WO2022168455A1

WO2022168455A1 - Wear assessment device

Info

Publication number: WO2022168455A1
Application number: PCT/JP2021/046024
Authority: WO
Inventors: 淳池田; 知康古田; 健走原島
Original assignee: 株式会社不二越
Priority date: 2021-02-08
Filing date: 2021-12-14
Publication date: 2022-08-11
Also published as: JPWO2022168455A1; CN115210040A

Abstract

[Problem] To provide a wear assessment device capable of performing automatic re-learning of machine learning without needing a specialist. [Solution] A wear assessment device 100 for a tool in a processing machine is characterized by comprising: a feature amount extraction unit 130 for acquiring a feature amount from an output of a sensor; a dynamic learning unit 140 for constructing a learning model on the basis of the feature amount; an inference unit 160 for inferring a degree of wear on the basis of the learning model constructed by the dynamic learning unit; a threshold value storage unit 180 for storing a threshold value of wear; and a wear assessment unit 190 for assessing wear on the basis of the degree of wear and the threshold value. The wear assessment device is also characterized in that: the inference unit determines whether inference based on the current learning model is appropriate, and if determining that the foregoing is not appropriate, then transmits a re-learning instruction to the dynamic learning unit; and the dynamic learning unit, upon receiving control data from the inference unit, performs re-learning on the basis of the threshold value and an additional feature amount different from the feature amount that was used when constructing the current learning model, and updates the learning model.

Description

Wear determination device

The present invention relates to a wear determination device that applies machine learning.

　Conventionally, there have been many proposals for applying machine learning to tool wear determination and tool life prediction. For example, a life prediction device disclosed in Patent Document 1 observes life-related data, creates a probabilistic model of the replacement life of consumable parts, and uses the created probabilistic model to observe the life-related data. predict the replacement life of According to Patent Document 1, it is possible to predict the service life of a consumable part of a manufacturing machine with a predetermined accuracy even when the collected data is small.

JP 2019-207576 A

However, in conventional machine learning, not limited to Patent Document 1, the wear state is once learned under specific conditions, a learning model is constructed, and the degree of wear is inferred under the same conditions based on the learning model. Then, even if the inference can be made with high accuracy when the conditions do not change, when the conditions such as the machining method, tools, and work change, the accuracy drops significantly, and there is a problem that appropriate inference cannot be performed (general compatibility issues). For this reason, even if the machine has been trained once, it must be re-learned each time the conditions change. However, it is extremely difficult to re-learn unless you are a machine learning expert, and there is a problem that it cannot be handled by on-site workers. there were.

Therefore, an object of the present invention is to provide a wear determination device that can automatically perform machine learning relearning without requiring an expert.

A representative configuration of the present invention is a tool wear determination device in a processing machine, which includes a sensor that measures the state of the workpiece or the tool, a feature amount extraction unit that acquires the feature amount from the output of the sensor, and the feature amount a dynamic learning unit that constructs a learning model based on the dynamic learning unit, an inference unit that infers the degree of wear based on the learning model constructed by the dynamic learning unit, a threshold storage unit that stores a wear threshold, and a degree of wear and the threshold The inference unit judges whether the inference based on the current learning model is appropriate or not, and instructs the dynamic learning unit to re-learn if it is determined not to be appropriate. , and when the dynamic learning unit receives the control data from the inference unit, it re-learns based on additional feature values and thresholds that are different from the feature values used to construct the current learning model, and the learning model is updated.

When conditions such as machining methods, tools, and workpieces change, the degree of wear inferred from feature values changes significantly. As a result, the determination by the wear determining unit will also be significantly faster or slower than the actual wear limit. However, by dynamically updating the learning model using feature values at the same time as inference, machine learning can be automatically relearned without the need for relearning by experts (generalizability). improvement).

Another representative configuration of the present invention is a tool wear determination device in a processing machine, which includes a sensor for measuring the state of the workpiece or the tool, a feature amount extraction unit for acquiring a feature amount from the output of the sensor, A dynamic learning unit that builds a learning model based on a feature amount, an inference unit that infers the degree of wear based on the learning model built by the dynamic learning unit, a threshold storage unit that stores a wear threshold, and a work A dynamic threshold setting unit that stores a threshold input by a user in a threshold storage unit, and a wear determination unit that determines wear based on the degree of wear and the threshold, and the inference unit determines whether the threshold has been changed. If it is determined that there has been a change, a re-learning instruction is sent to the dynamic learning unit, and when the dynamic learning unit receives control data from the inference unit, re-learning is performed based on the feature amount and the changed threshold value. and update the learning model.

