WO2020179063A1

WO2020179063A1 - Machine learning device, numerical control device, abnormality estimation device, and machine tool control system

Info

Publication number: WO2020179063A1
Application number: PCT/JP2019/009157
Authority: WO
Inventors: 恵介山中; 和行佐藤
Original assignee: 三菱電機株式会社
Priority date: 2019-03-07
Filing date: 2019-03-07
Publication date: 2020-09-10
Also published as: DE112019006825T5; JPWO2020179063A1; JP6608102B1; CN113498497A

Abstract

This machine learning device (1), which learns an abnormality generation condition of a machining tool for processing an object to be processed, is characterized by comprising: a state observation unit (11) that observes, as state variables, tool service life information indicating the service life of a tool and machining conditions of the machining tool (2); and a learning unit (12) that learns the abnormality generation condition according to a data set which is created on the basis of the state variables and presence/absence of a notification regarding the abnormality of the machining tool (2).

Description

Machine learning device, numerical control device, abnormality estimation device, and machine tool control system

The present invention relates to a machine learning device, a numerical control device, an abnormality estimating device, and a machine tool control system capable of estimating an abnormality of a machine tool controlled by a numerical control device.

A machine tool that uses a tool to process an object to be machined, when the actuator, such as a servomotor or spindle motor, is subjected to a load, the numerical control device will perform operations such as "overload," "overspeed deviation," "overheat. It has the function of notifying the abnormality of the machine and stopping the machining. When the abnormality of the machine tool is notified, it is necessary to review the processing conditions and try the processing many times, which requires labor and time.

Patent Document 1 discloses a machine learning device that observes electric energy, temperature, load, etc. of an electric motor as state variables and learns an operation command to a motor that avoids a machine tool abnormality.

JP, 2017-64837, A

The technology described in Patent Document 1 learns the operation command by observing the operation status of the electric motor as a state variable. However, the abnormality of the machine tool, that is, the operation status of the electric motor depends not only on the operation command but also on the tool status of the machine tool. The state of the tool of the machine tool varies depending on the elapsed time from the start of use and the usage situation, but depending on the state of the tool, whether or not an abnormality occurs may change. Therefore, the above-mentioned conventional technique has a problem in that the learning accuracy of the abnormality occurrence condition of the machine tool is reduced, and it is difficult to accurately estimate whether or not the abnormality has occurred.

The present invention has been made in view of the above, and an object of the present invention is to obtain a machine learning device capable of improving the learning accuracy of an abnormality occurrence condition of a machine tool.

In order to solve the above-mentioned problems and achieve the object, the present invention is a machine learning device that learns an abnormality occurrence condition of a machine tool that processes a workpiece, and tool life information indicating the life of the tool, A machining condition of the machine tool, a state observing unit that observes the condition variable as a state variable, and a learning unit that learns the abnormality generating condition according to a data set created based on the state variable and the presence/absence of notification of abnormality of the machine tool. It is characterized by being provided.

The machine learning device according to the present invention has an effect that it is possible to improve the learning accuracy of the abnormality occurrence condition of the machine tool.

The figure which shows the structure of the machine learning system concerning Embodiment 1 of this invention. The figure which shows the three-layer neural network model which the learning part shown in FIG. 1 can use. The figure which shows the structure of the control system of the machine tool concerning Embodiment 2 of this invention. The figure which shows the function structure of the abnormality estimation apparatus shown in FIG. The figure which shows an example of the display screen which HMI shown in FIG. 3 outputs. Flowchart for explaining the operation of the control system shown in FIG. FIG. 1 is a diagram showing dedicated hardware for realizing the functions of a machine learning device, a numerical control device, and an abnormality estimation device according to the first and second embodiments of the present invention. FIG. 3 is a diagram showing a configuration of a control circuit for realizing the functions of the machine learning device, the numerical control device, and the abnormality estimating device according to the first and second embodiments of the present invention.

