CN115210040A

CN115210040A - Wear determination device

Info

Publication number: CN115210040A
Application number: CN202180007968.5A
Authority: CN
Inventors: 池田淳; 古田知康; 原岛健走
Original assignee: Nachi Fujikoshi Corp
Current assignee: Nachi Fujikoshi Corp
Priority date: 2021-02-08
Filing date: 2021-12-14
Publication date: 2022-10-18
Also published as: JPWO2022168455A1; WO2022168455A1

Abstract

Provided is a wear determination device capable of automatically performing machine learning relearning without requiring an expert. A wear determination device (100) for a tool in a processing machine, comprising: a feature value extraction unit (130) that acquires a feature value from the output of the sensor; a dynamic learning unit (140) that constructs a learning model on the basis of the feature quantities; an estimation unit (160) that estimates the degree of wear based on the learning model constructed by the dynamic learning unit; a threshold value storage unit (180) that stores a wear threshold value; and a wear determination unit (190) that determines wear based on the degree of wear and a threshold value, wherein the estimation unit determines whether or not estimation based on the current learning model is appropriate, and when it is determined that the estimation is inappropriate, transmits a relearning instruction to the kinetic learning unit, and when the kinetic learning unit receives control data from the estimation unit, the kinetic learning unit performs relearning based on the threshold value and an additional feature amount different from the feature amount used when the current learning model is constructed, and updates the learning model.

Description

Wear determination device

Technical Field

The present invention relates to a wear determination device to which machine learning is applied.

Background

Conventionally, a large number of techniques have been proposed in which machine learning is applied to wear determination or life prediction of a tool. For example, a life prediction device disclosed in patent document 1 observes life related data to create a probabilistic model of the replacement life of a consumable part, and predicts the replacement life of the consumable part based on the observed life related data using the created probabilistic model. According to patent document 1, it is described that the life of a consumable part of a manufacturing machine can be predicted with a predetermined accuracy even at a stage where the collected data is small.

Documents of the prior art

Patent document

Patent document 1: japanese patent laid-open publication No. 2019-207576

Disclosure of Invention

Problems to be solved by the invention

However, in the conventional machine learning, not only in patent document 1, a learning model is first constructed by learning a wear state under a specific condition, and then the degree of wear is estimated under the same condition based on the learning model. Thus, there are problems as follows: although it is possible to estimate with high accuracy without changing the conditions, if the conditions of the machining method, the tool, the workpiece, and the like change, the accuracy significantly decreases, and appropriate estimation cannot be performed (problem of generalization). Therefore, even if learning is once performed, relearning is required every time conditions change, but relearning is very difficult to perform unless it is a machine learning specialist, and a worker on the spot cannot cope with it.

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

Means for solving the problems

A typical configuration of the present invention is a tool wear determination device for a processing machine, including: a sensor that measures a state of the workpiece or the tool; a feature value extraction unit that acquires a feature value from an output of the sensor; a dynamic learning unit that constructs a learning model based on the feature amount; an estimation unit that estimates a degree of wear based on the learning model constructed by the dynamic learning unit; a threshold value storage unit that stores a threshold value of wear; and a wear determination unit that determines wear based on the degree of wear and a threshold value, wherein the estimation unit determines whether or not estimation based on the current learning model is appropriate, and when it is determined that the estimation is not appropriate, transmits a relearning instruction to the dynamic learning unit, and when the dynamic learning unit receives control data from the estimation unit, the dynamic learning unit performs relearning based on the threshold value and an additional feature amount different from the feature amount used in constructing the current learning model, and updates the learning model.

When conditions of a machining method, a tool, a workpiece, and the like change, the degree of wear estimated from the characteristic amount greatly changes. Therefore, the determination by the wear determination unit is also greatly advanced or delayed from the actual wear limit. However, by dynamically updating the learning model using the feature quantities while performing estimation, it is possible to automatically perform machine learning relearning (improvement of generalization) without requiring expert relearning.

Another representative configuration of the present invention is a tool wear determination device for a processing machine, including: a sensor that measures the state of the workpiece or the tool; a feature value extraction unit that acquires a feature value from an output of the sensor; a dynamic learning unit that constructs a learning model based on the feature quantities; an estimation unit that estimates a degree of wear based on the learning model constructed by the dynamic learning unit; a threshold value storage unit that stores a threshold value of wear; a dynamic threshold value setting unit that stores a threshold value input by an operator in a threshold value storage unit; and a wear determination unit that determines wear based on the wear degree and the threshold value, wherein the estimation unit determines whether or not the threshold value has been changed, and when it is determined that the threshold value has been changed, transmits a relearning instruction to the dynamic learning unit, and when the dynamic learning unit receives the control data from the estimation unit, the dynamic learning unit performs relearning based on the feature amount and the changed threshold value, and updates the learning model.

