KR102655363B1 - Apparatus and method for measuring tool-wear of cnc machine based on vibration data - Google Patents

Abstract

The present invention relates to a device and method for measuring tool wear of CNC machine tools based on vibration data. According to one aspect of the present invention, the device for measuring tool wear of CNC machine tools based on vibration data includes a plurality of time points divided according to tool counter values. A learning vibration data input unit that receives learning vibration data including a plurality of measurement values corresponding to the learning vibration data, and a Fourier transform that converts the plurality of measurement values corresponding to the plurality of time points included in the learning vibration data into a frequency spectrum by applying the Fourier transform. A filter bank unit that generates a spectrogram image composed of energy values for each filter bank corresponding to multiple viewpoints and multiple frequencies by passing the frequency spectrum through the Mel filter bank, and corresponds to the spectrogram image and learning vibration data. It includes a learning unit that trains the CNN model by inputting the tool counter value into the preset CNN model.

Description

Device and method for measuring tool wear of CNC machine tools based on vibration data {APPARATUS AND METHOD FOR MEASURING TOOL-WEAR OF CNC MACHINE BASED ON VIBRATION DATA}

The present invention relates to a technology for measuring tool wear of CNC machine tools, and more specifically, to a technology that can measure tool wear of CNC machine tools using an artificial neural network.

During the machining process using computer numerical control (CNC), damage to tools attached to CNC machine tools causes the production of defective products, reduced production yield, and loss of raw materials. In particular, if tool damage leads to damage to the CNC machine tool, extensive damage may occur due to repair or replacement of the CNC machine tool, so managing the wear of the CNC machine tool tool is very important.

In relation to this, tool management has traditionally been carried out through direct monitoring by field workers and by replacing tools according to a set cycle or number of uses. However, the method of direct monitoring by field workers requires field workers to be present throughout the operation time of the facility, and when operating large-scale CNC facilities, the number of workers must be increased, which increases operating costs. Additionally, it was difficult for people to accurately detect tool damage due to noise from the equipment.

If tools are replaced according to a set cycle or number of uses, differences in the degree of wear depending on the number of uses depending on the work environment or work content cannot be considered. Therefore, there is a problem in that tools are replaced unnecessarily or are not replaced even if the wear is severe compared to the number of actual uses.

Republic of Korea Patent Application No. 10-2019-0159773 Republic of Korea Patent Application No. 10-2020-0129492

The purpose of the present invention is to solve the above problems, and to provide a device and method for measuring tool wear of CNC machine tools that can learn CNN models by converting high-dimensional vibration data into images using voice feature extraction techniques. .

Additionally, the purpose of the present invention is to provide an apparatus and method for measuring tool wear of a CNC machine tool that can obtain a predicted tool counter value by applying an image converted from vibration data to a CNN model.

The object of the present invention is not limited to the object mentioned above, and other objects not mentioned can be clearly understood from the description below.

In order to achieve the above-described object, the tool wear measurement device for CNC machine tools based on vibration data according to one aspect of the present invention provides learning vibration data including a plurality of measurement values corresponding to a plurality of time points divided according to the tool counter value. A learning vibration data input unit that receives input, a Fourier transform unit that converts a plurality of measurement values corresponding to multiple time points included in the learning vibration data into a frequency spectrum by applying Fourier transformation, and a plurality of measurement values by passing the frequency spectrum through a Mel filter bank. A filter bank unit that generates a spectrogram image composed of energy values for each filter bank corresponding to each viewpoint and multiple frequencies, and a tool counter value corresponding to the spectrogram image and learning vibration data are input into a preset CNN model to create a CNN model. It includes a learning unit that learns.

