KR20230063751A - Method, apparatus and system for predicting tool-wear of cnc machine based on artificial intelligence - Google Patents

Abstract

The present invention relates to a method for predicting tool wear of a CNC machine tool based on artificial intelligence, an apparatus for the same, and a system for the same, capable of predicting whether a tool is worn by using data measured by a sensor attached to a CNC machine tool. According to an aspect of the present invention, the method for predicting tool wear of a CNC machine tool based on artificial intelligence includes: a step of storing a tool counter value according to a driving number of the CNC machine tool, multiple measurement values corresponding to each of multiple sensors and measured by the multiple sensors attached to the CNC machine tool, and monitoring data including a G code transmitted to the CNC machine tool according to the input of a user depending on time; a step of generating a division data set comprising multiple pieces of monitoring data having the same tool counter value by dividing the multiple pieces of monitoring data stored depending on the time based on the tool counter value; a step of generating an uniform data set in which a time series is uniformized by removing the monitoring data including the G code corresponding to a predetermined removal code in the multiple pieces of monitoring data included in the division data set; a step of generating a resampling data set including a measurement value of a number according to a sampling rate by each item by resampling multiple measurement values by each item corresponding to each sensor included in the uniform data set based on the predetermined sampling rate; a step of generating a normalization data set by maximally and minimally normalizing the value included in the resampling data set by each item; and a step of predicting whether the tool is worn by applying the normalization data set to a learning model which is learned in advance.

Description

Tool wear prediction method, apparatus, and system of CNC machine tools based on artificial intelligence

The present invention relates to an artificial intelligence-based tool wear prediction method, apparatus, and system for predicting tool wear of a CNC machine tool based on artificial intelligence.

If the tool attached to the CNC equipment is damaged during the machining process using computer numerical control (CNC), the parts to be produced are produced as defective products, which reduces the production yield and causes enormous damage such as loss of raw materials. In addition, tool damage may cause damage to the CNC equipment itself, so it is very important to manage tool wear in CNC machine tools.

In order to solve this problem, a method in which a field worker periodically monitors the CNC equipment and a method in which a tool is replaced according to a preset cycle or number of uses is used. However, the monitoring method by the field worker has a problem in that it is difficult for the field worker to detect tool breakage due to the noise of the facility. In addition, since field workers must reside throughout the operation time of the CNC facility, it is inconvenient and in an environment where a large-scale CNC facility is operated, the number of field workers must be increased in proportion to the size of the facility, which increases operating costs.

The replacement method according to a predetermined cycle or number of uses has a problem in that the tool is replaced unnecessarily as the tool is replaced regardless of the actual tool wear, resulting in production inefficiency.

In this regard, a technology for predicting wear of a tool by applying a sensing value to a learning model has been disclosed, but most learning models basically learn based on time-series data with uniform measurement intervals, compared to actual CNC machine tools. The measured value measured from the sensor attached to has an irregular measurement cycle, and problems such as missing some measured value due to problems in the network environment occur, making it difficult to apply the data measured from the actual CNC machine tool to the learning model. .

Republic of Korea Patent Application No. 10-2019-0159773 Republic of Korea Patent Application No. 10-2017-0014840

An object of the present invention is to solve the above problems, and an artificial intelligence-based tool wear prediction method, apparatus, and to provide the system.

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

A method for predicting tool wear of a CNC machine tool based on artificial intelligence according to an aspect of the present invention for achieving the above object is a tool counter value according to the number of drives of the CNC machine tool input from the outside, and a plurality of attached CNC machine tools. Storing monitoring data over time, including a plurality of measured values corresponding to each of a plurality of sensors measured from two sensors and a G-code transmitted to a CNC machine tool according to a user's input; Generating a divided data set composed of a plurality of monitoring data having the same tool counter value by classifying the monitoring data based on the tool counter value, determining whether the plurality of monitoring data included in the divided data set corresponds to a preset removal code. Removing monitoring data including code and generating a uniform data set in which the time series is uniform, resampling a plurality of measurement values for each item corresponding to each sensor included in the uniform data set based on a preset sampling rate to obtain a uniform data set for each item Generating a resampling data set including the number of measured values according to the sampling rate, maximally-minimum normalizing the values included in the resampling data set for each item, generating a normalized data set, and learning the normalized data set in advance Applying to the model to predict tool wear.