In the conventional machine learning, it is necessary to set a threshold value before the start of the estimation program, and in order to change the threshold value, it is necessary to temporarily stop the estimation program and then restart the estimation program. However, according to the above configuration, since the threshold value can be dynamically updated, the operator can reset the threshold value without stopping the estimation program, and convenience can be improved.

Preferably, the sensor is a vibration sensor that acquires a vibration acceleration of the tool or the workpiece as a state of the tool or the workpiece, and the feature amount extraction unit acquires the feature amount using the vibration acceleration. The vibration acceleration is highly dependent on the degree of wear and is robust against changes in other conditions, and therefore is preferable as the state quantity for acquiring the feature quantity.

ADVANTAGEOUS EFFECTS OF INVENTION

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.

Drawings

Fig. 1 is a block diagram of a wear determination device according to the present embodiment.

Fig. 2 is a flowchart illustrating an operation of an estimation program of the wear determination device.

Fig. 3 is a diagram illustrating an example of the feature amount.

Fig. 4 is a diagram illustrating an example of the feature amount.

Detailed Description

Preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Dimensions, materials, specific numerical values, and the like shown in the above embodiments are merely examples for facilitating understanding of the present invention, and do not limit the present invention unless otherwise specified. Note that in the present specification and the drawings, elements having substantially the same function and configuration are denoted by the same reference numerals, and overlapping description is omitted, and elements not directly related to the present invention are omitted from illustration.

Fig. 1 is a block diagram of a wear determination device according to the present embodiment, and fig. 2 is a flowchart illustrating an operation of an estimation program of the wear determination device.

The wear determination device 100 is a device for determining 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. For each tool, it is assumed that, for example, machining is performed 100 to 150 times (machining is performed 100 to 150 workpieces), and the tool is replaced if the wear limit is reached. The flowchart of fig. 2 is executed for each process (each workpiece). First, the estimation program determines whether or not machining is being performed (step 300). If the machining is being performed, the data acquisition unit 120 acquires data (step 302).

The data acquisition unit 120 obtains outputs from a sensor 110 attached to the tool spindle and a sensor 112 attached to the workpiece spindle. Specifically, the data acquisition unit 120 includes an amplifier and a recorder.

In the present embodiment, the

sensors

110 and 112 are vibration sensors. In addition to vibration sensors, temperature sensors, displacement sensors, etc. may be considered. However, the values of the temperature sensor and the displacement sensor are also greatly affected by factors other than wear. In contrast, the vibration acceleration is highly dependent on the degree of wear, and is robust against changes in other conditions, and therefore is preferable as the state quantity for acquiring the feature quantity. In addition to the acceleration data obtained from the vibration sensor, the data acquisition unit may also be supplied with data such as "positional deviation" and "integrated power value" of the tool spindle and the workpiece spindle, which are obtained from an NC (Numerical Control), and thereby have the same function.

The feature extraction unit 130 extracts and accumulates feature quantities from the state quantities acquired by the sensors 110 and 112 (step 304). In the present embodiment, the vibration acceleration is used to acquire the feature amount. As the characteristic amount that can be obtained from the vibration acceleration, an absolute value average, RMS (Root Mean Square), STFT (short-time fourier transform), a fundamental frequency of a varying cutting force (cutting edge passing frequency), a harmonic thereof, and the like can be used.

Next, the inference program determines whether to adopt a mode for performing inference or a mode for performing initial learning (step 308) (step 306). The case where no inference is performed means that a learning model is not present yet. The case of performing inference is a case where a learning model is already constructed.

When the estimation is not performed (when the learning model does not exist), the operator 30 determines the wear limit and ends the one-time machining. Then, the inference program performs learning using the accumulated feature quantities (step 308), and constructs an initial learning model.

The dynamic learning unit 140 can perform both static learning for constructing a learning model from only the feature amount output from the feature amount extraction unit 130 and dynamic learning for performing relearning from a relearning instruction (flag) of the estimation unit. This dynamic learning is a feature of the present invention. As the algorithm for machine learning, a known algorithm such as a search algorithm or a genetic algorithm can be used. The dynamic learning unit 140 stores the constructed learning model in the learning storage unit 150.

When the estimation is performed in step 310, the estimation unit 160 estimates the degree of wear based on the learning model constructed by the dynamic learning unit 140 (step 310). That is, the estimation is performed using the learning model read out from the learning storage unit 150 and the feature values sequentially acquired from the feature value extraction unit 130. The degree of wear is a probabilistic model, and is intermediate data indicating that wear of this degree occurs with this degree of probability.

