JP7364452B2 - Machine learning device, nozzle condition prediction device, and control device - Google Patents

Description

The present invention relates to a machine learning device, a nozzle state prediction device, and a control device.

A laser processing machine performs laser processing such as cutting or drilling a workpiece by irradiating the workpiece with laser light focused by a lens through the opening of a nozzle. At that time, the laser processing machine blows away the spatter, which is the workpiece melted by the laser beam, by spraying gas (hereinafter also referred to as "assist gas") onto the workpiece at high speed from the nozzle opening along with the laser beam. can.
However, due to heat generated during laser irradiation, adhesion of blown spatter, etc., the nozzle may become dirty, scratched, or deformed.
FIG. 9 is a diagram showing an example of an opening of a nozzle. FIG. 9 shows the shape of the opening when the nozzle is normal (on the left) and when the nozzle is deformed (on the right) when viewed from directly above the nozzle (laser light incident side). As shown on the right side of FIG. 9, when the nozzle is deformed, the opening of the nozzle is deformed by the heat generated during laser irradiation.
In this case, the processing accuracy of laser processing decreases, so it is necessary to replace the nozzle.
In this regard, an alignment camera installed in the optical head that irradiates the laser beam is used to photograph the nozzle opening from the upstream side of the laser beam path. A known technique is to determine the lifespan of a nozzle by comparing the image of the opening of the nozzle. For example, see Patent Document 1.

Patent No. 6194464

However, it is difficult for workers to judge the lifespan of a nozzle by comparing the image of the nozzle opening in normal condition and the image of the nozzle opening after use, that is, to determine the timing of maintenance such as nozzle cleaning or replacement. Is required. Furthermore, it is difficult to detect slight scratches or deformations from images. Furthermore, even if scratches or deformations occur in areas that cannot be captured by the alignment camera (blind spots), it is difficult to detect them from images.

Therefore, before performing laser processing, it is desired to generate a trained model that can predict the quality of the nozzle condition when the laser processing is performed.

(1) One aspect of the machine learning device of the present disclosure provides laser processing conditions for arbitrary laser processing on an arbitrary workpiece by an arbitrary laser processing machine, a degree of deterioration of a nozzle before laser processing due to the laser processing conditions, an input data acquisition unit that acquires input data including the input data; a label acquisition unit that acquires label data indicating the degree of deterioration of the nozzle after laser processing under the laser processing conditions included in the input data; and the input data A learning unit that performs supervised learning to generate a trained model using the input data acquired by the acquisition unit and the label data acquired by the label acquisition unit.

(2) One aspect of the nozzle state prediction device of the present disclosure is to use the learned model generated by the machine learning device according to (1) and, prior to the laser processing to be performed by the laser processing machine, to perform the laser processing according to the laser processing. an input section for inputting laser processing conditions and the degree of deterioration of the nozzle before the laser processing; and an input section for inputting the laser processing conditions and the degree of deterioration of the nozzle before the laser processing; A prediction unit that predicts the degree of deterioration of the nozzle.

(3) One aspect of the control device of the present disclosure includes the nozzle state prediction device described in (2).

According to one aspect, before performing laser processing, it is possible to generate a trained model that can predict whether the state of the nozzle is good or bad when the laser processing is performed.

FIG. 1 is a functional block diagram showing an example of a functional configuration of a numerical control system according to an embodiment. 2 is a diagram showing an example of a learned model provided to the nozzle state prediction device of FIG. 1. FIG. It is a figure which shows an example of the degree of deterioration of a nozzle. It is a figure which shows an example of a nozzle management table. It is a flowchart explaining prediction processing of a nozzle state prediction device in an operation phase. 2 is a diagram showing an example of a learned model provided to the nozzle state prediction device of FIG. 1. FIG. FIG. 1 is a diagram showing an example of the configuration of a numerical control system. FIG. 1 is a diagram showing an example of the configuration of a numerical control system. FIG. 3 is a diagram showing an example of an opening of a nozzle.

An embodiment of the present disclosure will be described below with reference to the drawings.
<One embodiment>
FIG. 1 is a functional block diagram showing an example of the functional configuration of a numerical control system according to an embodiment. As shown in FIG. 1, the numerical control system includes a control device 10, a laser processing machine 20, a nozzle state prediction device 30, and a machine learning device 40.