According to another aspect of the present invention, the tool wear measurement device for CNC machine tools based on vibration data includes an evaluation vibration data input unit that receives evaluation vibration data including a plurality of measurement values corresponding to a plurality of time points divided according to tool counter values. , a Fourier transform unit that converts a plurality of measured values corresponding to a plurality of time points included in the evaluation vibration data into a frequency spectrum by applying the Fourier transform, and passes the frequency spectrum through a Mel filter bank to obtain a plurality of time points and multiple frequencies. A filter bank unit generates a spectrogram image composed of energy values for each corresponding filter bank, and learning vibration data including multiple measurement values corresponding to multiple viewpoints is Fourier transformed and passed through a Mel filter bank. It includes a tool counter value prediction unit that receives the tool counter value corresponding to the image and learning vibration data and inputs the spectrogram image for the evaluation vibration data into a pre-trained CNN model to obtain a predicted tool counter value.

A method of measuring tool wear of a CNC machine tool based on vibration data according to another aspect of the present invention includes the steps of receiving learning vibration data including a plurality of measurement values corresponding to a plurality of time points divided according to the tool counter value, learning Converting a plurality of measured values corresponding to a plurality of time points included in the vibration data into a frequency spectrum by applying Fourier transform, passing the frequency spectrum through a Mel filter bank to obtain a plurality of measured values corresponding to a plurality of time points and a plurality of frequencies. A step of generating a spectrogram image composed of energy values for each filter bank, a step of learning a CNN model by inputting the spectrogram image and the tool counter value corresponding to the learning vibration data into a preset CNN model, and a step of learning the CNN model according to the tool counter value. Step of receiving evaluation vibration data containing multiple measurement values corresponding to multiple time points, Fourier transforming the evaluation vibration data, and passing the spectrogram image through the Mel filter bank to input the prediction tool counter value into the CNN model. Includes acquisition steps.

According to the present invention, it is effective to provide an apparatus and method for measuring tool wear of CNC machine tools that can learn CNN models by converting high-dimensional vibration data into images using a voice feature extraction technique.

According to the present invention, it has the effect of providing an apparatus and method for measuring tool wear of a CNC machine tool that can obtain a predicted tool counter value by applying an image converted from vibration data to a CNN model.

Accordingly, it is possible to expect the effect of being able to accurately determine wear using the predicted tool counter value, regardless of the difference in wear depending on the number of times the CNC machine tool is used.

The effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description of the claims.

Figure 1 is a block diagram of a device for measuring tool wear of a CNC machine tool based on vibration data according to an embodiment of the present invention.
Figure 2 is a flow chart of a method for measuring tool wear of a CNC machine tool based on vibration data according to another embodiment of the present invention.
Figure 3 is an example diagram showing a data set for explaining vibration data corresponding to a tool counter value in embodiments of the present invention.
Figure 4 is an example diagram showing a spectrogram image generated in embodiments of the present invention.

The advantages and features of the present invention and methods for achieving them will become clear by referring to the embodiments described in detail below along with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below and will be implemented in various different forms, but the present embodiments only serve to ensure that the disclosure of the present invention is complete, and those skilled in the art It is provided to fully inform the person of the scope of the invention, and the present invention is defined only by the claims. Meanwhile, the terms used in this specification are for describing embodiments and are not intended to limit the present invention. As used herein, singular forms also include plural forms, unless specifically stated otherwise in the context.

Referring to Figure 1, the tool wear measurement device 10 of a CNC machine tool based on vibration data includes a learning vibration data input unit 101, an evaluation vibration data input unit 103, a data division unit 110, and a Fourier transform unit 120. ), a filter bank unit 130, a grayscale conversion unit 140, a size adjustment unit 150, a learning unit 160, a tool counter value prediction unit 170, and a wear determination unit 180. .

The learning vibration data input unit 101 can receive learning vibration data including a plurality of measurement values corresponding to a plurality of time points divided according to the tool counter value.

The evaluation vibration data input unit 103 may receive evaluation vibration data including a plurality of measured values corresponding to a plurality of time points divided according to the tool counter value.