When the

sensors

110 and 112 output outliers, the estimated value processing unit 162 performs weighting to prevent erroneous determination. Specifically, when the machining amount after the tool change is small, the output of the estimation unit 160 is multiplied by a large coefficient for the underestimation. Conversely, when the machining amount after the tool change is large, the coefficient for excessively small evaluation is set small. This improves the validity of the estimated value, thereby preventing erroneous determination.

The wear determination unit 190 determines the wear of the tool using the wear degree (estimated value) obtained via the estimated value processing unit 162 and the threshold value acquired from the threshold value storage unit 180 (step 318). The display device 200 displays the degree of wear output from the estimation unit 160 and the determination result (whether or not the wear limit has been reached) output from the wear determination unit 190 (step 318) (step 320). Here, the value in the threshold storage unit 180 is input with a set value before the first learning (step 308). However, the estimation program can be updated at any time by the operator 30 inputting from the dynamic threshold setting unit 170 during the loop of the estimation program.

Next, relearning will be described. After performing the estimation at step 310, the estimation unit 160 determines whether the estimation is appropriate (step 312). The determination as to whether the estimation is appropriate can be determined as inappropriate, for example, when the estimated value calculated by the current learning model loses the correlation with the number of processes. When it is estimated that the state is inappropriate, the estimation unit 160 transmits a relearning instruction 161 (control data) to the dynamic learning unit 140 (step 313).

The estimation program of the wear determination device 100 determines whether or not the threshold value has been changed (step 314). When the operator 30 inputs a new threshold value from the dynamic threshold value setting unit 170 to set the threshold value in the threshold value storage unit 180, the estimation unit 160 transmits a relearning instruction 161 (control data) to the dynamic learning unit 140 (step 315).

Then, when the estimation is performed (in the flow from step 310), the estimation program also determines whether or not 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. In the event that the wear limit has not been reached, a determination is made as to whether a relearning indication 161 has been issued (step 324). In the case where no relearning indication 161 is issued, return is made to step 300.

When the relearning instruction 161 is transmitted, the dynamic learning unit 140 performs relearning using the accumulated feature amount (step 326). As the feature amount used for the relearning at this time, a feature amount (hereinafter referred to as "additional feature amount") different from the feature amount used when the current learning model (latest learning model) is constructed is used. The additional feature may be transmitted from the estimation unit 160 to the dynamic learning unit 140 together with the relearning instruction 161, or may be directly transmitted from the feature extraction unit 130 to the dynamic learning unit 140.

For example, the first learning model is constructed with the tool a. In the conventional technique, the learning model is used to estimate the tool B and the tool C (step 310 and subsequent steps to step 310), and no further update of the learning model is performed. However, in the present invention, when the relearning instruction 161 is issued, the learning model can be updated using the feature values (additional feature values) of the tool B and the tool C.

If the number of machining passes (for example, 120 passes) in the first time of constructing the learning model by the tool a does not sufficiently reach the vicinity of the wear limit, learning is insufficient, and accurate estimation cannot be performed. In such a case, the operator 30 can make a decision by observing the trend of the determination result by the dynamic threshold setting unit 170 and can relearn the number of additional processes (for example, the feature amount from 121 times to 150 times is an additional feature amount).

When conditions of a machining method, a tool, a workpiece, and the like change, the degree of wear estimated from the characteristic amount greatly changes. Therefore, the determination by the wear determination unit 190 is also greatly advanced or delayed from the actual wear limit. However, as described above, by dynamically determining whether or not the estimation is appropriate using the feature quantity while performing the estimation and updating the learning model appropriately, it is possible to automatically perform the relearning of the machine learning (improvement of the generalization) without requiring the relearning by an expert.

In addition, in the conventional machine learning, it is necessary to set a threshold value before the start of the estimation program, and in order to change the threshold value, it is necessary to temporarily stop the estimation program and then restart the estimation program. However, according to the above configuration, since the threshold value can be dynamically updated, the operator can reset the threshold value without stopping the estimation program, and convenience can be improved.

When the relearning (step 326) is complete, the process immediately returns to step 310 to re-infer. In this case, since the infinite loop is formed if it is continuously determined that the estimation is not appropriate in step 312, the number of relearnings when the estimation is not appropriate is limited to an upper limit. When the number of relearnings reaches the upper limit, a message to that effect is displayed to the operator 30, and the process returns to step 300.