The control device 10, the laser processing machine 20, the nozzle state prediction device 30, and the machine learning device 40 may be directly connected to each other via a connection interface (not shown). Further, the control device 10, the laser processing machine 20, the nozzle state prediction device 30, and the machine learning device 40 may be interconnected via a network (not shown) such as a LAN (Local Area Network) or the Internet. In this case, the control device 10, the laser processing machine 20, the nozzle state prediction device 30, and the machine learning device 40 are equipped with a communication unit (not shown) for communicating with each other through such connections. Note that, as described later, the control device 10 may include a nozzle state prediction device 30 and a machine learning device 40. Further, the laser processing machine 20 may include the control device 10.

The control device 10 is a numerical control device known to those skilled in the art, generates an operation command based on control information, and transmits the generated operation command to the laser processing machine 20. Thereby, the control device 10 controls the operation of the laser processing machine 20. The control device 10 also outputs the control information to the nozzle state prediction device 30.
Note that the control information includes a processing program set in the control device 10 and laser processing conditions. The laser processing conditions include at least the output of the laser light (W), the frequency of the laser light (Hz), the laser irradiation time (or processing time), the type of assist gas to be used, and the blowout of the assist gas for the laser processing to be performed. The amount (L/sec) and a part of the focal length of the lens that condenses the laser beam may be included.
In addition, the control device 10 stores a list of identification information (hereinafter also referred to as "nozzle ID") regarding nozzles that can be selected for the laser processing machine 20 as a nozzle data table in a hard disk drive (not shown) or the like. Good too.
Note that the nozzle data table may include data such as the current degree of deterioration of the nozzle associated with each nozzle ID.

Further, the assist gas includes, for example, active gases such as oxygen, air, and nitrogen, and inert gases such as argon/helium. For example, when the material of the workpiece is carbon steel, low alloy steel, or stainless steel, an active gas such as oxygen, air, or nitrogen may be used as the assist gas. Further, when the material of the workpiece is an aluminum alloy, an active gas such as oxygen or air, or an inert gas such as argon/helium may be used as the assist gas. Further, when the material of the workpiece is a titanium alloy, an inert gas such as argon/helium or an active gas such as nitrogen may be used as the assist gas. Furthermore, when the material of the workpiece is acrylic, wood, paper, or fiber, an active gas such as nitrogen or air, or an inert gas such as argon/helium may be used as the assist gas.

The laser processing machine 20 is a machine that performs laser processing by irradiating a workpiece to be processed with laser light based on an operation command from the control device 10 . The laser processing machine 20 feeds back to the control device 10 information indicating the operating state based on the operation command from the control device 10 .

In the operation phase, the nozzle state prediction device 30 may acquire the laser processing conditions for the laser processing included in the control information from the control device 10, prior to the laser processing by the laser processing machine 20. Further, the nozzle state prediction device 30 may acquire information on the nozzle selected by the operator of the control device 10, including at least the current degree of deterioration of the nozzle, for example. The nozzle condition prediction device 30 inputs the acquired laser processing conditions and the current degree of deterioration of the nozzle into a trained model provided from a machine learning device 40, which will be described later, to predict whether the laser will use the selected nozzle. It is possible to predict the degree of deterioration of the nozzle after processing.

Before explaining the nozzle condition prediction device 30, "degree of nozzle deterioration" and machine learning for generating a learned model will be explained.

<About the degree of nozzle deterioration>
The "degree of nozzle deterioration" indicates the degree of dirt or deformation in the opening of the nozzle used in laser processing by the laser processing machine 20, as shown in FIG. For example, a new nozzle or a nozzle that has just undergone maintenance such as cleaning, as shown on the left side of FIG. 9, has a circular opening and is in the best condition, so there is no need for maintenance or replacement. The "degree of nozzle deterioration" is "0%". The "degree of deterioration of the nozzle" is determined as "100%" when the nozzle is used, and as the nozzle is used, spatter adheres to the opening and deformation due to heat during laser irradiation increases, for example, when the nozzle needs to be replaced. becomes.
The "degree of nozzle deterioration" was a percentage value in the range of "0%" to "100%", but for example, the "degree of nozzle deterioration" was a percentage value in the range of "0" to "1". It may be a value, etc.

<Machine learning device 40>
For example, the machine learning device 40 is configured to input data including, in advance, laser processing conditions for arbitrary laser processing on an arbitrary workpiece by an arbitrary laser processing machine, and the degree of deterioration of the nozzle before laser processing due to the laser processing conditions. get.
The machine learning device 40 also obtains data indicating the degree of deterioration of the nozzle after laser processing due to the laser processing conditions of the obtained input data as a label (correct answer).
The machine learning device 40 performs supervised learning using the acquired training data of input data and label pairs, and constructs a trained model to be described later.
By doing so, the machine learning device 40 can provide the constructed learned model to the nozzle state prediction device 30.
The machine learning device 40 will be specifically explained.