An apparatus for predicting tool wear of a CNC machine tool based on artificial intelligence according to another aspect of the present invention provides a tool counter value according to the number of times the CNC machine tool is driven from a CNC machine tool control module that controls the operation of the CNC machine tool, and a CNC machine tool Data collection that receives monitoring data including a plurality of measured values corresponding to each of the plurality of sensors measured from a plurality of sensors attached to the controller and G-code transmitted to the CNC machine tool according to user input and stored according to time A data division unit that divides a plurality of monitoring data stored according to time based on the tool counter value to create a divided data set composed of a plurality of monitoring data having the same tool counter value, and a plurality of monitoring data included in the divided data set A time series equalization unit that removes monitoring data including a G code corresponding to a preset removal code among data and creates a uniform data set in which the time series is uniform, and responds to each sensor included in the uniform data set based on a preset sampling rate Resampling a plurality of measured values for each item to generate a resampling data set including the number of measured values according to the sampling rate for each item Includes a resampling unit, a normalization unit that normalizes the values included in the resampling data set by item to minimum and maximum, and generates a normalized data set, and a tool wear prediction unit that predicts tool wear by applying the normalized data set to a pre-learned learning model do.

A tool wear prediction system for a CNC machine tool based on artificial intelligence according to another aspect of the present invention transfers the G-code input from the outside to the CNC machine tool to control the operation of the CNC machine tool, CNC machine tool that externally transmits monitoring data including a tool counter value according to the tool counter value, a plurality of measurement values corresponding to each of the plurality of sensors measured from a plurality of sensors attached to the CNC machine tool, and a G-code transmitted to the CNC machine tool. It receives monitoring data from the machine control module and CNC machine tool control module and stores it according to time. Creates a split data set, removes monitoring data including a G code corresponding to a preset removal code among a plurality of monitoring data included in the split data set, creates a uniform data set with a uniform time series, and sets a sampling rate Based on , a plurality of measured values for each item corresponding to each sensor included in the uniform data set are resampled to generate a resampling data set including the number of measured values according to the sampling rate for each item, and the value included in the resampling data set and a tool wear prediction device for predicting tool wear by item-by-item maximum-minimum normalization, generating a normalization data set, and applying the normalization data set to a pre-learned learning model.

According to the present invention, there is an effect of providing an artificial intelligence-based tool wear prediction method, apparatus, and system for predicting tool wear using data measured by sensors attached to the CNC machine tool.

According to the present invention, it is possible to increase wear prediction accuracy by predicting tool wear based on measured data.

Therefore, an effect of increasing production efficiency by preventing unnecessary tool replacement can be expected.

In addition, it has the effect of preventing the damage of the CNC equipment by stopping the operation of the machine tool before the tool is damaged according to the prediction result of whether the tool is worn or not.

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.

1 is a block diagram of a tool wear prediction system for a CNC machine tool based on artificial intelligence according to an embodiment of the present invention.
2 is a flowchart of a method for predicting tool wear of a CNC machine tool based on artificial intelligence according to another embodiment of the present invention.
3 is a flowchart illustrating a step of generating a resampling data set in more detail in the method for predicting tool wear of a CNC machine tool based on artificial intelligence according to another embodiment of the present invention.
4 is an exemplary diagram for explaining monitoring data stored according to time in embodiments of the present invention.
5 is an exemplary view for explaining a divided data set generated from monitoring data stored according to time according to embodiments of the present invention.
6 is an exemplary diagram illustrating a time-series homogenized uniform data set according to embodiments of the present invention.
7 is an exemplary diagram illustrating a graph of a plurality of measurement values measured over time from one sensor and a number of measurement values corresponding to a reference value selected by downsampling the plurality of measurement values.

Advantages and features of the present invention, and methods of achieving them, will become clear with reference to the detailed description of the following embodiments taken in conjunction with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but will be implemented in a variety of different forms, only the present embodiments make the disclosure of the present invention complete, and those skilled in the art in the art to which the present invention belongs It is provided to fully inform the person of the scope of the invention, and the invention is defined only by the recitation of the claims. Meanwhile, terms used in this specification are for describing embodiments, and are not intended to limit the present invention. In this specification, singular forms also include plural forms unless specifically stated otherwise in a phrase.