Here, the learning vibration data refers to the vibration data used for learning the CNN model among the vibration data generated for each CNC machine tool, and the evaluation vibration data refers to the object of wear judgment among the vibration data generated for each CNC machine tool. This means vibration data.

Referring to Figure 3, the data set is a plurality of vibration data (1010) consisting of a plurality of measurement values measured at a plurality of points in time from a vibration sensor attached to the CNC machine tool, and is obtained from the tool counter of the CNC machine tool and contains a plurality of vibration data. It may include a plurality of tool counter values 1020 corresponding to each piece of data.

According to the present invention, vibration data for each tool counter value can be input as learning vibration data or evaluation vibration data from a dataset consisting of a plurality of tool counter values and a plurality of vibration data corresponding to each of the plurality of tool counter values.

The data division unit 110 divides the plurality of measured values corresponding to a plurality of time points included in the learning vibration data or the evaluation vibration data into a plurality of time points corresponding to a plurality of time points with a preset length in chronological order from the first time point among the plurality of time points. By extracting the measured values, multiple split data can be created.

The data division unit 110 applies a window of a preset length from the first time point to a plurality of measurement values corresponding to a plurality of time points included in the learning vibration data or evaluation vibration data to determine the measurement values corresponding to a plurality of time points according to time. Split data can be created by extracting a set number of measured values.

The data division unit 110 moves a window of a preset length by a preset interval and generates segmented data, thereby generating segmented data in which at least one or more measurement values overlap with the segmented data generated at the previous position of the window. .

For example, the data division unit 110 overlaps 512 measurement values over time in vibration data consisting of 20,000 measurement values corresponding to a plurality of time points and divides them into 2048 divisions consisting of a plurality of measurement values corresponding to a preset length. Data can be generated.

The Fourier transform unit 120 may receive learning vibration data or evaluation vibration data and convert them into a frequency spectrum by applying Fourier transform to a plurality of measured values corresponding to a plurality of time points included in the vibration data.

The Fourier transform unit 120 may apply Fourier transform to each of the plurality of split data generated by the data splitter 110 for the learning vibration data or the evaluation vibration data to convert them into a plurality of frequency spectra.

The Fourier transform unit 120 may apply Fourier transform to a plurality of measured values corresponding to a plurality of time points included in the segmented data and convert them into a frequency spectrum composed of values corresponding to a plurality of frequencies.

The filter bank unit 130 passes the frequency spectrum converted by the Fourier transform unit 120 through a Mel-filter bank, thereby converting the frequency spectrum into energy values for each filter bank corresponding to a plurality of viewpoints and a plurality of frequencies. It may be to generate a configured spectrogram image.

The filter bank unit 130 may generate a plurality of spectrogram images for learning vibration data or evaluation vibration data by passing each of the plurality of frequency spectra converted in the Fourier transform unit 120 through a Mel filter bank. .

According to the above configuration, high-dimensional vibration data including a plurality of measurement values corresponding to a plurality of viewpoints can be converted into a two-dimensional image. According to the present invention, vibration data can be converted into an image using a voice feature extraction technique and applied to a CNN model.

In addition, by generating a plurality of divided data from each vibration data, a plurality of images can be generated for each tool counter value. Accordingly, prediction performance can be improved by generating a large number of images to be used for training the CNN model while reducing the amount of individually processed data.

Figure 4 is an example diagram showing a spectrogram image generated in embodiments of the present invention. Spectrogram images can express the characteristics of vibration data in two dimensions, but if the range of energy values is wide, performance may be poor.

In order to prevent the above problem, the present invention may include a configuration for converting the spectrogram image to grayscale and changing it to a preset size before applying it to the CNN model.

The grayscale conversion unit 140 converts the energy value of each filter bank of the spectrogram image generated for the learning vibration data or evaluation vibration data by the filter bank unit 130 into a grayscale value and can generate a grayscale image. .