As shown in FIG. 1, the machine learning device 40 includes an input data acquisition section 401, a label acquisition section 402, a learning section 403, and a storage section 404.
In the learning phase, the input data acquisition unit 401 acquires, via a communication unit (not shown), laser processing conditions for arbitrary laser processing on an arbitrary workpiece using an arbitrary laser processing machine, and the nozzle condition before laser processing according to the laser processing conditions. The degree of deterioration is acquired from the control device 10 or the like as input data.
As mentioned above, the laser processing conditions include the output of the laser light (W), the frequency of the laser light (Hz), the laser irradiation time (or processing time), the type of assist gas used, and the blowout of the assist gas. This includes the amount (L/sec), the focal length of the lens that condenses the laser beam, etc.
The input data acquisition unit 401 outputs the acquired input data to the storage unit 404.

The label acquisition unit 402 acquires data indicating the degree of deterioration of the nozzle after laser processing due to the laser processing conditions included in the input data as label data (correct data), and stores the acquired label data in the storage unit 404. Output.
Note that, as described above, the label is a value ranging from "0%" to "100%".

The learning unit 403 receives the above-described combination of input data and label as training data, and performs supervised learning using the received training data to determine the laser processing conditions to be performed on the workpiece to be processed, A learned model 351 that predicts the degree of nozzle deterioration after laser processing is constructed based on the current degree of nozzle deterioration of the selected nozzle.
The learning unit 403 then provides the constructed learned model 351 to the nozzle state prediction device 30.

Here, it is desirable to prepare a large amount of training data for performing supervised learning. For example, an expert may create training data in which the degree of deterioration of the nozzle before and after laser processing is evaluated for each combination of every laser processing condition and every nozzle. Alternatively, the training data may be acquired from each of the control devices 10 in various locations actually operating in a customer's factory or the like.

FIG. 2 is a diagram showing an example of the trained model 351 provided to the nozzle state prediction device 30 of FIG. 1. Here, as shown in FIG. 2, the trained model 351 uses the laser processing conditions to be performed and the current degree of nozzle deterioration of the selected nozzle as input layers, and the nozzle after laser processing under these laser processing conditions. A multilayer neural network whose output layer is data indicating the degree of deterioration of is illustrated.
Here, the laser processing conditions to be performed include the output of the laser beam, the frequency of the laser beam, the laser irradiation time, the type of assist gas, the ejection amount of the assist gas, the focal length of the lens, and the like.

In addition, when the learning unit 403 acquires new teacher data after constructing the trained model 351, the learning unit 403 further performs supervised learning on the trained model 351. may be updated.
Further, the machine learning device 40 may share the trained model 351 with other machine learning devices 40 to perform supervised learning. By doing so, it becomes possible to perform supervised learning by distributing the trained models 351 in each of the plurality of machine learning devices 40, and it becomes possible to improve the efficiency of supervised learning.

The supervised learning described above may be performed by online learning, batch learning, or mini-batch learning.
Online learning is a learning method in which supervised learning is performed immediately every time laser processing is performed by the laser processing machine 20 and training data is created. In addition, batch learning means that while laser processing is performed by the laser processing machine 20 and training data is created repeatedly, a plurality of training data are collected according to the repetition, and all the collected training data are This is a learning method that uses supervised learning to perform supervised learning. Furthermore, mini-batch learning is a learning method that is intermediate between online learning and batch learning, in which supervised learning is performed every time a certain amount of training data is accumulated.

The storage unit 404 is a RAM (Random Access Memory) or the like, and stores the input data acquired by the input data acquisition unit 401, the label data acquired by the label acquisition unit 402, and the learned model 351 constructed by the learning unit 403. etc. to be memorized.
The machine learning for generating the learned model 351 included in the nozzle state prediction device 30 has been described above.
Next, the nozzle state prediction device 30 in the operation phase will be explained. In the following explanation, the learned model shown in Figure 2 is used as the learned model, that is, the laser processing conditions to be performed and the degree of deterioration of the selected nozzle are used as the input layer, and the "degree of nozzle deterioration" is used as the output layer. An example of using a multilayer neural network with .

<Nozzle state prediction device 30 in operation phase>
As shown in FIG. 1, the nozzle state prediction device 30 in the operation phase includes an input section 301, a prediction section 302, a determination section 303, a notification section 304, and a storage section 305.
Note that the nozzle state prediction device 30 includes an arithmetic processing device (not shown) such as a CPU (Central Processing Unit) in order to realize the operations of the functional blocks shown in FIG. In addition, the nozzle state prediction device 30 is used in an auxiliary storage device (not shown) such as a ROM (Read Only Memory) or an HDD that stores various control programs, or in an arithmetic processing unit that is temporarily required to execute the programs. A main storage device (not shown) such as a RAM is provided for storing data.