With conventional machine learning, it was necessary to set the threshold before starting the inference program, and to change the threshold, it was necessary to stop and restart the inference program. However, according to the above configuration, the threshold can be dynamically updated, so that the operator can reset the threshold without stopping the inference program, thereby improving convenience.

It is preferable that the sensor is a vibration sensor that acquires vibration acceleration as the state of the tool or workpiece, and the feature quantity extraction unit acquires the feature quantity using the vibration acceleration. Vibration acceleration is highly dependent on the degree of wear and is robust against changes in other conditions, so it is preferable as a state quantity for obtaining feature quantities.

According to the present invention, it is possible to provide a wear determination device capable of automatically performing machine learning relearning without requiring an expert.

1 is a block diagram of a wear determination device according to an embodiment; FIG. 4 is a flowchart for explaining the operation of an inference program of the wear determination device; It is a figure explaining the example of a feature-value. It is a figure explaining the example of a feature-value.

Preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings. The dimensions, materials, and other specific numerical values shown in these embodiments are merely examples for facilitating understanding of the invention, and do not limit the invention unless otherwise specified. In the present specification and drawings, elements having substantially the same functions and configurations are denoted by the same reference numerals to omit redundant description, and elements that are not directly related to the present invention are omitted from the drawings. do.

FIG. 1 is a block diagram of the wear determination device according to this embodiment, and FIG. 2 is a flow chart explaining the operation of the inference program of the wear determination device.

The wear determination device 100 is a device that determines wear of the tool 12 of the processing machine 10 that processes the workpiece 20 . The processing machine 10 is a processing machine that mainly performs cutting. It is assumed that machining is performed, for example, 100 to 150 times (100 to 150 workpieces are machined) per tool, and the tool is replaced when the wear limit is reached. The flowchart in FIG. 2 is executed for each machining (one workpiece). First, the inference program determines whether machining is in progress (step 300). If it is being processed, the data acquisition unit 120 acquires data (step 302).

The data acquisition unit 120 obtains outputs from the sensor 110 attached to the tool spindle and the sensor 112 attached to the work spindle. The data acquisition unit 120 specifically includes an amplifier and a logger.

The

sensors

110 and 112 in this embodiment are vibration sensors. Other than the vibration sensor, a temperature sensor, a displacement sensor, etc. can be considered. However, the values of temperature sensors and displacement sensors are greatly affected by factors other than wear. On the other hand, the vibration acceleration is highly dependent on the degree of wear and is robust against changes in other conditions, so it is preferable as the state quantity for acquiring the feature quantity. In addition to the acceleration data obtained from the vibration sensor, equivalent can have a function.

The feature quantity extraction unit 130 extracts and accumulates feature quantities from the state quantities acquired by the sensors 110 and 112 (step 304). In this embodiment, the feature amount is acquired using the vibration acceleration. The feature values that can be obtained from the vibration acceleration are the absolute value average, RMS (root mean square), STFT (short-time Fouriertranform), and the fundamental frequency of fluctuating cutting force (cutting force). blade passing frequency) and its harmonics can be used.

The inference program then determines (step 306) whether to perform inference or to perform initial learning (step 308). When no inference is performed, that is when the learning model does not yet exist. Inference is performed when a learning model has already been constructed.

When inference is not performed (when the learning model does not yet exist), the worker 30 judges the wear limit and completes one machining. The inference program then performs learning using the accumulated feature amounts (step 308) to construct the initial learning model.

The dynamic learning unit 140 performs static learning in which a learning model is constructed only from the feature quantity output from the feature quantity extraction unit 130, and dynamic learning in which re-learning is performed according to a re-learning instruction (flag) of the inference unit. You can do both. This dynamic learning is a feature of the present invention. A known algorithm such as a search algorithm or a genetic algorithm can be used as the machine learning algorithm. The dynamic learning unit 140 stores the constructed learning model in the learning storage unit 150 .

When performing inference in step 310, the inference section 160 infers the degree of wear based on the learning model constructed by the dynamic learning section 140 (step 310). That is, inference is performed using the learning model read out from the learning storage unit 150 and the feature amount sequentially acquired from the feature amount extraction unit 130 . The degree of wear is a probabilistic model, and is intermediate data with the meaning of the degree of wear with this probability.