A tool wear prediction method of an artificial intelligence-based CNC machine tool according to another embodiment of the present invention is an artificial intelligence-based CNC tool wear prediction system 10 of an artificial intelligence-based CNC machine tool according to an embodiment of the present invention. It can be performed by the tool wear prediction device 200 of the machine tool.

For convenience of description below, in FIGS. 1 to 3, reference numerals of functionally identical functional units are identical and overlapping descriptions will not be made.

Referring to FIG. 1, the tool wear prediction system 10 of an artificial intelligence-based CNC machine tool according to an embodiment of the present invention includes a CNC machine tool control module 100 and an artificial intelligence-based tool wear prediction device of a CNC machine tool. It may be composed of (200).

The CNC machine tool control module 100 transfers G-CODE input from the outside to the CNC machine tool to control the operation of the CNC machine tool, and the tool counter value according to the number of times the CNC machine tool is driven and the CNC machine tool Monitoring data including a plurality of measurement values corresponding to each of the plurality of sensors measured by a plurality of sensors attached to the machine and the G-code transmitted to the CNC machine tool may be externally transmitted.

A plurality of sensors attached to the CNC machine tool include a vibration sensor, a gyro sensor, a current sensor that measures the motor current of the CNC machine tool, a speed measurement sensor that measures the spindle motor speed of the CNC machine tool, and a feed rate sensor that measures the feed rate. It may contain.

Here, the monitoring data is a tool counter value, a plurality of vibration sensors, a plurality of gyro sensors, a motor current measurement sensor, a spindle motor speed measurement sensor, a plurality of measurement values corresponding to each feed rate sensor, a measurement time point, and a CNC work at the measurement time point. It may be to include a jigod input into the machine.

The tool wear prediction device 200 collects the monitoring data transmitted from the CNC machine tool control module 100, and the time series is uniformed from the collected monitoring data, and a data set including the number of measured values corresponding to the preset sampling rate , and apply it to the pre-learned learning model to predict tool wear of CNC machine tools to predict tool wear of CNC machine tools.

The tool wear prediction device 200 includes a data collection unit 210, a data division unit 220, a time series equalization unit 230, a resampling unit 240, a normalization unit 250, a tool wear estimation unit 260, and a communication unit. (270) may be included.

The data collection unit 210 receives the monitoring data transmitted from the CNC machine tool control module 100 and stores them according to time in chronological order (S100).

Referring to FIG. 4 , monitoring data stored in the data collection unit 210 includes a device ID (equ_id) for identifying a CNC machine tool, an input time point (time), and a plurality of measured values (current_voltage) corresponding to a plurality of sensors. , sensor1, sensor2, sensor3, sensor4, sensor5, sensor6, current_tool, spindle, feedrate), a tool counter value (tool_count), and a g code (exec_gcode).

Here, the plurality of measurement values corresponding to the plurality of sensors are a plurality of vibration values (sensor1, sensor2, sensor3) according to the vibration of the CNC machine tool, and a plurality of angular velocity values (sensor4, sensor5, sensor6) according to the movement of the CNC machine tool. ), the current value (current_tool) of the CNC machine tool motor, the speed (spindle) of the spindle motor, and the feedrate value (feedrate), but is not limited thereto.

The data collection unit 210 may be to store monitoring data according to time based on the input time point at which the monitoring data is input from the CNC machine tool control module 100, and in another example, from the CNC machine tool control module 100 Monitoring data including the measurement time point may be received and stored according to time based on the measurement point.

The data division unit 220 divides the plurality of monitoring data stored according to time in the data collection unit 210 based on the tool counter value and generates a divided data set composed of a plurality of monitoring data having the same tool counter value ( S200).

Referring to FIG. 5 , the data partitioning unit 220 bundles a plurality of monitoring data according to time with the same tool counter value (tool_count) to form a first divided data set 301 composed of a plurality of monitoring data having tool_count = 71; A second divided data set 302 composed of a plurality of monitoring data with tool_count = 72 and a third divided data set 303 composed of a plurality of monitoring data with tool_count = 73 may be generated for each tool counter value (S200 ).