Then, in the nozzle condition prediction device 30, the arithmetic processing unit reads the OS and application software from the auxiliary storage device, expands the read OS and application software to the main storage device, and performs arithmetic processing based on these OS and application software. I do. Based on this calculation result, the nozzle state prediction device 30 controls each piece of hardware. Thereby, the processing by the functional blocks of FIG. 1 is realized. That is, the nozzle state prediction device 30 can be realized by cooperation of hardware and software.

Prior to laser processing by the laser processing machine 20, the input unit 301 inputs, for example, the laser processing conditions to be performed from the control information of the control device 10 and the nozzle ID indicating the nozzle selected by the operator. do. The input unit 301 obtains the current degree of deterioration of the nozzle based on a nozzle management table 352 stored in the storage unit 305, which will be described later, and the input nozzle ID.
The input unit 301 outputs the acquired laser processing conditions to be performed from now on and the current degree of deterioration of the nozzle selected by the operator to the prediction unit 302.

The prediction unit 302 inputs the laser processing conditions to be performed and the current degree of deterioration of the nozzle selected by the operator into the learned model 351 of FIG. 2, and calculates the degree of deterioration of the selected nozzle after laser processing. Predict.
Thereby, the nozzle condition prediction device 30 can predict whether the nozzle condition will be good or bad when the laser processing is performed, before performing the laser processing.

The determining unit 303 determines whether the selected nozzle is appropriate based on the degree of deterioration of the nozzle after laser processing predicted by the predicting unit 302.
More specifically, the determining unit 303 determines whether the selected nozzle is appropriate based on a comparison between the predicted value of the degree of deterioration of the nozzle and a preset threshold value, and if the selected nozzle is not appropriate, the nozzle is Determine the best time to perform maintenance or replace the nozzle.

FIG. 3 is a diagram showing an example of the degree of deterioration of a nozzle. As shown in Figure 3, for example, even if the degree of deterioration of the nozzle is "0%" immediately after replacement or maintenance, the nozzle may still suffer from spatter adhesion due to laser processing or melting (deformation) due to heat during laser irradiation. The degree of deterioration changes depending on the For example, when the degree of deterioration of the nozzle is an appropriate value in the range of "0%" to "30%", the nozzle functions properly. On the other hand, if the degree of deterioration of the nozzle is outside the range of appropriate values, the nozzle will not function properly. In this case, the worker needs to perform maintenance such as cleaning the nozzle or replace the nozzle so that the degree of deterioration of the nozzle becomes an appropriate value.
In the following description, the degree of nozzle deterioration of "30%" is taken as the lower limit threshold value α ₁ , and the degree of nozzle deterioration of "80%" is taken as the upper limit threshold value α ₂ . Note that the threshold values α ₁ and α ₂ may be set as appropriate depending on the processing accuracy of the workpiece to be processed, the environment in which the laser processing machine 20 is installed, and the like.

The determining unit 303 determines whether the degree of nozzle deterioration predicted by the predicting unit 302 is an appropriate value that is equal to or less than the lower limit threshold value _α1 . If the predicted degree of deterioration of the nozzle is an appropriate value, the determining unit 303 determines not to perform maintenance or replacement of the nozzle before laser processing according to the laser processing conditions input by the input unit 301.
On the other hand, if the predicted degree of deterioration of the nozzle is higher than the lower threshold α ₁ and lower than the upper threshold α ₂ , the determining unit 303 performs nozzle maintenance before laser processing according to the input laser processing conditions. Determine the timing. Furthermore, when the predicted degree of deterioration of the nozzle is higher than the upper limit threshold _α2 , the determining unit 303 determines that it is time to replace the nozzle before laser processing according to the input laser processing conditions.
As a result, the numerical control system predicts the degree of deterioration of the nozzle selected by the operator prior to laser processing by the laser processing machine 20 after laser processing under the laser processing conditions to be performed. The optimal time to perform maintenance or replacement can be easily detected. In other words, the numerical control system can detect in advance whether maintenance or replacement of the nozzle is required during laser processing.

Note that the operator may adjust the laser processing conditions. For example, if there are multiple workpieces to be machined included in the laser processing conditions, the operator may adjust the number of workpieces to be machined so that the degree of nozzle deterioration after laser processing does not exceed the lower limit threshold _α1 . As a result, the laser irradiation time may be adjusted. Thereby, the numerical control system can predict in advance the timing at which nozzle maintenance or replacement will be required after laser processing has been performed, before starting laser processing. Then, the numerical control system may, for example, stop the laser processing machine 20 and perform nozzle maintenance or replace the nozzle at a suitable stage of laser processing.
Furthermore, the determining unit 303 may automatically adjust the laser processing conditions as described later.