A machine learning device, a numerical control device, an abnormality estimation device, and a machine tool control system according to the embodiment of the present invention will be described below in detail with reference to the drawings. The present invention is not limited to this embodiment.

Embodiment 1.
FIG. 1 is a diagram showing a configuration of a machine learning system 100 according to a first embodiment of the present invention. The machine learning system 100 includes a machine learning device 1, a machine tool 2, a numerical controller 3, and an HMI (Human Machine Interface) 4.

The machine learning device 1 is a device that learns an abnormality occurrence condition that is a condition for causing an abnormality of the machine tool 2. Since the machine learning device 1 determines that an abnormality has occurred by using the presence or absence of the abnormality notification, it can be said that the machine learning device 1 is an apparatus that learns an abnormality notification condition that is a condition for notifying the abnormality of the machine tool 2. The machine tool 2 is a device that processes an object to be machined by changing the position of a tool. The numerical control device 3 is a device that numerically controls the machine tool 2. The HMI4 is an input / output device for exchanging information between humans and machines. The user can use the HMI 4 to input the processing conditions of the machine tool 2 into the numerical controller 3.

The machine tool 2 has a drive unit 21 and a detection unit 22. The drive unit 21 has a function of changing the position of the tool, and includes, for example, an actuator such as a servo motor or a spindle motor, a control unit that controls the actuator according to a command from the numerical control device 3, and the like. The detection unit 22 has a function of detecting the state of the machine tool 2. The state of the machine tool 2 detected by the detection unit 22 is, for example, the temperature of the machine tool 2. The temperature of the machine tool 2 includes the temperature of the servo motor and the temperature of the spindle motor included in the machine tool 2. The detection unit 22 outputs the detected information to the numerical control device 3.

When the numerical controller 3 detects a state such as “overload”, “excessive speed deviation”, or “overheat” of the servo motor, the spindle motor, or the like, based on the information output from the detection unit 22 of the machine tool 2, An abnormality notification called an alarm may be output to the user to stop the machine tool 2. The drive unit 21 stops when the numerical control device 3 outputs an abnormality notification. The drive unit 21 outputs information indicating the presence / absence of an abnormality notification to the machine learning device 1.

The numerical control device 3 receives, as processing conditions, information input by the user using the HMI 4 and information detected by the detection unit 22. The numerical control device 3 numerically controls the machine tool 2 according to a numerical control program based on the received machining conditions. Here, the information input from the HMI 4 includes, for example, information indicating the property of the workpiece and information indicating the operating state of the machine tool 2. The information indicating the property of the processing target is, for example, the material of the processing target. The information indicating the operating state of the machine tool 2 is, for example, the depth of cut, which is the information indicating the movement amount of the tool, the feed speed of the servo motor, and the rotation speed of the spindle motor.

Even if the processing conditions such as the material of the object to be machined and the depth of cut are the same, the load on the servo motor and spindle motor may increase due to the wear state of the tool, and an error notification may be output. The numerical control device 3 has tool life information indicating the type of tool mounted on the machine tool 2 and the life of each tool. The tool life information is information indicating the tool life such as the number of times the tool is used and the usage time. In the present embodiment, the machine learning device 1 performs machine learning based on the tool life information. Therefore, the numerical control device 3 outputs tool life information and machining conditions to the machine learning device 1. Note that here, the numerical control device 3 uses the information input using the HMI 4 and the information output by the detection unit 22 as the processing conditions, but the numerical control device 3 analyzes the analysis results such as the control program and parameters. It may be processing conditions. In other words, the numerical controller 3 has information input using an input device such as the HMI 4, information detected by the detection unit 22 provided in the machine tool 2, and a control program for controlling the machine tool 2. The processing condition including at least one of the information acquired from and is output.

The machine learning device 1 follows a data set created based on a combination of a state observation unit 11 that observes tool life information and machining conditions output by the numerical control device 3 as state variables, and a combination of the state variables and whether or not there is an abnormality notification. And a learning unit 12 that learns the abnormality occurrence condition of the machine tool 2. Here, the data set is data in which the state variable and the presence/absence of the abnormality notification are associated with each other.