The grayscale conversion unit 140 applies the RGB value according to the energy value for each filter bank of the spectrogram image generated for the learning vibration data or the evaluation vibration data by the filter bank unit 130 to a preset color conversion model to grayscale. You can convert it to a value and create a grayscale image.

The size adjuster 150 may change the size of the grayscale image to a preset size and output the resized image.

For example, the size adjustment unit 150 may adjust the size of the grayscale image to a preset size (64, 64, 1) and output it.

The learning unit 160 may learn a CNN model by inputting the spectrogram image generated for the learning vibration data by the filter bank unit 130 and the tool counter value corresponding to the learning vibration data into a preset CNN model.

The learning unit 160 inputs the resized image output from the size adjusting unit 150 for the spectrogram image for the learning vibration data and the tool counter value corresponding to the learning vibration data into a preset CNN model and constructs the CNN model. It can be learned.

The tool counter value prediction unit 170 inputs the spectrogram image generated for the evaluation vibration data in the filter bank unit 130 into the CNN model learned in the learning unit 160 to obtain a predicted tool counter value. .

The tool counter value prediction unit 170 inputs the size-adjusted image output from the size adjustment unit 150 for the spectrogram image for the evaluation vibration data into the CNN model learned in the learning unit 160 to predict the tool counter value. can be obtained.

The wear determination unit 180 outputs one of a plurality of preset wear classification values according to the predicted tool counter value and can determine whether the tool is worn.

The wear determination unit 180 may output a preset first wear classification value if the predicted tool counter value does not exceed a preset critical counter value.

When the predicted tool counter value exceeds the preset threshold counter value, the wear determination unit 180 calculates a difference value according to the difference between the predicted tool counter value and the tool counter value corresponding to the evaluation vibration data, and the difference value is set to the preset threshold value. If the difference value does not exceed the preset second wear classification value, the preset wear classification value can be output, and if the difference value exceeds the preset threshold difference value, the preset third wear classification value can be output.

When the first wear classification value is output from the wear determination unit 180, it may be recognized that tool replacement is not necessary. When the second wear classification value is output from the wear determination unit 180, it may be recognized that the tool needs to be replaced. When the third wear classification value is output from the wear determination unit 180, it may be recognized that tool replacement and on-site environmental inspection are necessary.

According to the present invention, a predicted tool counter value can be obtained through analysis of vibration data separately from the actually measured tool counter value, and the wear of the tool can be determined using the predicted tool counter value. Therefore, wear can be accurately determined even if wear varies depending on the environment in which the CNC machine tool is installed or the work performed.

The method of measuring tool wear of a CNC machine tool based on vibration data according to another embodiment of the present invention can be performed by a device for measuring tool wear of a CNC machine tool based on vibration data according to an embodiment of the present invention.

For convenience of description below, in FIGS. 1 and 2, the reference numerals of functional parts are identical and are not described in duplicate.

The tool wear measurement device 10 of a CNC machine tool receives learning vibration data containing a plurality of measurement values corresponding to a plurality of time points divided according to the tool counter value (S10), and receives a plurality of measurement values included in the learning vibration data. A plurality of split data can be generated by extracting a plurality of measurement values corresponding to a plurality of time points of a preset length in chronological order starting from the first of the plurality of time points from the plurality of measurement values corresponding to the time points (S20). .

The tool wear measurement device 10 of a CNC machine tool converts each of the plurality of segmented data into a plurality of frequency spectra by applying Fourier transform (S30), and passes each of the plurality of frequency spectra through the Mel filter bank, respectively, to obtain a plurality of frequency spectra. A plurality of spectrogram images composed of energy values for each filter bank corresponding to each of the viewpoints and plural frequencies can be generated (S40).

The tool wear measurement device 10 of a CNC machine tool converts the energy value for each filter bank of the spectrogram image generated in step S40 into a grayscale value, generates a grayscale image (S50), and measures the size of the grayscale image. The CNN model can be trained by changing the size to a set size (S60) and inputting the image resized to the preset size and the tool counter value corresponding to the learning vibration data into the preset CNN model (S70).