For example, the notification unit 304 outputs the determination result by the determination unit 303 as to whether the selected nozzle is appropriate to an output device (not shown) such as a liquid crystal display included in the control device 10 and/or the laser processing machine 20. You may. For example, if the selected nozzle is appropriate, the control device 10 may cause the laser processing machine 20 to perform laser processing under the input laser processing conditions. On the other hand, if the selected nozzle is not appropriate, the notification unit 304 may output an instruction for maintenance or replacement of the determined nozzle to the control device 10 and/or the output device (not shown) of the laser processing machine 20. . Further, the notification unit 304 may notify by voice via a speaker (not shown).

The storage unit 305 is a ROM, HDD, or the like, and may store a learned model 351 and a nozzle management table 352 along with various control programs.

<Nozzle management table 352>
FIG. 4 is a diagram showing an example of the nozzle management table 352. As shown in FIG. 4, the nozzle management table 352 includes "nozzle IDs" for identifying all the nozzles managed to be usable for laser processing. Further, the nozzle management table 352 includes "degree of nozzle deterioration" corresponding to the nozzle ID.

The "nozzle ID" in the nozzle management table 352 is information for identifying the nozzle selected by the operator of the control device 10, and is set in advance by the operator. In FIG. 4, the nozzle ID is set, for example, as a number from 1 to n uniquely assigned to each nozzle, but it may also be set as an alphabet. Note that n is an integer of 2 or more.

The "degree of nozzle deterioration" in the nozzle management table 352 stores the current degree of deterioration of the nozzle determined by the operator after the previous laser processing of the nozzle to which each of the above-mentioned nozzle IDs has been assigned. That is, the "degree of nozzle deterioration" in the nozzle management table 352 is input and updated by the operator after determining the degree of nozzle deterioration each time the nozzle is used for laser processing.

<Prediction processing of the nozzle state prediction device 30 in the operation phase>
Next, the operation related to prediction processing of the nozzle state prediction device 30 according to this embodiment will be explained.
FIG. 5 is a flowchart illustrating the prediction process of the nozzle state prediction device 30 in the operation phase.

In step S11, prior to laser processing by the laser processing machine 20, the input unit 301 acquires the laser processing conditions to be performed from the control information of the control device 10, and also acquires the laser processing conditions to be performed from now on based on the nozzle ID of the nozzle selected by the operator. , the degree of nozzle deterioration is acquired from the nozzle management table 352.

In step S12, the prediction unit 302 inputs the laser processing conditions to be performed and the degree of deterioration of the nozzle selected by the operator, which were acquired in step S11, into the learned model 351, and Predict the "degree of nozzle deterioration".

In step S13, the determining unit 303 determines whether the selected nozzle is appropriate based on the comparison between the degree of deterioration of the nozzle predicted in step S12 and the threshold values α ₁ and α ₂ .

In step S14, if the nozzle is determined to be inappropriate in step S13, the notification unit 304 sends an instruction to maintain or replace the nozzle before laser processing to an output device (not shown) of the control device 10 and/or the laser processing machine 20. Output to.

As described above, the nozzle state prediction device 30 according to one embodiment determines, from the control information of the control device 10, the laser processing conditions to be performed regarding the laser processing and the laser processing conditions selected by the operator, prior to laser processing by the laser processing machine 20. The degree of deterioration of the selected nozzle is input to the learned model 351 to predict the degree of deterioration of the selected nozzle after laser processing. Thereby, the nozzle condition prediction device 30 can predict whether the nozzle condition is good or bad (degree of deterioration) when the laser processing is performed, before performing the laser processing. Then, the nozzle condition prediction device 30 compares the predicted degree of deterioration of the nozzle with preset threshold values α ₁ and α ₂ to determine whether maintenance or replacement of the nozzle is required during laser processing. can be detected in advance.
That is, if the degree of deterioration of the selected nozzle exceeds the threshold α ₁ or the threshold α ₂ , the nozzle condition prediction device 30 can detect this as the optimal time for maintenance or replacement of the nozzle before laser processing. The burden on workers can be reduced.
Further, since the trained model 351 is generated using teacher data of various nozzles, the nozzle state prediction device 30 can predict the quality (degree of deterioration) of the states of various nozzles.