Fig. 3 and 4 are diagrams illustrating examples of the feature amount. As described above, in the present embodiment, the characteristic amount is acquired using the vibration acceleration which is the state amount having robustness against the condition other than the wear.

Fig. 3 (a) shows an absolute value average of the vibration acceleration corresponding to the machining number as the feature amount. Alternatively, the sum of the areas made of the amplitudes may also be calculated. Fig. 3 (b) shows RMS (Root Mean Square) of the amplitude value of the vibration acceleration corresponding to the machining number.

Referring to fig. 3 (a) and (b), it is understood that the absolute average value or RMS increases approximately in proportion to the increase in the number of processes. In this way, the processed number and the average absolute value of the vibration acceleration or the processed number and the RMS of the amplitude value of the vibration acceleration have a correlation. Based on the correlation, machine learning is performed by using the relation between the absolute value average or RMS and the degree of wear, and thus the degree of wear can be estimated from the absolute value average or RMS for each machining.

Fig. 4 shows STFT (short-time fourier transform) of the vibration acceleration of a workpiece (gear) due to 5-pass machining. No.1 in fig. 4 (a) is data of a new tool, and No.18 in fig. 4 (b) is data of a tool that becomes a wear limit.

Referring to fig. 4, the horizontal axis represents time, and it is understood that a spectrum appears in each process pass. The fifth pass is a finishing process, and is a long pass. As a whole, it can be seen from the fact that the frequency intensity of No.18 in fig. 4 (b) is higher, that the vibration is increased when the wear is increased. In addition, around 800Hz at the fifth pass, more spectrum is shown in fig. 4 (b) where the wear is progressing. By performing machine learning by using such a difference in the frequency distribution of the STFT as a feature, it is possible to determine whether the product is a new product or a wear limit. Further, as the difference in the frequency distribution, the distribution may be digitized, or the distribution image may be subjected to image processing as a pattern.

Further, the feature amount may be not only a single index but also a combination of a plurality of indexes. This can further improve the reliability of estimation. Further, since any index should have a correlation with the number of processes, it is preferable that the correlation between the number of processes and the feature value is always evaluated, and the determination result is not displayed when there is no correlation. This can prevent display misjudgment.

Although the preferred embodiments of the present invention have been described above with reference to the drawings, it goes without saying that the present invention is not limited to such examples. It is clear that those skilled in the art can conceive various modifications and variations within the scope of the claims and that these modifications and variations naturally fall within the technical scope of the present invention.

Industrial applicability

The present invention can be used as a wear determination device to which machine learning is applied.

Description of the reference numerals

10: processing the workpiece; 12: a cutter; 20: a workpiece; 30: an operator; 100: a wear determination device; 110: a sensor; 112: a sensor; 120: a data acquisition unit; 130: a feature value extraction unit; 140: a dynamic learning unit; 150: a learning storage unit; 160: an estimation unit; 161: a relearning instruction; 162: an estimated value processing unit; 170: a dynamic threshold setting unit; 180: a threshold value storage unit; 190: a wear determination section; 200: a display device.

Claims

1. A wear determination device for a tool in a processing machine, comprising:

a sensor that measures the state of the workpiece or the tool;

a feature value extraction unit that acquires a feature value from an output of the sensor;

a dynamic learning unit that constructs a learning model based on the feature amount;

an estimation unit that estimates a degree of wear based on the learning model constructed by the dynamic learning unit;

a threshold value storage unit that stores a threshold value of wear; and

a wear determination section that determines wear based on the degree of wear and the threshold value,

wherein the estimation unit determines whether or not estimation based on the current learning model is appropriate, and if it is determined that the estimation is not appropriate, transmits a relearning instruction to the dynamic learning unit,

the dynamic learning unit, upon receiving control data from the estimation unit, performs relearning based on the threshold value and an additional feature amount different from a feature amount used in constructing a current learning model, and updates the learning model.

2. A wear determination device for a tool in a processing machine, comprising:

a sensor that measures a state of the workpiece or the tool;

a threshold value storage unit that stores a threshold value of wear;

a dynamic threshold value setting unit that stores a threshold value input by an operator in the threshold value storage unit; and

wherein the estimation unit determines whether or not the threshold value has been changed, and transmits a relearning instruction to the dynamic learning unit when it is determined that the threshold value has been changed,

the dynamic learning unit, upon receiving control data from the estimation unit, performs relearning based on the feature amount and the changed threshold value, and updates the learning model.

3. The wear determination device according to claim 1 or 2,

the sensor is a vibration sensor that acquires a vibration acceleration of the tool or the workpiece as a state of the tool or the workpiece, and the feature amount extraction unit acquires the feature amount using the vibration acceleration.