ㆍThe elements to observe as state variables are the elements related to the occurrence of anomaly notification. For example, in order to prevent the motor from overheating, an abnormality notification is output when the temperature of the motor exceeds a preset threshold value. Further, when the object to be processed is a hard material, the load on the motor increases. Even when the depth of cut is large or the feed speed of the servo motor or the rotation speed of the spindle motor is high, the load on the motor increases. Further, when the tool mounted on the machine tool 2 is in a worn state, the load on the motor may increase even under the same machining conditions, and an abnormality notification may occur. For this reason, it is important to observe the wear state of the tool. In the present embodiment, the wear state of the tool is determined using the tool life information. It can be inferred that a tool with a short tool life has a large amount of wear.

The machine learning device 1 is a device separate from the numerical control device 3 in FIG. 1, but the numerical control device 3 may include the machine learning device 1 therein. Further, the machine learning device 1 may exist on the cloud server.

The learning unit 12 learns the abnormality occurrence condition of the machine tool 2 by so-called supervised learning according to, for example, a neural network model. Here, supervised learning refers to a model in which a large set of certain input and result data sets are given to the machine learning device 1 to learn the characteristics of those data sets and to estimate the result from the input.

A neural network is composed of an input layer composed of multiple neurons, an intermediate layer composed of multiple neurons, and an output layer composed of multiple neurons. The intermediate layer is also called a hidden layer, and may be one layer or two or more layers.

FIG. 2 is a diagram showing a three-layer neural network model that can be used by the learning unit 12 shown in FIG. In this model, when a plurality of inputs are input to the input layers X1-X8, the values are multiplied by the weight W1 (w11-w116) and input to the intermediate layers Y1-Y2. The values of the intermediate layers Y1-Y2 are multiplied by the weight W2 (w21-w22) and output from the output layer Z1. This output result changes depending on the values of the weights W1 and W2.

In the present embodiment, the neural network uses the so-called supervised learning to detect abnormalities in accordance with a data set created based on a combination of machining conditions and tool life information observed by the state observation unit 11 and presence/absence of abnormality notification. Learn the occurrence conditions.

In other words, the neural network performs learning by adjusting the weights W1 and W2 so that the processing conditions and tool life information are input to the input layer and the result output from the output layer approaches whether or not there is an abnormality notification.

In addition, the learning unit 12 can also learn the abnormal occurrence condition by so-called unsupervised learning. The unsupervised learning is to give only a large amount of input data to the machine learning device 1 to learn what distribution the input data has, and to give the corresponding input data without learning the corresponding teacher output data. It is a method of learning a device that performs compression, classification, shaping, and so on. In unsupervised learning, the features in a dataset can be clustered into similar ones. The learning unit 12 can realize the prediction of the output by using the clustering result and assigning the output so as to optimize the clustering result by setting some reference. There is also a method called semi-supervised learning as an intermediate problem setting between supervised learning and unsupervised learning. In the semi-supervised learning, there is only a part of input and output data sets, and in other cases, learning is performed using only input data.

The learning unit 12 can use deep learning for learning the extraction of the feature amount itself, or can use other known methods such as genetic programming, functional logic programming, and support vector machine.

Note that in the example shown in FIG. 1, the machine learning device 1 acquires information such as machining conditions and tool life information from the single numerical control device 3, but the present embodiment is not limited to this example. The state observation unit 11 may observe the information acquired from the plurality of numerical control devices 3 as state variables, and the learning unit 12 may learn the abnormality occurrence condition based on the data sets for the plurality of numerical control devices 3. .. Here, the learning unit 12 may acquire a data set from a plurality of numerical control devices 3 used at the same site, or may acquire a data set from a plurality of numerical control devices 3 operating independently at different sites. You may. Further, the numerical control device 3 for which the data set is to be acquired can be added in the middle or can be removed from the target in the middle. Further, with respect to the machine tool 2 to be controlled by a certain numerical control device 3, the machine learning device 1 that has learned the abnormality occurrence condition is attached to a different numerical control device 3, and the machine tool 2 to be controlled by this numerical control device 3 The learning result can be updated by re-learning the abnormal condition.