Although one embodiment has been described above, the nozzle state prediction device 30 and the machine learning device 40 are not limited to the above-described embodiment, and include modifications, improvements, etc. within a range that can achieve the purpose.
<Modification 1>
In the above-described embodiment, the machine learning device 40 is exemplified as a device different from the control device 10, the laser processing machine 20, and the nozzle state prediction device 30. , the laser processing machine 20 or the nozzle state prediction device 30 may be provided.

<Modification 2>
In the above-described embodiment, the nozzle state prediction device 30 is exemplified as a device different from the control device 10 and the laser processing machine 20, but some or all of the functions of the nozzle state prediction device 30 can be performed by the control device 10 or the laser processing machine. 20 may be provided.
Alternatively, a server may include a part or all of the input section 301, the prediction section 302, the determination section 303, the notification section 304, and the storage section 305 of the nozzle state prediction device 30, for example. Further, each function of the nozzle state prediction device 30 may be realized using a virtual server function or the like on the cloud.
Further, the nozzle state prediction device 30 may be a distributed processing system in which each function of the nozzle state prediction device 30 is distributed to a plurality of servers as appropriate.

<Modification 3>
For example, in the above embodiment, the learned model 351 in FIG. 3 is a multilayer model in which the laser processing conditions to be performed and the degree of deterioration of one nozzle are used as input layers, and the "degree of deterioration of a nozzle" is used as an output layer. Although it is a neural network, it is not limited to this. For example, as shown in FIG. 6, the learned model 351 uses M nozzles as input layers, including the laser processing conditions to be performed and the degree of deterioration of each of a plurality of nozzles (M: M≧2) selected by the operator. It may be a neural network whose output layer is the "degree of nozzle deterioration" of each nozzle.
As a result, when there are M nozzles that can be used for laser processing to be performed, the operator can select the M nozzle IDs, and the nozzle condition prediction device 30 can predict the "nozzle deterioration" of each of the M nozzles. It is possible to perform predictions in parallel on the degree of ``intensity'' and to quickly select a nozzle suitable for laser processing, thereby increasing the efficiency of laser processing.
FIG. 6 is a diagram showing an example of a trained model provided to the nozzle state prediction device 30 of FIG. 1.
Note that the machine learning device 40 constructs learned models 351 in advance for each of M from 2 to n, and the nozzle state prediction device 30 constructs learned models 351 according to the number of nozzles selected by the operator. You may choose.

<Modification 4>
For example, in the above-described embodiment, if the degree of deterioration of the nozzle selected by the operator after laser processing exceeds the threshold α ₁ or the threshold α ₂ , the determination unit 303 of the nozzle condition prediction device 30 Although this is determined as the time to perform maintenance or replacement on the selected nozzle before processing, it is not limited thereto.
As described above, the determining unit 303 determines whether the degree of deterioration of the nozzle after laser processing exceeds a threshold value, for example, when the laser irradiation time under the laser processing conditions includes a plurality of workpieces (for example, 10 workpieces) to be laser processed. To avoid this, the laser irradiation time may be adjusted to adjust the number of workpieces.
More specifically, for example, if the degree of deterioration of the nozzle exceeds a threshold by laser processing a plurality of workpieces with the "laser irradiation time" of the laser processing conditions to be performed from now on, the determining unit 303 further determines that ( By setting the "laser irradiation time" in the laser processing conditions (according to the operator's instructions) to an integral multiple (k times: k≧1) of the laser processing time required for one workpiece, and having the prediction unit 302 predict the degree of nozzle deterioration. , a value of k (for example, 6, etc.) that does not exceed a threshold value may be searched.
Then, the determining unit 303 determines, for example, after the laser processing of the k workpieces as the time to perform maintenance or replacement of the selected nozzle. The numerical control system of FIG. 1 can stop the laser processing machine 20 and perform maintenance or replacement of the nozzle at a determined time.
Thereby, the numerical control system can predict in advance the timing of how much laser processing will require maintenance or replacement of the nozzle before starting laser processing. In addition, the numerical control system can avoid the need for nozzle maintenance or replacement during laser processing, and can avoid deterioration in processing quality.