The CNC machine tool performs work for a preset time based on the G-code entered by the user, but the sensor installed on the CNC machine tool measures the measurement value when a change in the measurement item occurs regardless of whether the work is being performed. Unnecessary monitoring data at unrelated time points is collected, and accordingly, a problem occurs in which time intervals corresponding to divided data sets divided based on tool counter values are not unified.

The time series equalization unit 230 removes monitoring data including a G code corresponding to a preset removal code among a plurality of monitoring data included in each divided data set generated by the data division unit 220, and performs actual work. It may be to generate a uniform data set composed of monitoring data corresponding to a time point and having a uniform time series (S300).

The G-code is input from the user to control the machine tool to perform a specific machining operation, and the CNC machine tool performs the operation during a predetermined machining time according to the G-code input from the user.

Here, the removal code is a code for performing position adjustment of the machine tool before and after the operation, and may be, for example, "G28U0W0" or "G0X130.Z30."

Referring to FIGS. 5 and 6, the time series smoothing unit 230 performs a first segment including a G code corresponding to G0X130.Z30., which is a preset removal code, in the second divided data set 302 corresponding to tool_count=72. It may be to remove the monitoring data and the last monitoring data including the G-code corresponding to G28U0W0 to generate a uniform data set consisting only of the monitoring data corresponding to the time when the actual operation was performed (310).

According to the present invention, by removing the monitoring data including the G-code corresponding to the removal code, which is a code for performing position adjustment of the machine tool before and after the work, it corresponds to a certain machining time when the actual work was performed according to the G-code. It is possible to create a uniform data set in which the time series are homogenized, consisting only of the monitored data.

That is, the time series equalization unit 230 eliminates unnecessary monitoring data unrelated to the work operation before and after the work based on the G-code that controls the machine tool, so that the time series corresponding to the time when the actual work is performed is uniform data. A set can be created.

The resampling unit 240 receives the uniform data set generated by the time series equalization unit 230 and resamples a plurality of measured values for each item corresponding to each sensor included in the uniform data set based on a preset sampling rate to obtain It may be to generate a resampling data set including the number of measured values according to a preset sampling rate (S400).

Specifically, the resampling unit 240 calculates a reference value by multiplying the work time according to the difference between the earliest time point and the last time point among a plurality of time points corresponding to a plurality of monitoring data included in the divided data set by a preset sampling rate, (S410), it may be to compare whether the number of monitoring data included in the divided data set exceeds a reference value (S420).

When the number of monitoring data included in the divided data set exceeds the reference value, the resampling unit 240 selects a plurality of measurement values over time for each item corresponding to each sensor included in the divided data set by the number according to the reference value. and generating a downsampled resampling data set to include as many selected measurement values as the number according to the reference value for each item (S430).

Hereinafter, the generation of a resampling data set will be described in more detail with reference to the following example. The following examples are arbitrarily set for convenience of explanation, and are not limited thereto.

The table below shows a split data set as an example.

point of view vibration sensor gyro sensor feed rate sensor 00:07.6 15 0.0366 984 00:08.6 15 0.041 983 00:09.7 12 0.0356 981 00:10.9 17 0.0381 982 00:22.9 19 0.0356 792 00:23.9 18 0.0376 795 00:25.6 19 0.0366 794

When the reference value is divided by 5 and the number of monitoring data included in the data set exceeds the reference value, the resampling unit 240 corresponds to each sensor, that is, to the vibration sensor, among a plurality of measurement values included in the divided data set. Among the plurality of measured values (15, 15, 12, 17, 19, 18, 19) according to the time of Among the measured values (0.0366, 0.041, 0.0356, 0.0381, 0.0356, 0.0376, 0.0366), the number (5) measured values corresponding to the reference value is selected, and a plurality of measured values (984, 983, 981, 982, 792, 795, 794), select the number (5) of the measured values corresponding to the reference value, and downsample to include as many measured values as the number according to the reference value for each item corresponding to each sensor. You can create resampling data sets.