The inference value processing unit 162 performs weighting to prevent erroneous determination when the

sensors

110 and 112 output outliers. Specifically, when the amount of machining after tool replacement is small, the output of the inference section 160 is multiplied by a factor that greatly underestimates it. Conversely, if the amount of machining after the tool is replaced is large, the underestimation coefficient is reduced. This improves the validity of the inference value and prevents erroneous determination.

The wear determining unit 190 determines tool wear using the degree of wear (inference value) obtained via the inference value processing unit 162 and the threshold value obtained from the threshold storage unit 180 (step 318). The display device 200 displays the degree of wear output by the inference unit 160 and the determination result (whether or not the wear limit has been reached) output by the wear determination unit 190 (step 318) (step 320). Here, as the value of the threshold storage unit 180, a set value is inputted in advance before the first learning (step 308) is performed. However, even while the inference program is looping, it can be updated at any time by input from the dynamic threshold setting unit 170 by the operator 30 .

Next, I will explain re-learning. After reasoning in step 310, reasoning unit 160 determines whether the reasoning was appropriate (step 312). Whether the inference is appropriate or not can be determined to be inappropriate, for example, when the inference value calculated by the current learning model loses the correlation with the number of processes. If the inference is not appropriate, the inference unit 160 sends a re-learning instruction 161 (control data) to the dynamic learning unit 140 (step 313).

The inference program of the wear determination device 100 also determines whether the threshold has been changed (step 314). If a new threshold is set in the threshold storage unit 180 by the operator 30 inputting from the dynamic threshold setting unit 170, the inference unit 160 transmits a relearning instruction 161 (control data) to the dynamic learning unit 140 ( step 315).

Also when making an inference (flow from step 310), the inference program determines whether the current tool has reached the wear limit (step 322). When the wear limit is reached, a signal to stop the processing machine is transmitted to the processing machine 10 (step 323), and the process returns to step 300. If the wear limit has not yet been reached, it is determined whether or not a re-learning instruction 161 has been issued (step 324). If the relearning instruction 161 has not been issued, the process returns to step 300 .

If the re-learning instruction 161 has been sent, the dynamic learning unit 140 performs re-learning using the accumulated feature amount (step 326). At this time, the feature amount used for re-learning is a feature amount (hereinafter referred to as "additional feature amount") different from the feature amount used when constructing the current learning model (latest learning model). The additional feature amount may be sent from the inference unit 160 together with the relearning instruction 161 to the dynamic learning unit 140 , or may be sent directly from the feature amount extraction unit 130 to the dynamic learning unit 140 .

For example, suppose you built the first learning model with tool A. In the conventional technique, this learning model is used to make inferences about tools B and C (from step 310), and the learning model is not updated any more. However, in the present invention, when the re-learning instruction 161 is issued, the feature amount (additional feature amount) of tool B and tool C can also be used to update the learning model.

Also, if the number of processes (for example, 120 times) when building the learning model with tool A does not sufficiently reach the wear limit, the learning will be insufficient and correct inference will not be possible. . In such a case, the operator 30 makes a judgment while looking at the trend of the judgment result in the dynamic threshold value setting unit 170, and further adds the number of processing (for example, the feature amount of 121 times to 150 times is the additional feature amount). can be re-learned.

When conditions such as machining methods, tools, and workpieces change, the degree of wear inferred from feature values changes significantly. As a result, the determination by the wear determining unit 190 will also be significantly faster or slower than the actual wear limit. However, as explained above, by dynamically judging whether or not the inference is appropriate using the feature value at the same time as the inference and updating the learning model as appropriate, machine learning can be performed without the need for retraining by experts. Re-learning can be performed automatically (improved generalizability).

Also, in conventional machine learning, it was necessary to set the threshold before starting the inference program, and to change the threshold, it was necessary to stop and restart the inference program. However, according to the above configuration, the threshold can be dynamically updated, so that the operator can reset the threshold without stopping the inference program, thereby improving convenience.

When re-learning (step 326) is completed, immediately return to step 310 and redo the inference. In this case, if the inference continues in step 312, an infinite loop will occur. When the number of re-learning times reaches the upper limit, a message to that effect is displayed to the worker 30 and the process returns to step 300 .

　Figures 3 and 4 are diagrams illustrating examples of feature amounts. As described above, in the present embodiment, the feature quantity is acquired using the vibration acceleration, which is a state quantity that is robust against conditions other than wear.

Fig. 3(a) shows the average absolute value of vibration acceleration with respect to the number of processes as a feature quantity. Alternatively, the sum of the areas created by the amplitudes may be calculated. FIG. 3B shows RMS (Root Mean Square) of the amplitude value of vibration acceleration with respect to the number of processes.