<Modification 5>
For example, in the above-described embodiment, the nozzle state prediction device 30 uses the learned model 351 provided by the machine learning device 40 to determine whether the nozzle state prediction device 30 uses the learned model 351 provided by the machine learning device 40 to perform the laser processing after laser processing based on the laser processing conditions to be performed, which are acquired from one control device 10. Although the degree of deterioration of the nozzle selected by the operator has been predicted, the present invention is not limited thereto. For example, as shown in FIG. 7, the server 50 stores the learned model 351 generated by the machine learning device 40, and the server 50 stores the learned model 351 generated by the machine learning device 40, and ) may share the trained model 351 (m is an integer of 2 or more). Thereby, the learned model 351 can be applied even if a new laser processing machine, control device, and nozzle state prediction device are installed.
Note that each of the nozzle condition prediction devices 30A(1)-30A(m) is connected to each of the control devices 10A(1)-10A(m), and each of the control devices 10A(1)-10A(m) is connected to each of the control devices 10A(1)-10A(m). , and laser processing machines 20A(1) to 20A(m).
Further, each of the control devices 10A(1) to 10A(m) corresponds to the control device 10 in FIG. Each of the laser processing machines 20A(1) to 20A(m) corresponds to the laser processing machine 20 in FIG. Each of the nozzle state prediction devices 30A(1) to 30A(m) corresponds to the nozzle state prediction device 30 in FIG.
Alternatively, as shown in FIG. 8, the server 50 operates as, for example, the nozzle state prediction device 30, and the server 50 operates as the nozzle state prediction device 30, and informs each of the control devices 10A(1) to 10A(m) connected to the network 60 about the laser information to be performed from now on. The degree of deterioration of the nozzle under processing conditions may be predicted. Thereby, the learned model 351 can be applied even if a new laser processing machine and control device are installed.

In one embodiment, each function included in the nozzle state prediction device 30 and the machine learning device 40 can be realized by hardware, software, or a combination thereof. Here, being realized by software means being realized by a computer reading and executing a program.

Each component included in the nozzle state prediction device 30 and the machine learning device 40 can be realized by hardware including an electronic circuit, software, or a combination thereof. When realized by software, programs constituting this software are installed on the computer. Furthermore, these programs may be recorded on removable media and distributed to users, or may be distributed by being downloaded to users' computers via a network. In addition, when configured with hardware, part or all of the functions of each component included in the above device may be implemented using, for example, an ASIC (Application Specific Integrated Circuit), a gate array, an FPGA (Field Programmable Gate Array), a CPLD ( It can be configured with an integrated circuit (IC) such as a complex programmable logic device.

The program can be stored and provided to the computer using various types of non-transitory computer readable media. Non-transitory computer-readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic recording media (e.g., flexible disks, magnetic tape, hard disk drives), magneto-optical recording media (e.g., magneto-optical disks), CD-ROMs (Read Only Memory), and CD-ROMs. R, CD-R/W, semiconductor memory (including, for example, mask ROM, PROM (Programmable ROM), EPROM (Erasable PROM), flash ROM, RAM). Programs may also be stored in various types of temporary computer-readable media. (Transitory computer readable medium). Examples of transitory computer readable media include electrical signals, optical signals, and electromagnetic waves. Transient computer readable media include electrical wires, optical fibers, etc. The program can be supplied to the computer via a wired communication path or a wireless communication path.

Note that the step of writing a program to be recorded on a recording medium includes not only processes that are performed in chronological order, but also processes that are not necessarily performed in chronological order but are executed in parallel or individually. It also includes.

In other words, the machine learning device, nozzle state prediction device, and control device of the present disclosure can take various embodiments having the following configurations.

(1) The machine learning device 40 of the present disclosure includes laser processing conditions for arbitrary laser processing on an arbitrary workpiece by an arbitrary laser processing machine 20, and a degree of deterioration of a nozzle before laser processing due to the laser processing conditions. An input data acquisition unit 401 that acquires input data; a label acquisition unit 402 that acquires label data indicating the degree of deterioration of the nozzle after laser processing due to laser processing conditions included in the input data; The learning unit 403 executes supervised learning using the acquired input data and the label data acquired by the label acquisition unit 402 to generate a trained model 351.
According to this machine learning device 40, before performing laser processing, it is possible to generate a learned model that can predict the quality (degree of deterioration) of the nozzle condition when the laser processing is performed.

(2) In the machine learning device 40 described in (1), the laser processing conditions include at least the output of the laser beam, the frequency of the laser beam, the processing time or the laser irradiation time, the type of gas used, the amount of gas ejected, and a part of the focal length of the lens that condenses the laser beam.
By doing so, the machine learning device 40 can generate a learned model 351 that can accurately predict the quality (degree of deterioration) of the nozzle condition.

(3) The nozzle state prediction device 30 of the present disclosure uses the learned model 351 generated by the machine learning device described in (1) or (2), and the laser An input section for inputting the laser processing conditions related to processing and the degree of deterioration of the nozzle before laser processing, and the learned model 351 are used to input the laser processing conditions after laser processing according to the laser processing conditions input by the input section 301. A prediction unit 302 that predicts the degree of deterioration of the nozzle is provided.
According to this nozzle condition prediction device 30, before performing laser processing, it is possible to predict whether the nozzle condition will be good or bad (degree of deterioration) when the laser processing is performed.