As described above, according to the machine learning system 100 according to the first embodiment of the present invention, in addition to the machining conditions of the machine tool 2, the machine learning device 1 uses the tool life information indicating the life of the tool as a state variable. Then, the abnormality occurrence condition of the machine tool 2 is learned according to the data set created based on the state variable and the presence or absence of the abnormality notification. By adopting such a configuration, it becomes possible to improve the learning accuracy of the abnormality occurrence condition as compared with the case where the tool life information is not used.

Embodiment 2.
FIG. 3 is a diagram showing a configuration of a control system 200 of the machine tool 2 according to the second exemplary embodiment of the present invention. The control system 200 has a function of controlling the machine tool 2 by using the learning result of the machine learning system 100 described in the first embodiment. Specifically, in addition to the configuration of the machine learning system 100, the control system 200 determines that the machine tool 2 has an abnormality based on the learning result of the machine learning device 1 and the processing conditions of the processing process to be executed. It has an abnormality estimation device 5 for estimating whether or not it occurs. Hereinafter, the description of the same configuration as that of the first embodiment will be omitted, and the portions different from the first embodiment will be mainly described.

The abnormality estimating device 5 presupposes the tool life indicated by the input tool life information based on the learning result of the machine learning device 1 before actually operating the machine tool 2, and performs the machining under the input machining conditions. It is estimated whether or not an abnormality occurs when the machine 2 is operated. The abnormality estimation device 5 outputs the estimation result to the HMI 4 and notifies the user.

FIG. 4 is a diagram showing a functional configuration of the abnormality estimation device 5 shown in FIG. The abnormality estimation device 5 includes a learning result acquisition unit 51, an estimation condition acquisition unit 52, an estimation unit 53, and a notification unit 54. The learning result acquisition unit 51 acquires the learning result of the abnormality occurrence condition of the machine tool 2 from the learning unit 12 of the machine learning device 1. For example, the learning result acquisition unit 51 may acquire the learning result every time the learning result of the learning unit 12 is updated, or may periodically acquire the learning result. The learning result acquisition unit 51 outputs the acquired learning result to the estimation unit 53.

The speculative condition acquisition unit 52 acquires, from the numerical control device 3, speculative conditions regarding whether or not an abnormality has occurred in the machine tool 2, including machining conditions of the machine tool 2 and tool life information. The estimation condition acquisition unit 52 outputs the acquired estimation conditions to the estimation unit 53.

The estimating unit 53 estimates whether or not an abnormality has occurred in the machine tool 2 based on the learning result of the abnormality occurrence condition and the estimation condition. The estimation unit 53 outputs the estimation result to the notification unit 54. The notification unit 54 notifies the user of the estimation result. Specifically, the notification unit 54 can notify the user of the estimation result by outputting the estimation result to the HMI 4 and displaying the estimation result on the HMI 4.

FIG. 5 is a diagram showing an example of the display screen 60 output by the HMI 4 shown in FIG. The display screen 60 shows whether or not an abnormality occurs in the machine tool 2 when the machining condition input area 61 that receives an input of the machining condition of the machine tool 2 and the machining condition displayed in the machining condition input area 61 are used. An estimation result display area 62 for displaying an estimation result indicating the, and a machining start button 63 which is an operation unit for instructing the start of machining using the machining condition displayed in the machining condition input area 61. By using the display screen 60 displayed by the HMI 4, the user inputs the processing conditions, and when the input processing conditions are used, the estimated result of whether or not an abnormality occurs in the machine tool 2 is actually machined. It becomes possible to know without moving the machine 2. Then, the user can start the actual processing by using the processing condition that the abnormality is estimated to cause no abnormality.

