JP2024017966A - Machine tool monitoring device using sensor and system - Google Patents

Abstract

To provide a technique capable of immediately starting the operation of a detection device which can estimate the state of a tool.SOLUTION: There is provided a detecting device for estimating a state of a tool in a machine tool facility which is attached with the tool being a replaceable consumable supply and which performs steps. The detecting device comprises: initial condition setting means which sets a use period of the tool on the basis of a material to be machined by the machine tool facility and usage of the tool; selection means which selects a machine learning model corresponding to the tool on the basis of the output from the initial condition setting means; means which deploys the selected machine learning model; a sensor which is configured to detect a signal generated from the step performed by the tool; and data analysis and input means which is configured to analyze time-series data obtained from the sensor as a combination of plural types of parameters, divide the time-series data into each use period of the tool, and input the divided time-series data to the machine learning model selected and deployed so as to correspond to the tool to make the machine learning model learn the data.SELECTED DRAWING: Figure 1

Description

The present invention relates to a machine tool monitoring device and system using sensors.

Many machine tools (machine tool equipment), such as machining centers and NC lathes, are equipped with consumable tools. The timing of replacing consumables is important, and if replacement is delayed, the accuracy of the products to be processed will be adversely affected. However, if you replace it too soon, the cost of expensive tools will increase unnecessarily. There are limits to determining when it is time to replace a product that relies on human intuition, and as the labor shortage worsens, there is a need for technical solutions that are not individualistic.

For example, Patent Document 1 discloses a non-contact type that is located at a predetermined distance from a workpiece to be plasticized and detects acoustic emissions that are generated when the workpiece is processed by a plastic processing device and propagated in the atmosphere. an acoustic emission sensor, an analysis section that performs time-frequency analysis of the detection signal of the non-contact acoustic emission sensor, and a frequency component that is equal to or higher than a predetermined threshold in a predetermined frequency band based on the analysis result of the analysis section. A plastic working abnormality detection device is disclosed that includes a diagnostic unit that detects the occurrence of a processing abnormality in the case of a plastic working abnormality, and it is described that a signal acquired from a sensor is analyzed as a time-frequency map.

However, the prior art as described above has the following problems. Generally, machine tools are influenced by the environment where they are installed (climate, atmospheric pressure, sea level, nature of the ground, etc.). This is especially noticeable in machine tools that perform precision work such as micromachining. For this reason, it is known in the field that even machine tools of the same model number actually have different characteristics. The technology in Patent Document 1 states that it is desirable to use different frequency bands depending on the phenomenon to be detected, but this is just another way of saying that the analysis must be re-performed for each individual machine tool. do not have. In other words, the people in the field must individually and specifically optimize the analysis, and in the end it is undeniable that this is a method that depends on the individual.

Additionally, methods have been proposed for detecting abnormalities in machine tools using machine learning. For example, Patent Document 2 discloses an estimated load utilization method using a sensor and an information processing device, which estimates fluctuations in load applied to a tool of a machine tool, and utilizes information on the estimated load. an information processing device converting data obtained from the sensor that measures the variation in the load applied to the tool attached to the device that drives the tool into feature quantities; Estimating a component due to the variation in the load and a component due to disturbance due to the feature quantity by machine learning, and estimating the transition of a probability distribution or probability density function that the variation in the feature quantity for each process follows. a step of calculating a probability of a value that the feature amount can take based on the result of the estimated probability distribution or the transition of the probability density function; The method is characterized in that it includes a step of detecting an abnormality in the feature amount by comparing the probability that the feature value can be regarded as a normal value.

However, Patent Document 2 does not suggest how to prepare and train a machine learning model. This is evidence that Patent Document 2 only assumes that the same machine learning model always functions. In the field, as mentioned above, each machine tool has its own unique characteristics, so such a convenient assumption cannot be said to be realistic. In other words, the technology of Patent Document 2 also has the problem of requiring individualized work to maintain the machine learning model on site.