(4) In the nozzle condition prediction device 30 described in (3), it is determined whether the nozzle is appropriate based on a comparison between the degree of deterioration of the nozzle predicted by the prediction unit 302 and a preset threshold value. You may further include a determining unit 303 that determines whether the
By doing so, the nozzle condition prediction device 30 can accurately determine whether the nozzle condition is good or bad (degree of deterioration) when the laser processing is performed, before performing the laser processing.

(5) In the nozzle condition prediction device 30 described in (4), if the nozzle is not appropriate, the determining unit 303 may determine whether it is time to perform maintenance or replacement of the nozzle before laser processing. .
By doing so, the nozzle state prediction device 30 can avoid maintenance or replacement of the nozzle during laser processing, and can avoid deterioration in processing quality.

(6) In the nozzle condition prediction device 30 described in (4), if the nozzle is not appropriate, the determining unit 303 compares the degree of deterioration of the nozzle predicted by the predicting unit 302 with a preset threshold value. Based on this, the laser processing conditions may be changed so that the degree of deterioration of the nozzle does not exceed the threshold value.
By doing so, the nozzle condition prediction device 30 can predict in advance the timing of how much laser processing will require maintenance or replacement of the nozzle before starting laser processing.

(7) In the nozzle state prediction device 30 according to any one of (3) to (6), the trained model 351 is provided in the server 50 that is accessible from the nozzle state prediction device 30 via the network 60. You can.
By doing so, the learned model 351 can be applied even if a new control device 10, laser processing machine 20, and nozzle state prediction device 30 are installed.

(8) The nozzle state prediction device 30 described in any one of (3) to (7) may include the machine learning device 40 described in (1) or (2).
By doing so, the nozzle state prediction device 30 can achieve the same effects as any of the above (1) to (7).

(9) The control device 10 of the present disclosure may include the nozzle state prediction device 30.
According to this control device 10, the same effects as any of the above (1) to (8) can be achieved.

10 Control device 20 Laser processing machine 30 Nozzle state prediction device 40 Machine learning device 301 Input section 302 Prediction section 303 Determination section 304 Notification section 305 Storage section 351 Learned model 352 Nozzle management table 401 Input data acquisition section 402 Label acquisition section 403 Learning Section 404 Storage section

Claims (9)

an input data acquisition unit that acquires input data including laser processing conditions for arbitrary laser processing on an arbitrary workpiece by an arbitrary laser processing machine, and a degree of deterioration of a nozzle before laser processing due to the laser processing conditions;
a label acquisition unit that acquires label data indicating a degree of deterioration of the nozzle after laser processing according to the laser processing conditions included in the input data;
a learning unit that performs supervised learning to generate a trained model using the input data acquired by the input data acquisition unit and the label data acquired by the label acquisition unit;
A machine learning device equipped with The laser processing conditions include at least the output of the laser beam, the frequency of the laser beam, the processing time or the laser irradiation time, the type of gas used, the ejection amount of the gas, and the focal length of the lens that focuses the laser beam. The machine learning device according to claim 1, comprising a part of. A learned model generated by the machine learning device according to claim 1 or claim 2,
an input unit for inputting laser processing conditions for the laser processing and the degree of deterioration of the nozzle before the laser processing, prior to the laser processing to be performed by the laser processing machine;
a prediction unit that uses the learned model to predict the degree of deterioration of the nozzle after laser processing according to the laser processing conditions input by the input unit;
A nozzle condition prediction device. The nozzle according to claim 3, further comprising a determining unit that determines whether the nozzle is appropriate based on a comparison between the degree of deterioration of the nozzle predicted by the predicting unit and a preset threshold value. Condition prediction device. The nozzle condition prediction device according to claim 4, wherein the determining unit determines whether or not it is time to perform maintenance or replacement of the nozzle before the laser processing, if the nozzle is not appropriate. When the nozzle is inappropriate, the determining unit determines whether the degree of deterioration of the nozzle exceeds the threshold based on a comparison between the degree of deterioration of the nozzle predicted by the prediction unit and a preset threshold. The nozzle state prediction device according to claim 4, wherein the laser processing conditions are changed so that the laser processing conditions do not occur. The nozzle state prediction device according to any one of claims 3 to 6, wherein the learned model is provided in a server that is accessible from the nozzle state prediction device via a network. The nozzle state prediction device according to any one of claims 3 to 7, comprising the machine learning device according to claim 1 or 2. A control device comprising the nozzle state prediction device according to any one of claims 3 to 8.

Images