Afterwards, the tool wear measurement device 10 of the CNC machine tool receives evaluation vibration data including a plurality of measurement values corresponding to a plurality of time points divided according to the tool counter value (S80), and the evaluation vibration data included in the evaluation vibration data is input (S80). A plurality of split data can be generated by extracting a plurality of measurement values corresponding to a plurality of time points of a preset length in chronological order starting from the first of the plurality of time points from a plurality of measurement values corresponding to a plurality of time points ( S90).

The tool wear measurement device 10 of a CNC machine tool converts each of the plurality of divided data into a plurality of frequency spectra by applying the Fourier transform (S100), and passes each of the plurality of frequency spectra through the Mel filter bank to obtain a plurality of frequency spectra. A plurality of spectrogram images consisting of energy values for each filter bank corresponding to each of the viewpoints and a plurality of frequencies are generated (S110), and the energy values for each filter bank of the spectrogram image are converted to grayscale values to generate a grayscale image. And (S120), the size of the grayscale image can be changed to a preset size (S130).

The tool wear measurement device 10 of a CNC machine tool inputs an image resized to a preset size in step S130 into a pre-trained CNN model in step S70 to obtain a predicted tool counter value (S140), and adds the predicted tool counter value to the predicted tool counter value. Accordingly, one of a plurality of preset wear classification values can be output to determine whether the tool is worn (S150).

Meanwhile, the blocks in the attached block diagram and the steps in the flow chart are computer instructions that are mounted on the processor or memory of an electronic device capable of data processing (e.g., general-purpose computer, special-purpose computer, portable laptop computer, network computer) and perform designated functions. It can be implemented as instructions. Because these computer program instructions can be stored in computer-readable memory, the functions described in the blocks of a block diagram or the steps of a flow diagram can also be produced as an article containing instruction means to perform them.

Those skilled in the art to which the present invention pertains will understand that the present invention can be implemented in other specific forms without changing its technical idea or essential features. Therefore, the embodiments described above should be understood in all respects as illustrative and not restrictive. The scope of the present invention is indicated by the scope of the claims described below rather than the detailed description above, and all changes or modified forms derived from the scope of the claims and their equivalent concepts should be construed as being included in the scope of the present invention.

10: Tool wear measurement device for CNC machine tools based on vibration data
101: Learning vibration data input unit
103: Evaluation vibration data input unit
110: data division unit
120: Fourier transform unit
130: Filter bank unit
140: Grayscale conversion unit
150: Size adjustment part
160: Learning Department
170: Tool counter value prediction unit
180: Wear judgment unit

Claims (11)