FIG. 6 is a flowchart for explaining the operation of the control system 200 shown in FIG. The user of the control system 200 inputs the processing conditions of the machine tool 2 using the display screen 60 of the HMI 4 (step S101). When the processing condition is input to the processing condition input area 61 of the display screen 60, the numerical control device 3 inputs the processing condition and tool life information to the abnormality estimating device 5 (step S102).

The estimating unit 53 of the abnormality estimating device 5 estimates whether or not the abnormality of the machine tool 2 is notified based on the input machining conditions and tool life information (step S103). It is assumed that the abnormality estimation device 5 has acquired the learning result from the machine learning device 1 in advance and performs the estimation process of step S103 based on the acquired learning result. The estimation unit 53 outputs the estimation result to the notification unit 54.

When the notification unit 54 of the abnormality estimation device 5 outputs the estimation result to the HMI 4, the HMI 4 notifies the user by displaying the estimation result (step S104). The numerical control device 3 determines whether or not the machining conditions have been changed (step S105). For example, when notifying the user of the estimation result by displaying the estimation result in the estimation result display area 62 of the display screen 60 shown in FIG. 5, the user changes the processing condition displayed in the processing condition input area 61. The processing condition can be changed by inputting Alternatively, the user can instruct the start of processing by operating the processing start button 63 on the display screen 60.

If the processing conditions are changed (step S105: Yes), the process is repeated from step S101. When the machining conditions are not changed (step S105: No), that is, when the machining start button 63 is operated in the example of the display screen 60, the numerical control device 3 starts the numerical control, so that the machine tool 2 operates as follows. The processing is executed (step S106).

When the machining process is executed, the numerical control device 3 and the drive unit 21 of the machine tool 2 input the machining result as training data to the learning unit 12 (step S107). Specifically, as described in the first embodiment, the machining conditions when the machine tool 2 is actually operated, the tool life information, and the presence/absence of the abnormality notification are input to the machine learning device 1.

As described above, according to the control system 200 according to the second embodiment of the present invention, the machining conditions of the machine tool 2 are appropriate in advance based on the learning result of the machine learning device 1, that is, the machine tool. It is possible to actually drive the machine tool 2 after confirming that the machining conditions are such that no abnormality occurs in No. 2. Here, since the machine learning device 1 learns the abnormality occurrence condition based on the tool life information, it is possible to accurately estimate whether or not an abnormality occurs in the machine tool 2 by using a highly accurate learning result. It becomes possible to do.

In this way, by notifying the user of the estimated result of whether or not there is an abnormality notification before executing the processing, it becomes easy to specify the optimum processing conditions. In addition, it is possible to shorten the time until the processing conditions actually used are determined.

Next, a hardware configuration for realizing the functions of the machine learning device 1, the numerical control device 3, and the abnormality estimation device 5 according to the first and second embodiments of the present invention will be described. The functions of the machine learning device 1, the numerical control device 3, and the abnormality estimation device 5 are realized by processing circuits. These processing circuits may be realized by dedicated hardware, or may be control circuits using a CPU (Central Processing Unit).

When the above processing circuits are realized by dedicated hardware, these are realized by the processing circuit 90 shown in FIG. 7. FIG. 7 is a diagram showing dedicated hardware for realizing the functions of the machine learning device 1, the numerical control device 3, and the abnormality estimation device 5 according to the first and second embodiments of the present invention. The processing circuit 90 is a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a combination thereof.

When the above processing circuit is realized by a control circuit using a CPU, this control circuit is, for example, the control circuit 91 having the configuration shown in FIG. FIG. 8 is a diagram showing a configuration of a control circuit 91 for realizing the functions of the machine learning device 1, the numerical control device 3, and the abnormality estimating device 5 according to the first and second embodiments of the present invention. As shown in FIG. 8, the control circuit 91 includes a processor 92 and a memory 93. The processor 92 is a CPU and is also called a central processing unit, a processing unit, an arithmetic unit, a microprocessor, a microcomputer, a DSP (Digital Signal Processor), or the like. The memory 93 is, for example, a nonvolatile or volatile semiconductor memory such as a RAM (Random Access Memory), a ROM (Read Only Memory), a flash memory, an EPROM (Erasable Programmable ROM), and an EEPROM (registered trademark) (Electrically EPROM), Magnetic disks, flexible disks, optical disks, compact disks, mini disks, DVDs (Digital Versatile Disk), etc.