Here, selecting the number of measurement values corresponding to the reference value from a plurality of measurement values according to time for each item corresponding to each sensor is divided into groups corresponding to the reference value and selecting a value that makes the largest triangle in each group. Select using the Largest Triangle Three Buckets (LTTB) technique.

This is to avoid the problem of eliminating the pattern of measured values over time due to the smoothing of measured values when using the average, which is commonly used in downsampling. There is an advantage in reducing the amount of data while maintaining the characteristic pattern of measured values according to .

Specifically, the resampling unit 240 generates groups of the number according to the reference value (S431), and selects the best time among a plurality of measurement values according to time for each item corresponding to each sensor in a plurality of monitoring data included in the divided data set. The corresponding measurement value is included in the first group, the measurement value corresponding to the last time point is included in the last group, and the plurality of measurement values are included in a plurality of groups excluding the first group and the last group among the groups of numbers according to the reference value. A plurality of measurement values, excluding the measurement values corresponding to the best and last times, are uniformly distributed in chronological order, and the plurality of measurement values corresponding to each item included in the split data set are divided into groups corresponding to the reference value in chronological order. It may be distributed (S433), and scores for measurement values included in each group except for the first group and the last group may be calculated (S435).

Here, the resampling unit 240 selects any one measured value included in the first group, which is any one group except for the first group and the last group, in a coordinate area composed of coordinate axes corresponding to a plurality of groups and coordinate axes corresponding to the measured values. When the first coordinate corresponding to , the second coordinate corresponding to the average value of the measurement values included in the second group, which is the group after the first group, and the third coordinate corresponding to the measurement value included in the last group are respectively arranged, The width of the triangle generated by connecting the first to third coordinates may be calculated as a score for the measurement value included in the first group, and a score for each measurement value included in each group may be calculated ( S435).

Thereafter, the resampling unit 240 selects a measurement value having the highest score among measurement values included in each group except for the first group and the last group (S437), and assigns the first group for each item corresponding to each sensor. Resampling data downsampled to consist of the number of measurement values corresponding to the standard value for each item, including the included measurement values, the measurement values selected from each group except for the first and last groups, and the measurement values included in the last group. It may be to create a set (S479).

7 is a number corresponding to a reference value selected by downsampling a plurality of measured values and a plurality of measured values measured over time from a sensor on a coordinate area composed of a coordinate axis corresponding to time and a coordinate axis corresponding to measured values. Referring to FIG. 6 as an example diagram showing the measured values, the first line 610 connecting the measured values selected by the number (eg, 20) corresponding to the reference value using the LTTB technique is a plurality of measured values over time. It can be seen that the characteristic pattern of the second line 620 connecting the measured values is reflected.

That is, the resampling unit 240 does not use the average, which is commonly used in downsampling, but divides into a number of groups corresponding to the reference value and selects a value that makes the largest triangle in each group. LTTB (Largest Triangle Three Buckets) technique Using , it is possible to downsample so that a unique pattern of measured values over time is maintained.

Therefore, by downsampling an excessive amount of measured values than the preset standard, the quality of tool wear prediction due to downsampling is not deteriorated, and the effect of preventing an increase in calculation time due to an excessive amount of data being input to the learning model can be expected. there is.

And, if the number of monitoring data included in the divided data set does not exceed the reference value, the resampling unit 240 corresponds to each sensor among a plurality of measurement values over time corresponding to a plurality of sensors included in the divided data set. Generates a resampling data set that is upsampled to include the number of measurements corresponding to the reference value according to time for each item corresponding to each sensor by applying a one-dimensional nearest-neighbor interpolation to a plurality of measurement values over time for each item to be It may be (S440).

This is in consideration of the characteristics of a sensor that measures a measurement value when a change occurs, and the resampling unit 240 performs upsampling using a one-dimensional nearest neighbor interpolation method to compensate for a measurement value that is missing depending on a communication situation or the like.

In addition, through upsampling, a plurality of measured values according to time have a step function on a time series, so that noise processing is strong.

The normalization unit 250 may normalize the maximum-minimum measurement values included in the resampling data set generated by the resampling unit 240 for each item and generate a normalized data set (S500).

The normalization unit 250 uses the maximum value among the measurement values included in each item and the minimum value included in each item for the measurement value included in the resampling data set, so that the measurement value included in each item has a value between 0 and 1. It may be normalized as much as possible, and a normalized data set obtained by maximum-minimum normalization of a plurality of measurement values included in the resampling data set for each item may be generated (S500).

The normalization unit 250 calculates a normalized value for each of a plurality of measured values belonging to each item included in the resampling data set using the following equation, and calculates a measured value for each item with respect to the measured value of each item. It may be to generate a normalized data set in which a plurality of measured values are maximally-minimum normalized for each item while replacing with one normalized value (S500).

은 X가 속한 항목의 측정값 중 최솟값이며,

는 X가 속한 항목의 측정값 중 최댓값을 의미한다.where X is a measurement value belonging to any one item,

is the minimum value of the measurements of the item to which X belongs,

means the maximum value among the measured values of the item to which X belongs.

Table 2 below shows resampling data as an example to explain normalization.

vibration sensor gyro sensor feed rate sensor 15 0.0366 984 15 0.041 983 12 0.0356 981 17 0.0381 982 19 0.0356 792

Table 3 below shows the normalized data set from Table 2.

vibration sensor gyro sensor feed rate sensor 0.43 0.19 One 0.43 One 0.99 0 0 0.98 0.71 0.46 0.99 One 0 0

Referring to Tables 2 and 3, the normalization unit 250 subtracts the measurement value 15 of the item corresponding to the vibration sensor from 12, the minimum value of the measurement value of the item corresponding to the vibration sensor, and subtracts 12 from 15 to obtain 3 as the vibration sensor. It may be to calculate a normalized value of 0.43 for 15 by dividing by 7, which is the difference between the maximum value of 19 and the minimum value of 12, among the measured values of the item corresponding to .

In addition, the minimum value 0.0356 of the measured values corresponding to the gyro sensor is subtracted from each measured value corresponding to the gyro sensor (0.0366, 0.041, 0.0356, 0.0381, 0.0356), and the difference between the maximum value 0.0381 and the minimum value 0.0356 among the measured values corresponding to the gyro sensor is By dividing by 0.0054, a normalized value normalized to a value between 0 and 1 can be calculated.

The normalization unit 250 may normalize a plurality of measurement values for each item corresponding to each sensor to generate a normal data set composed of normalized values for each item.

According to the configuration described above, a plurality of measured values corresponding to each of a plurality of sensors (vibration sensor, gyro sensor, current sensor of CNC machine tool motor, speed measurement sensor of spindle, and feed rate sensor) and all having different scales. A normalized data set can be created by converting to a normalized normalized value between 0 and 1 with itemwise min-max normalization.

The tool wear predictor 260 may apply the normalized data set generated by the normalizer 250 to a pre-learned learning model to predict tool wear (S600).

Here, the learning model is composed of a plurality of rows corresponding to time and a plurality of columns corresponding to a plurality of sensors, a data set including a plurality of measurement values measured by each of the plurality of sensors according to time, and a data set for each data set. A first classification value corresponding to tool normality or a second classification value corresponding to tool abnormality is input as a supervised learning value, and a data set composed of values corresponding to a plurality of measurement values measured by a plurality of sensors over time is input. It may be learned to receive and output either a first classification value corresponding to tool normality or a second classification value corresponding to tool abnormality.

The second classification value may mean that the tool is worn out.

The learning model may have a convolutional neural network structure. Specifically, the learning model may be a 1d convolutional layer, for example, using Keras' 1D ResNet, but is not limited thereto.

The tool wear prediction unit 260 inputs the normalization data set generated by the normalization unit 250 to a pre-learned learning model, and for the normalization data set, the first classification value corresponding to the normal tool in the learning model (for example, 0) or the second classification value (for example, 1) corresponding to tool abnormality is output, and tool wear can be predicted.

When the tool wear prediction unit 260 predicts tool wear, the communication unit 270 sets a preset stop command code for stopping the operation of the CNC machine tool by the CNC machine tool control module 100 that controls the operation of the CNC machine tool. By sending, it may be to stop the operation of the CNC machine tool.