Japanese Patent Application Publication No. 2022-052946 Patent No. 6952318

In view of the above-mentioned problems with the conventional technology, it is possible to estimate the condition of tools at actual sites where various machine tools with replaceable consumables are used, without placing an excessive burden on individual personnel at the site. There is a need for new technology that allows detection equipment to be put into operation immediately.

That is, the present invention can provide the following aspects.

Aspect 1.
A detection device for estimating the state of a tool in machine tool equipment that is equipped with a tool that is a replaceable consumable item and performs a process,
initial condition setting means for setting a usage period of the tool based on the material to be machined with the machine tool equipment and the use of the tool;
Selection means for selecting a machine learning model corresponding to the tool based on the output from the initial condition setting means;
a means of deploying the selected machine learning model; and
a sensor configured to detect a signal generated from a process performed by the tool;
The time-series data obtained from the sensor is analyzed as a combination of multiple types of parameters, divided by the period of use of the tool, and the divided time-series data is selected and deployed to correspond to the tool. A detection device comprising: a data analysis/input means configured to input data to a machine learning model for learning.

Aspect 2.
There is a plurality of the tools, and the machine tool equipment is configured to continuously perform a plurality of types of processes,
The initial condition setting means is configured to set the order and period of use of each tool based on the material to be machined with the machine tool equipment and the use of the tool,
The selection means is configured to select a machine learning model corresponding to each of the plurality of tools based on the output from the initial condition setting means,
The data analysis/input means analyzes the time series data obtained from the sensor as a combination of multiple types of parameters, divides it according to the usage period of each tool, and applies the divided time series data to each tool. The detection device according to aspect 1, wherein the detection device is configured to input and learn a machine learning model selected and deployed to perform learning.

Aspect 3.
The sensor is
an acoustic emission sensor configured to detect elastic waves generated from a process performed in the machine tool equipment;
an image sensor configured to visually detect the state of a tool mounted on the machine tool equipment;
a vibration sensor configured to detect vibrations generated from a process performed on the machine tool equipment; and a temperature sensor configured to detect a temperature generated from a process performed on the machine tool equipment. 3. The detection device according to aspect 1 or 2, characterized in that the detection device includes one or more of:

Aspect 4.
4. The detection device according to aspect 3, wherein the sensors are a combination of two or more types.

Aspect 5.
The detection device further includes:
Contains an interface for connecting to a network,
The selection means is configured to select a machine learning model stored in an external server via the interface,
5. The detection device according to aspects 1 to 4, wherein the data analysis/input means is configured to deploy the machine learning model selected by the selection means via the interface.

Aspect 6.
The detection device according to aspect 5, configured to upload the learning result dataset learned by the data analysis/input means to an external server via the interface.

Aspect 7.
7. The detection device according to aspects 1 to 6, wherein the machine tool equipment is a machining center, an injection molding machine, a press machine, an NC lathe, or a polishing machine.

Aspect 8.
A server configured to connect to a network,
a storage means for storing multiple types of machine learning models;
including a communication means for connecting with an external detection device via a network,
Via the communication means,
Receiving a signal for selecting a machine learning model corresponding to a tool for which a period of use is set based on the material to be machined by external machine tool equipment and the use of the tool;
A server configured to: transmit a selected machine learning model over a network.

Aspect 9.
Further, via the communication means,
Time-series data obtained from a sensor configured to detect signals generated from the process performed by a tool is divided by the period of use of the tool, analyzed as a combination of multiple types of parameters, and analyzed to correspond to the tool. Receive a learning result dataset obtained by inputting and learning a machine learning model selected and deployed in
9. The server according to aspect 8, wherein the server is configured to store the received learning result data set in the storage means.

Aspect 10.
The detection device according to aspects 1 to 7,
The server according to aspects 8 to 9,
A system comprising a network configured to connect the detection device and the server.

Aspect 11.
A program including instructions executable by a computer having a processor, which is connected to machine tool equipment that carries out a process and is equipped with a tool that is a replaceable consumable item,
The instruction is executed by the processor,
using the processor to set a usage period of the tool based on the material to be machined with the machine tool equipment and the use of the tool;
Selecting a machine learning model corresponding to the tool based on the output from the initial condition setting means using the processor;
deploying a selected machine learning model using the processor;
using a sensor to detect a signal generated from a process performed by the tool;
Using the processor, the time-series data obtained from the sensor is analyzed as a combination of multiple types of parameters, divided by the period of use of the tool, and the divided time-series data is divided into pieces corresponding to the tool. A program configured to input and train a machine learning model selected and deployed in a machine learning model.

The present invention makes it possible to immediately operate a detection device that can estimate the condition of a tool at an actual site where various machine tools with replaceable consumables are used, without placing an excessive burden on the individual at the site. This has the remarkable effect that it can be started immediately.

1 shows a configuration example of a system according to an embodiment of the present invention. 3 shows a configuration example of a system according to another embodiment of the present invention. An example of time series data obtained by a system according to an embodiment of the present invention is shown. Another example of time series data obtained by a system according to an embodiment of the present invention is shown. An example of a correlation curve obtained using the time series data of FIG. 4 is shown. Another example of a correlation curve obtained using the time series data of FIG. 4 is shown. An example of time series data obtained from a machine tool that uses multiple types of tools is shown. An example of time series data obtained from an AE sensor is shown.

An embodiment of the present invention provides a detection device for inferring the condition of a tool that can be used in machine tool equipment that performs a process with a tool that is a replaceable consumable item. The type of machine tool equipment does not particularly matter as long as it is a type that can be equipped with tools, but examples include machining centers, injection molding machines, press machines, NC lathes, and polishing machines.

FIG. 1 is a block diagram showing the configuration of a detection device 100 as an example of the detection device. The detection device 100 in FIG. 1 mainly includes an initial condition setting means 110 for setting the usage period of a tool, a selection means 120 for selecting a machine learning model corresponding to the tool, and a deployment means for deploying the selected machine learning model. 130, a sensor 140 configured to detect signals generated from operations performed by the tool, and data analysis configured to input and train a machine learning model selected and deployed in response to the tool.・Input means 150 are included. The detection device 100 is configured to be connectable to an external machine tool 97 and an external server 99 via a network interface 160 (via a wireless or wired network). The detection device 100 may be a terminal having arithmetic means (processor, microprocessor, microcontroller, etc.) having arithmetic functions, and may be a single terminal or a system consisting of multiple terminals. You can. In a preferred embodiment, the detection device 100 may be an edge terminal (one or more) that can be connected to a cloud server. Such edge terminals may be, for example, single-board computers, stick PCs, smartphones, tablet PCs, etc. that can be used for IoT applications.

The initial condition setting means 110 receives data regarding the material to be machined and the use of the tool in the machine tool 97, and determines the usage period of the tool (i.e., the time interval in which the tool is used in a certain process) based on the received data. , time cycle).

For example, if the machine tool is a single-shot press machine, it can be set as the length of time for which the press part, which is a tool, applies pressure. Further, when the machine tool is a machining center, the order and period of using each of multiple types of tools (drill, tap, milling cutter, end mill, reamer, side cutter, etc.) can be determined as a sequence. Further, if the machine tool is an injection molding machine, initial conditions can be set in the same manner as described above assuming multiple types of tools such as die plates, tie bars, ejector pins, and molds.

In acquiring the data regarding the material to be machined and the use of the tool, for example, it may be recognized from image data taken of the inside of the machine tool 97 by an imaging means, or it may be acquired from the operating condition setting data of the machine tool. You can do it like this.

The initial conditions thus determined are fed back to the PLC (control unit) 98 of the machine tool 97, and the operation of the machine tool 97 can be controlled.

The selection means 120 has a function of selecting a machine learning model corresponding to the tool. The selection means 120 can be connected to a network (LAN, WAN, Internet, etc.) via the network interface 160, access the external server 99, and select among multiple types of machine learning models stored in the server 99. conduct. The selection can be made based on the output from the initial condition setting means 110 described above.

In some embodiments, for example, the output from the initial condition setting means 110 may include a photographed image of the tool, and the selection means 120 may analyze the image to identify or infer the type of tool. Alternatively, the image may be sent to the server 99 and applied to the image recognition function of the server 99.

In another embodiment, the output from the initial condition setting means 110 includes data indicating the type of tool installed in the machine tool 97 (for example, data indicating the model, shape, source of the tool, etc.). Good too. The data may be interpreted by the selection means 120 to identify or estimate the type of tool, or the data may be sent to the server 99 to identify or estimate the type of tool.

The multiple types of machine learning models that the server 99 stores are not particularly limited. The deploying means 130 deploys the machine learning model selected from among these by the selecting means 120 as being appropriate for the initial conditions of the machine tool 97. Note that in this specification, "deploy" refers to introducing a machine learning model from outside the detection device 100 (for example, via a network), and (storing it in a storage (not shown)) to the machine using the processor of the detection device 100. It refers to making a learning model functional. In addition, in the figure, for clarity, the deployed machine learning model is depicted in the block representing the deployment means 130. The reason for adopting such a configuration is that it is not necessary to store a large number of machine learning models in the storage of the detection device 100. These effects are particularly useful when the detection device 100 is an edge terminal from the perspective of saving limited storage capacity.

The sensor 140 then detects a signal generated from a process performed by a tool attached to the machine tool 97. Examples of the sensor 140 include an acoustic emission sensor (AE sensor) configured to detect elastic waves generated from a process, and an image sensor (including an image sensor) configured to visually detect the state of a tool to be mounted. ), a vibration sensor configured to detect vibrations generated from the process, a temperature sensor configured to detect temperature generated from the process, and the like. In the example shown in FIG. 1, the sensor 140 is a single type of sensor (there may be multiple sensors of the same type). In another embodiment, sensor 140 may be a combination of multiple types of sensors, as shown in FIG. 2 (described in detail below).

The signal detected by the sensor 140 can be treated as time series data (data that plots the correlation between signal strength and time) in association with the time at which the signal was detected (which can be determined by a clocking means (not shown)). can. The data analysis/input means 150 then analyzes the time series data obtained from the sensor 140 as a combination of multiple types of parameters (for example, in this example, signal strength and time can be set as parameters). For analysis, the time series data can be divided by the period of use of the tool, and the divided time series data can be input to the above-described deployed machine learning model for learning.

Figure 3 shows a time plot of the signal strength (denoted as Parameter 1) obtained using a sensor during one cycle of a tool's use (in this example, from 0 to 97 seconds). This is an example. Machine learning using signal strength and time as feature quantities can be performed by dividing the original time series data, processing it to cut out each time unit, and inputting it to a machine learning model for learning. can be done.

In a preferred embodiment, the data analysis/input means 150 may further upload the training result dataset to an external server via the network interface 160. With such a configuration, it is possible to feed back the learning results in the detection device 100 to the server and optimize the machine learning model group stored in the server. In another embodiment, the detection device 100 may not perform such uploads, but may specialize only in deploying from external servers.

In preferred embodiments, a combination of multiple types of sensors may be used. FIG. 2 is a block diagram showing the configuration of another detection device 200 having three types of sensors 240, 242, 244. The detection device 200 is the same as the detection device 100 except for the number of types of sensors. In the detection device 200, three types of outputs (time plots of Parameters 1 to 3) can be obtained from each sensor as shown in Figure 4. The data analysis/input means 250 can then obtain a correlation curve (drawn as a straight line in the example shown) between each parameter, as shown in FIG. More preferably, it is desirable to remove outliers and obtain a correlation curve as shown in FIG. Note that in another embodiment, it is also possible to acquire multiple types of parameters from one type of sensor and obtain an output as shown in FIG. 4.

By inputting the data related to the correlation curve obtained as shown in Figure 5 or Figure 6 (for example, the slope of the correlation curve, the differential value, etc.) as a feature quantity into the deployed machine learning model and making it learn, the machine can become even more accurate. The effect of enabling learning can be obtained.

A detection device according to an embodiment can be suitably used for a machine tool configured to be equipped with a plurality of types of tools and to perform a plurality of types of processes continuously. When using multiple types of tools, the cycles used for each tool are different, as shown in Figure 7, so it has traditionally been difficult to accurately estimate the condition of each tool individually using simple machine learning. Met. Therefore, in the detection device 100, the initial condition setting means 110 sets the order and period of use of each tool based on the material to be machined by the machine tool 97 and the purpose of the tool. Based on the output from the initial condition setting means 110, the selection means 120 can select a machine learning model corresponding to each of the plurality of tools. The data analysis/input means 150 analyzes the time-series data obtained from the sensor 140 as a combination of multiple types of parameters, divides it by the usage period of each tool, and divides the divided time-series data into correspondences for each tool. It is possible to input and train a machine learning model that has been selected and deployed to do so. It will be obvious that such a configuration can be similarly implemented in the detection device 200 of FIG.

In a preferred embodiment, the present detection device has an acoustic emission sensor (AE sensor) as a sensor, so that multiple types of parameters can be obtained from the elastic wave, such as maximum amplitude value (amplitude), effective value (threshold value, Parameters such as RMS), energy (integral value, energy), number of hits, rise time, number of counts, and waveform duration can be obtained. For example, time-series data regarding combinations of three types of parameters: maximum amplitude value, effective value, and energy can be obtained. FIG. 8 shows an example of time-series data obtained from the AE sensor (an example obtained for three types of parameters). For example, the data analysis/input means included in the detection device may take a correlation between these parameters.

In a preferred embodiment, the present detection device includes an image sensor (such as a camera) as a sensor, thereby making it easier to estimate the condition of a tool that is particularly fragile and whose shape is likely to deteriorate over time. More preferably, the present detection device includes a combination of an image sensor and an AE sensor, so that the state of each tool can be estimated with higher accuracy. For example, since this detection device includes a combination of two types, an AE sensor and a vibration sensor, it is possible to predict the occurrence of a primary phenomenon in a tool (such as a break inside the tool that cannot be directly observed from the outside), while also detecting the occurrence of a vibration sensor. By detecting the occurrence of vibration in the tool, it is possible to accurately estimate that the tool is nearing the end of its life.

In some embodiments of the invention, a network connectable server may also be provided. The server includes a storage means (which may be any storage) that stores multiple types of machine learning models, and an external detection device (which may be the detection device 100 or detection device 200 described above, for example) via the network. and a communication means for connecting with. Through the communication means, the server selects a machine learning model corresponding to the tool for which the usage period has been set based on the material to be machined by external machine tool equipment and the purpose of the tool. Can receive signals. The server can then transmit the selected machine learning model over the network via the communication means.

Preferably, the server further includes, via the communication means, a sensor configured to detect signals generated from the process performed by the tool, which may for example be sensor 140 or sensors 240, 242, 244 as described above. This data was obtained by dividing the time-series data by tool use period, analyzing it as a combination of multiple types of parameters, and inputting and learning it into a machine learning model that was selected and deployed to correspond to the tool. A learning result dataset can be received. The server may then store the received learning result data set in the storage means.

In some embodiments of the invention, a system may be provided that includes a detection device and a server as described above, and a network configured to connect the detection device and the server.

In some embodiments of the present invention, a program may be provided that includes instructions executable by a computer having a processor that connects to machine tool equipment that carries a replaceable consumable tool and performs a process. When the instruction is executed by the processor, the processor may perform one or more of the following functions in random order. In other words, the usage period of the tool should be set based on the material to be machined with the machine tool equipment and the purpose of the tool. Selecting a machine learning model corresponding to the tool based on the output from the initial condition setting means. Deploying selected machine learning models. Using sensors to detect signals generated by the process performed by the tool. The time-series data obtained from the sensor is analyzed as a combination of multiple types of parameters, divided by the period of use of the tool, and the divided time-series data is processed into a machine learning model that is selected and deployed to correspond to the tool. input and learn.

In some embodiments, a computer readable medium (SSD, HDD, optical media, etc.) may also be provided that stores the program described above.

Claims (11)

A detection device for estimating the state of a tool in machine tool equipment that is equipped with a tool that is a replaceable consumable item and performs a process,
initial condition setting means for setting a usage period of the tool based on the material to be machined with the machine tool equipment and the use of the tool;
Selection means for selecting a machine learning model corresponding to the tool based on the output from the initial condition setting means;
a means of deploying the selected machine learning model; and
a sensor configured to detect a signal generated from a process performed by the tool;
The time-series data obtained from the sensor is analyzed as a combination of multiple types of parameters, divided by the period of use of the tool, and the divided time-series data is selected and deployed to correspond to the tool. A detection device comprising: a data analysis/input means configured to input data to a machine learning model for learning. There is a plurality of the tools, and the machine tool equipment is configured to continuously perform a plurality of types of processes,
The initial condition setting means is configured to set the order and period of use of each tool based on the material to be machined with the machine tool equipment and the use of the tool,
The selection means is configured to select a machine learning model corresponding to each of the plurality of tools based on the output from the initial condition setting means,
The data analysis/input means analyzes the time series data obtained from the sensor as a combination of multiple types of parameters, divides it according to the usage period of each tool, and applies the divided time series data to each tool. 2. The detection device according to claim 1, wherein the detection device is configured to input and learn a machine learning model selected and deployed to perform learning. The sensor is
an acoustic emission sensor configured to detect elastic waves generated from a process performed in the machine tool equipment;
an image sensor configured to visually detect the state of a tool mounted on the machine tool equipment;
a vibration sensor configured to detect vibrations generated from a process performed on the machine tool equipment; and a temperature sensor configured to detect a temperature generated from a process performed on the machine tool equipment. 3. The detection device according to claim 1, characterized in that the detection device includes one or more types of. 4. The detection device according to claim 3, wherein the sensors are a combination of two or more types. The detection device further includes:
Contains an interface for connecting to a network,
The selection means is configured to select a machine learning model stored in an external server via the interface,
3. The detection device according to claim 1, wherein the data analysis/input means is configured to deploy the machine learning model selected by the selection means via the interface. 6. The detection device according to claim 5, wherein the detection device is configured to upload the learning result data set learned by the data analysis/input means to an external server via the interface. 3. The detection device according to claim 1, wherein the machine tool equipment is a machining center, an injection molding machine, a press machine, an NC lathe, or a polishing machine. A server configured to connect to a network,
a storage means for storing multiple types of machine learning models;
including a communication means for connecting with an external detection device via a network,
Via the communication means,
Receiving a signal for selecting a machine learning model corresponding to a tool for which a period of use is set based on the material to be machined by external machine tool equipment and the use of the tool;
A server configured to: transmit a selected machine learning model over a network. Further, via the communication means,
Time-series data obtained from a sensor configured to detect signals generated from the process performed by a tool is divided by the period of use of the tool, analyzed as a combination of multiple types of parameters, and analyzed to correspond to the tool. Receive a learning result dataset obtained by inputting and learning a machine learning model selected and deployed in
9. The server according to claim 8, wherein the server is configured to store the received learning result data set in the storage means. The detection device according to claim 1,
The server according to claim 8,
A system comprising a network configured to connect the detection device and the server. A program including instructions executable by a computer having a processor, which is connected to machine tool equipment that carries out a process and is equipped with a tool that is a replaceable consumable item,
The instruction is executed by the processor,
using the processor to set a usage period of the tool based on the material to be machined with the machine tool equipment and the use of the tool;
Selecting a machine learning model corresponding to the tool based on the output from the initial condition setting means using the processor;
deploying a selected machine learning model using the processor;
using a sensor to detect a signal generated from a process performed by the tool;
Using the processor, the time-series data obtained from the sensor is analyzed as a combination of multiple types of parameters, divided by the period of use of the tool, and the divided time-series data is divided into pieces corresponding to the tool. A program configured to input and train a machine learning model selected and deployed in a machine learning model.