When the above processing circuit is realized by the control circuit 91, it is realized by the processor 92 reading and executing a program stored in the memory 93 and corresponding to the processing of each component. The memory 93 is also used as a temporary memory in each process executed by the processor 92.

The configurations described in the above embodiments are examples of the content of the present invention, and can be combined with another known technique, and the configurations of the configurations are not deviated from the scope not departing from the gist of the present invention. It is also possible to omit or change the part.

For example, the machine learning device 1 according to the first and second embodiments may be built in the numerical control device 3 or may exist on the cloud server. Further, although the abnormality estimation device 5 of the second embodiment is different from the numerical control device 3 and the machine learning device 1 in the above, it is the same as at least one of the numerical control device 3 and the machine learning device 1. May be realized on the above device. Further, the abnormality estimation device 5 may exist on the cloud server.

1 machine learning device, 2 machine tools, 3 numerical control device, 4 HMI, 5 abnormality estimation device, 11 state observation unit, 12 learning unit, 21 driving unit, 22 detection unit, 51 learning result acquisition unit, 52 inference condition acquisition unit , 53 estimation section, 54 notification section, 60 display screen, 61 processing condition input area, 62 estimation result display area, 63 processing start button, 90 processing circuit, 91 control circuit, 92 processor, 93 memory, 100 machine learning system, 200 Control system.

Claims

It is a machine learning device that learns the abnormal occurrence conditions of machine tools that process objects to be machined.
A state observer that observes tool life information indicating the life of a tool and machining conditions of the machine tool as state variables.
A learning unit that learns the abnormality occurrence condition according to a data set created based on the state variable and the presence/absence of notification of abnormality of the machine tool,
A machine learning device comprising:
The machining condition includes at least one of information indicating a property of the machining object, information indicating an operation state of the machine tool, and information indicating a type of the tool. The machine learning device described in.
The information indicating the operation state of the machine tool includes at least one of information indicating the amount of movement of the tool, a feed speed of a servo motor provided in the machine tool, and a temperature of the machine tool. The machine learning device according to claim 2.
The machine learning device according to any one of claims 1 to 3, wherein the state observing unit acquires the tool life information from a numerical control device that numerically controls the machine tool.
The machine learning device according to any one of claims 1 to 4, wherein the tool life information is calculated from the number of times of use and the time of use of the tool.
The processing condition is at least one of information input using an input device, information detected by a detection unit provided in the machine tool, and information acquired from a control program for controlling the machine tool. 6. The machine learning device according to claim 1, wherein the machine learning device includes one.
An estimation unit that estimates whether or not an abnormality of the machine tool will occur, based on an estimation condition that includes the machining condition and the tool life information, and a learning result of the abnormality occurrence condition,
A notification unit for notifying the estimation result of the estimation unit,
The machine learning apparatus according to any one of claims 1 to 6, further comprising.
A numerical control device comprising the machine learning device according to any one of claims 1 to 7.
A learning result acquisition unit that acquires the learning result of the abnormality occurrence condition from the machine learning device according to any one of claims 1 to 7,
An estimation condition acquisition unit that acquires estimation conditions including the machining conditions and tool life information,
An estimation unit that estimates whether or not an abnormality occurs in the machine tool based on the estimation conditions and the learning result.
A notification unit that notifies the estimation result of the estimation unit,
An anomaly estimation device comprising:
The machine learning device according to any one of claims 1 to 7.
An abnormality estimation device according to claim 9;
An output device for outputting a display screen including a processing condition input area for receiving the input of the processing condition, an estimation result display area for displaying an estimation result of the abnormality estimation device, and an operation unit for instructing the start of processing. When,
A control system for machine tools, comprising: