estimation device, which is used to estimate the processing state of laser processing in a second device that is different from a first device. The processing state of the first device is estimated by a first estimation model. The first estimation model is a model that uses machine learning to estimate the processing state of the first device by establishing corresponding teaching data with first waveform data as explanatory variables and the processing state of the first device as the target variable. The first waveform data includes a first waveform and a second waveform for at least two of the first thermal radiation, first visible light, first reflected light, and first laser light observed from the workpiece during laser processing based on the first device. The processing state estimation device includes: The information acquisition unit, during laser processing based on the second device, observes second thermal radiation, second visible light, second reflected light and second laser from the workpiece, and acquires second waveform data including a third waveform and a fourth waveform for at least two of the second thermal radiation, second visible light, second reflected light and second laser. The information transformation unit multiplies the third waveform by a first coefficient calculated based on the relationship between the third waveform and the first waveform, and multiplies the fourth waveform by a second coefficient calculated based on the relationship between the fourth waveform and the second waveform to generate first transformed data. Storage unit, storing the first presumed model; and The estimation unit, based on the first estimation model, estimates the processing state of the second device according to the first transformation data.

2. The processing state estimation device according to claim 1, wherein, The first coefficient and the second coefficient are determined such that the first transformed data is generated while keeping the ratio of the average values ​​of the first waveform and the second waveform constant.

3. A presumption model generation apparatus, which generates a presumption model for presuming the processing state of laser processing in a first apparatus and a second apparatus. The presumption model generation device includes: The information acquisition unit observes third thermal radiation, third visible light, third reflected light, and third laser light from the workpiece during laser processing based on the first device, and observes fourth thermal radiation, fourth visible light, fourth reflected light, and fourth laser light from the workpiece during laser processing based on the second device, and acquires third waveform data including a fifth waveform and a sixth waveform for at least two of the third thermal radiation, the third visible light, the third reflected light, and the third laser light, and fourth waveform data including a seventh waveform and an eighth waveform for at least two of the fourth thermal radiation, the fourth visible light, the fourth reflected light, and the fourth laser light; The information transformation unit multiplies the fifth waveform by a third coefficient calculated based on the relationship between the fifth waveform and the seventh waveform, and multiplies the sixth waveform by a fourth coefficient calculated based on the relationship between the sixth waveform and the eighth waveform to generate second transformed data. The estimation model generation unit generates a second estimation model estimating the processing state of the first device and a third estimation model estimating the processing state of the second device; and The storage unit stores the second presumption model and the third presumption model. The presumption model generation unit, Using the third waveform data as an explanatory variable and the processing state as the target variable, corresponding teaching data is established, and machine learning is performed to generate the second inference model. Using the second transformation data as an explanatory variable and the processing state as the target variable to establish corresponding teaching data, machine learning is performed to generate the third inference model.

4. The presumption model generation apparatus according to claim 3, wherein, The third and fourth coefficients are determined such that the second transformation data is generated while keeping the ratio of the average value of the seventh waveform to that of the eighth waveform constant.

5. A processing state estimation device, which estimates the processing state of laser processing, comprising: The storage unit stores the third presumption model generated by the presumption model generation apparatus according to claim 3 or 4; The information acquisition unit acquires the fourth waveform data; and The estimation unit estimates the processing state based on the third estimation model and the fourth waveform data.

Images