A learning vibration data input unit that receives learning vibration data including a plurality of measurement values corresponding to a plurality of time points divided according to the tool counter value;
a Fourier transform unit that converts a plurality of measured values corresponding to a plurality of time points included in the learning vibration data into a frequency spectrum by applying Fourier transform;
A filter bank unit that generates a spectrogram image composed of energy values for each filter bank corresponding to a plurality of viewpoints and a plurality of frequencies by passing the frequency spectrum through a Mel filter bank; and
A tool wear measurement device for a CNC machine tool based on vibration data including a learning unit that inputs the spectrogram image and the tool counter value corresponding to the learning vibration data into a preset CNN model to train the CNN model. According to paragraph 1,
By extracting a plurality of measurement values corresponding to a plurality of time points preset in time order starting from the first of the plurality of time points from a plurality of measurement values corresponding to a plurality of time points included in the learning vibration data, a plurality of split data are generated. It further includes a data division unit that generates,
The Fourier transform unit
Applying Fourier transform to each of the plurality of divided data to convert them into a plurality of frequency spectra.
Tool wear measurement device for CNC machine tools. According to paragraph 1,
A grayscale conversion unit that converts the energy value of each filter bank of the spectrogram image generated for the learning vibration data by the filter bank unit into a grayscale value and generates a grayscale image; and
Further comprising a size adjustment unit that changes the size of the grayscale image to a preset size and outputs it.
A tool wear measurement device for CNC machine tools based on vibration data. According to paragraph 3,
The learning department
Learning the CNN model by inputting the image output from the size control unit for the learning vibration data and the tool counter value corresponding to the learning vibration data into a preset CNN model.
A tool wear measurement device for CNC machine tools based on vibration data. An evaluation vibration data input unit that receives evaluation vibration data including a plurality of measurement values corresponding to a plurality of time points divided according to the tool counter value;
a Fourier transform unit that converts a plurality of measured values corresponding to a plurality of time points included in the evaluation vibration data into a frequency spectrum by applying Fourier transform;
A filter bank unit that generates a spectrogram image composed of energy values for each filter bank corresponding to a plurality of viewpoints and a plurality of frequencies by passing the frequency spectrum through a Mel filter bank; and
A CNN model trained in advance by receiving an image generated by Fourier transforming learning vibration data containing multiple measurement values corresponding to multiple viewpoints and passing it through a Mel filter bank, as well as a tool counter value corresponding to the learning vibration data. A tool counter value prediction unit that obtains a predicted tool counter value by inputting a spectrogram image for the evaluation vibration data to a tool wear measurement device of a CNC machine tool based on vibration data, including a tool counter value prediction unit. According to clause 5,
A tool wear measurement device for a CNC machine tool further comprising a wear determination unit that outputs one of a plurality of preset wear classification values according to the predicted tool counter value and determines whether the tool is worn. In the tool wear measurement method performed by a tool wear measurement device of a CNC machine tool based on vibration data,
Receiving learning vibration data including a plurality of measurement values corresponding to a plurality of time points divided according to the tool counter value;
Converting a plurality of measured values corresponding to a plurality of time points included in the learning vibration data into a frequency spectrum by applying Fourier transform;
Passing the frequency spectrum through a Mel filter bank to generate a spectrogram image composed of energy values for each filter bank corresponding to a plurality of viewpoints and a plurality of frequencies;
Inputting the spectrogram image and the tool counter value corresponding to the learning vibration data into a preset CNN model to train the CNN model;
Receiving evaluation vibration data including a plurality of measured values corresponding to a plurality of time points divided according to the tool counter value; and
A method for measuring tool wear of a CNC machine tool based on vibration data including; obtaining a predicted tool counter value by inputting a spectrogram image obtained by Fourier transforming the evaluated vibration data and passing it through a Mel filter bank into the CNN model. . In clause 7,
After receiving the learning vibration data and before converting it to the frequency spectrum,
By extracting a plurality of measurement values corresponding to a plurality of time points preset in time order starting from the first of the plurality of time points from a plurality of measurement values corresponding to a plurality of time points included in the learning vibration data, a plurality of split data are generated. generating a; further comprising,
The step of converting to the frequency spectrum is
Converting each of the plurality of divided data into a plurality of frequency spectra by applying Fourier transform.
Method for measuring tool wear of CNC machine tools. In clause 7,
After generating the spectrogram image and before training the CNN model,
Converting energy values for each filter bank of the spectrogram image generated for the learning vibration data into grayscale values and generating a grayscale image; and
A method of measuring tool wear of a CNC machine tool based on vibration data, further comprising changing the size of the grayscale image to a preset size and outputting it. According to clause 9,
The step of training the CNN model is
Learning the CNN model by inputting the grayscale image changed to the preset size and the tool counter value corresponding to the learning vibration data into the preset CNN model.
A method of measuring tool wear of CNC machine tools based on vibration data. In clause 7,
After obtaining the prediction tool counter value,
A method for measuring tool wear of a CNC machine tool based on vibration data, further comprising outputting one of a plurality of preset wear classification values according to the predicted tool counter value and determining whether the tool is worn.

Images