According to the present invention, tool wear of a machine tool can be predicted using a measurement value measured from a sensor installed in an actual CNC machine tool.

Accordingly, there is an advantage in that it is possible to predict tool wear more objectively than monitoring by field workers, and to prevent unnecessary tool replacement.

In addition, unnecessary monitoring data is removed based on the G-code that controls the CNC machine tool, and upsampling or downsampling is performed based on a preset sampling rate to prevent an increase in the computation time of the learning model and reduce wear prediction accuracy. don't let

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

10: Tool wear prediction system for CNC machine tools based on artificial intelligence
100: CNC machine tool control module
200: Tool wear prediction device for CNC machine tools based on artificial intelligence
210: data collection unit
220: data division
230: time series equalization unit
240: resampling unit
250: normalization unit
260: tool wear prediction unit
270: Ministry of Communication

Claims (10)

The tool counter value according to the number of driving of the CNC machine tool input from the outside, the plurality of measured values corresponding to each of the plurality of sensors measured from the plurality of sensors attached to the CNC machine tool, and the CNC machine tool according to the user's input Storing the monitoring data including the G-code transmitted to the machine according to time;
Generating a divided data set composed of a plurality of monitoring data having the same tool counter value by dividing a plurality of monitoring data stored according to time based on a tool counter value;
removing monitoring data including a G code corresponding to a predetermined removal code among a plurality of pieces of monitoring data included in the divided data set, and generating a uniform data set having a uniform time series;
generating a resampling data set including the number of measurement values according to the sampling rate for each item by resampling a plurality of measurement values for each item corresponding to each sensor included in the uniform data set based on a predetermined sampling rate;
generating a normalized data set by maximum-minimum normalizing the values included in the resampling data set for each item; and
Applying the normalized data set to a pre-learned learning model to predict tool wear; artificial intelligence-based tool wear prediction method of a CNC machine tool comprising the. According to claim 1,
If tool wear is predicted in the step of predicting tool wear,
After the step of predicting whether or not the tool is worn,
Transmitting a preset stop command code for stopping the operation of the CNC machine tool to a CNC machine tool control module that controls the operation of the CNC machine tool, so that the operation of the CNC machine tool is stopped; artificial intelligence-based including A tool wear prediction method for CNC machine tools. According to claim 1,
The step of generating the resampling data set is
Calculating a reference value by multiplying a work time according to a difference between a first time point and a last time point among a plurality of time points corresponding to a plurality of monitoring data included in the divided data set by a preset sampling rate;
Comparing whether the number of monitoring data included in the divided data set exceeds the reference value;
When the number of monitoring data included in the split data set exceeds the reference value, a plurality of measurement values over time for each item corresponding to each sensor included in the split data set are selected as many as the number according to the reference value, and Generating a downsampled resampling data set to include as many selected measurement values as the number according to the reference value
A tool wear prediction method for CNC machine tools based on artificial intelligence. The tool counter value according to the number of driving of the CNC machine tool from the CNC machine tool control module that controls the operation of the CNC machine tool, and the plurality of measurements corresponding to each of the plurality of sensors measured from the plurality of sensors attached to the CNC machine tool a data collection unit for receiving monitoring data including a value and a G-code transmitted to a CNC machine tool according to a user's input and storing the data according to time;
a data division unit configured to divide a plurality of monitoring data stored according to time based on a tool counter value to generate a divided data set composed of a plurality of monitoring data having the same tool counter value;
a time series smoothing unit removing monitoring data including a G code corresponding to a predetermined removal code among a plurality of pieces of monitoring data included in the split data set, and generating a uniform data set in which time series are uniform;
Based on a predetermined sampling rate, resampling a plurality of measurement values for each item corresponding to each sensor included in the uniform data set to generate a resampling data set including the number of measurement values according to the sampling rate for each item resampling unit;
a normalization unit for maximum-minimum normalization of values included in the resampling data set for each item, and generating a normalization data set; and
A tool wear prediction device for an artificial intelligence-based CNC machine tool comprising a; tool wear prediction unit for predicting tool wear by applying the normalized data set to a pre-learned learning model. According to claim 4,