Referring to FIGS. 3(a) and 3(b), it can be seen that the absolute value average or RMS increases almost proportionally as the number of processes increases. Thus, there is a correlation between the number of processes and the average absolute value of the vibration acceleration, or the RMS of the number of processes and the amplitude value of the vibration acceleration. Based on this correlation, it is possible to infer the degree of wear from the average absolute value or RMS for each machining by performing machine learning using the relationship between the average absolute value or RMS and the degree of wear as a feature.

Fig. 4 shows the STFT (short-time Fouriertranform) of the vibration acceleration due to 5-pass machining of a workpiece (gear). No. in FIG. 1 is the data of a new tool, No. 1 in FIG. 18 is the data of the tool at the wear limit.

Referring to FIG. 4, the horizontal axis is time, and it can be seen that the spectrum appears for each machining pass. Note that the fifth pass is a finishing process, which takes a long time. Overall, No. in FIG. 4(b). Since the frequency intensity of 18 is higher, it can be seen that vibration increases as wear progresses. Further, in FIG. 4B where wear has progressed, more spectra appear near 800 Hz in the fifth pass. By performing machine learning using the difference in frequency distribution of such STFTs as a feature, it is possible to determine whether the product is new or the wear limit. As for the difference in frequency distribution, the distribution may be digitized, or the distribution image may be used as a pattern for image processing.

Furthermore, not only a single index but also multiple indices may be combined as a feature quantity. This can further improve the reliability of the inference. Moreover, since any index should have a correlation with the number of processes, it is preferable to always evaluate the correlation between the number of processes and the feature amount, and not to display the determination result when there is no correlation. As a result, it is possible to prevent an erroneous determination from being displayed.

Although the preferred embodiments of the present invention have been described above with reference to the accompanying drawings, it goes without saying that the present invention is not limited to such examples. It is obvious that a person skilled in the art can conceive of various modifications or modifications within the scope described in the claims, and these also belong to the technical scope of the present invention. Understood.

The present invention can be used as a wear determination device that applies machine learning.

DESCRIPTION OF SYMBOLS 10... Processing machine, 12... Tool, 20... Work, 30... Worker, 100... Wear determination apparatus, 110... Sensor, 112... Sensor, 120... Data acquisition part, 130... Feature-value extraction part, 140... Dynamic learning Unit 150 Learning storage unit 160 Inference unit 161 Re-learning instruction 162 Inference value processing unit 170 Dynamic threshold setting unit 180 Threshold storage unit 190 Wear determination unit 200 Display device

Claims

A wear determination device for a tool in a processing machine,
a sensor that measures the state of the workpiece or tool;
A feature quantity extraction unit that acquires a feature quantity from the output of the sensor;
a dynamic learning unit that builds a learning model based on the feature amount;
an inference unit that infers the degree of wear based on the learning model constructed by the dynamic learning unit;
a threshold storage unit that stores a wear threshold;
A wear determination unit that determines wear based on the degree of wear and the threshold,
The inference unit determines whether or not the inference based on the current learning model is appropriate, and if it is determined to be inappropriate, transmits a re-learning instruction to the dynamic learning unit,
When the dynamic learning unit receives the control data from the inference unit, the dynamic learning unit re-learns based on the additional feature amount different from the feature amount used when constructing the current learning model and the threshold, and the learning model A wear determination device characterized by updating the.
A wear determination device for a tool in a processing machine,
a sensor that measures the state of the workpiece or tool;
A feature quantity extraction unit that acquires a feature quantity from the output of the sensor;
a dynamic learning unit that builds a learning model based on the feature amount;
an inference unit that infers the degree of wear based on the learning model constructed by the dynamic learning unit;
a threshold storage unit that stores a wear threshold;
a dynamic threshold setting unit that stores the threshold input by the operator in the threshold storage unit;
A wear determination unit that determines wear based on the degree of wear and the threshold,
The inference unit determines whether or not the threshold has been changed, and if it is determined that the threshold has been changed, transmits a re-learning instruction to the dynamic learning unit,
The wear determination device, wherein the dynamic learning unit performs re-learning based on the feature amount and the changed threshold to update the learning model when receiving the control data from the inference unit.
3. The wear determination according to claim 1, wherein the sensor is a vibration sensor that acquires vibration acceleration as the state of the tool or workpiece, and the feature quantity extraction unit acquires the feature quantity using the vibration acceleration. Device.