If the tool wear predictor predicts tool wear, a predetermined stop command code for stopping the operation of the CNC machine tool is transmitted to the CNC machine tool control module that controls the operation of the CNC machine tool, and the operation of the CNC machine tool is performed. Tool wear prediction device for artificial intelligence-based CNC machine tools further comprising a; communication unit to stop this. According to claim 4,
The resampling unit
A reference value is calculated by multiplying the work time according to the difference between the earliest time and the last time among the plurality of time points corresponding to the plurality of monitoring data included in the divided data set by the preset sampling rate, and It compares whether the number of monitoring data exceeds the reference value, and if the number of monitoring data included in the divided data set exceeds the reference value, a plurality of monitoring data according to time for each item corresponding to each sensor included in the divided data set is compared. Selecting as many measured values as the number according to the reference value and generating a downsampled resampling data set to include as many measured values as the number according to the reference value for each item.
A tool wear prediction device for CNC machine tools based on artificial intelligence. The G-code input from the outside is transferred to the CNC machine tool to control the operation of the CNC machine tool, and the tool counter value according to the number of driving of the CNC machine tool and the measured value from a plurality of sensors attached to the CNC machine tool A CNC machine tool control module that externally transmits monitoring data including a plurality of measured values corresponding to a plurality of sensors and a G-code transmitted to the CNC machine tool; and
A divided data set composed of a plurality of monitoring data having the same tool counter value by receiving monitoring data from the CNC machine tool control module and storing it according to time, and dividing the plurality of monitoring data stored according to time based on the tool counter value. Creates a uniform data set with a uniform time series by removing monitoring data including a G code corresponding to a preset removal code among a plurality of monitoring data included in the divided data set, and based on a preset sampling rate resampling a plurality of measurement values for each item corresponding to each sensor included in the uniform data set to generate a resampling data set including the number of measurement values according to the sampling rate for each item, and included in the resampling data set An artificial intelligence-based CNC machine tool including a tool wear prediction device that normalizes the maximum-minimum value for each item, generates a normalized data set, and predicts tool wear by applying the normalized data set to a pre-learned learning model. tool wear prediction system. According to claim 7,
The tool wear prediction device
By inputting the normalized data set into a pre-trained learning model, when tool wear is predicted, a predetermined stop command code for stopping the operation of the CNC machine tool is transmitted to the CNC machine tool control module,
The CNC machine tool control module
Controlling the operation of the CNC machine tool to stop when the stop command code is input
A tool wear prediction system for CNC machine tools based on artificial intelligence. According to claim 7
The tool wear prediction device
A data collection unit that receives monitoring data from the CNC machine tool control module and stores it according to time;
a data division unit configured to divide the plurality of monitoring data stored according to time in the data collection unit based on a tool counter value to generate a divided data set composed of a plurality of monitoring data having the same tool counter value;
A time series smoothing unit for removing monitoring data including a G code corresponding to a preset removal code among a plurality of pieces of monitoring data included in the split data set and generating a uniform data set in which time series are uniform;
A resampling unit resampling a plurality of measurement values for each item corresponding to each sensor included in the uniform data set based on a predetermined sampling rate to generate a resampling data set including the number of measurement values according to the sampling rate for each item and,
a normalization unit for maximum-minimum normalization of values included in the resampling data set for each item, and generating a normalization data set;
And a tool wear predictor for predicting tool wear by applying the normalized data set to a learning model pre-learned to predict tool wear by receiving a data set consisting of a plurality of measured values over time.
A tool wear prediction system for CNC machine tools based on artificial intelligence. According to claim 9,
The resampling unit
A reference value is calculated by multiplying the work time according to the difference between the earliest time and the last time among the plurality of time points corresponding to the plurality of monitoring data included in the divided data set by the preset sampling rate, and It compares whether the number of monitoring data exceeds the reference value, and if the number of monitoring data included in the divided data set exceeds the reference value, a plurality of monitoring data according to time for each item corresponding to each sensor included in the divided data set is compared. Selecting as many measured values as the number according to the reference value and generating a downsampled resampling data set to include as many measured values as the number according to the reference value for each item.
A tool wear prediction system for CNC machine tools based on artificial intelligence.

Images