KR102550417B1 - Apparatus and method for notifying calibration alarms for cnc machine device - Google Patents

Abstract

The present invention relates to a tool calibration alarm device and method for automatically alarming the calibration timing of a machine tool tool. A data collection unit that collects data about, a data analysis unit that learns the data collected through the data collection unit by an AI algorithm to derive an AI prediction value for predicting tool wear, and an AI prediction value derived through the data analysis unit , It includes a tool correction point detection unit that detects the tool correction time point while comparing it with data collected while applying to the actual process, and a notification unit that displays and informs the tool correction point detected by the tool correction point detection unit through a visual or audible method.

Description

Tool calibration alarm device and method for machine tools {APPARATUS AND METHOD FOR NOTIFYING CALIBRATION ALARMS FOR CNC MACHINE DEVICE}

The present invention relates to a tool calibration alarm device and method for automatically alarming the calibration timing of a tool for a machine tool.

In general, when processing a workpiece using a machine tool, no matter how precise the position control of the workpiece is, the length and diameter of the tool (processing tip) to be processed are inaccurate or deformed by the heat generated during processing, making accurate processing impossible. do. In addition, as the machining continues, the tool itself wears out and the outer diameter of the workpiece becomes larger than the actual size, resulting in a further decrease in precision. Therefore, there is a need to compensate for an error in the position of a tool before or during processing or according to a request of an operator.

1 is a view showing a calibration process of a tool for a conventional machine tool.

As shown in FIG. 1, according to the conventional calibration process, processing equipment data is collected from machine tool equipment, but managed and controlled with values based on experience. The data only searches the history and solves the problem by subjectively interpreting the problem.

2 is a graph showing the outer diameter of a workpiece and the wear amount of a tool according to the product quantity as a result of operation by a conventional process.

Looking at the graph on the left, it is a graph in which the outer diameter of the workpiece changes according to the number of products produced. That is, it can be seen that the outer diameter of the workpiece gradually increases as the number of manufactured products increases. However, the sections where the outer diameter of the workpiece gradually increases and then rapidly decreases, that is, sections A, B, and C, are sections in which the tool is calibrated manually by workers. As shown, when the error of the tool is corrected, the outer diameter of the workpiece is reduced to the actual size, but as the number of products is accumulated, the outer diameter of the workpiece increases again. You can see that it gradually changes to a steep slope.

The graph on the right shows the amount of wear of machining tips (tools) according to the cumulative number of jobs. It can be seen that the wear amount of the machining TIP (tool) gradually increases as the cumulative number of operations increases.

As such, in the prior art, when the quantity of produced products reaches a certain quantity by using the characteristic that the wear amount of tools increases as the quantity of manufactured goods increases, the tools are calibrated through the operator's experience or visual confirmation. At this time, even if the quantity of produced products reaches a certain quantity, it is found through the operator's experience or visual confirmation because there is no alarm system in the conventional CNC itself.

In other words, the conventional calibration process is not managed and controlled by an automatic system, but is controlled only by relying on the naked eye during patrol activities by managers and field workers. It can cause huge losses in terms of maintenance.

Therefore, in the prior art, there is a need for a process that enables more objective and scientific decision-making, rather than setting and interpreting based on experience.

Republic of Korea Registered Utility Model Publication No. 20-0149798 (Announced on June 15, 1999)

The present invention is to achieve the above needs, predicting the amount of tool wear using a spindle load and the quantity of products produced during processing through a machine tool, and notifying the correction time of the tool for the machine tool with an alarm. An object of the present invention is to provide a correction device and method for a tool for a machine tool that informs field workers of an accurate correction time.

In particular, the present invention is aimed at zeroing quality problems and maximizing productivity improvement by applying AI modeling that can be predicted similarly to the pattern of actual tool calibration by learning the spindle load of a machine tool according to the quantity of products produced by an AI algorithm is doing

To this end, the calibration alarm device for a tool for a machine tool according to an embodiment of the present invention collects data on the spindle load value generated by the operation of the CNC control unit and the quantity of products processed by the operation of the CNC control unit. collection department; a data analysis unit learning the data collected through the data collection unit by an AI algorithm to derive an AI prediction value for predicting tool wear; a tool correction point detector for detecting a tool correction point while comparing the predicted AI value derived through the data analysis unit with data collected while applying the data to an actual process; and a notification unit for displaying and notifying the tool correction time point detected by the tool correction point detection unit through a visual or audible method, wherein the AI prediction value derived through the data analysis unit is similar to a spindle load value pattern appearing in an actual process. The tool correction time point detection unit compares the AI prediction value derived through the data analysis unit with data collected while applying to the actual process to predict the tool correction time point, and the derived AI prediction value and the actual process While comparing patterns of spindle load values appearing in , it is possible to detect a tool correction point immediately before the point at which the derived AI predictive value rapidly decreases.

Meanwhile, a correction alarm method for a tool for a machine tool according to an embodiment of the present invention is a method for a correction alarm device for a tool for a machine tool, wherein the data collection unit of the device measures the spindle load value generated by the operation of the CNC control unit and Collecting data on the quantity of products processed by the operation of the CNC control unit; Deriving, by the data analysis unit of the device, an AI prediction value for predicting tool wear by learning the data collected through the data collection step by an AI algorithm; Detecting, by a tool correction point detection unit of the device, a tool correction point of time while comparing the AI prediction value derived through the analyzing step with data collected while applying it to an actual process; and notifying, by a notification unit of the device, the tool correction time point detected through the step of detecting the tool correction time point, expressed in a visual or auditory method, and the AI predicted value derived through the data analysis unit is actually It corresponds similarly to the pattern of the spindle load value appearing in the process, and the step of detecting the tool correction time point compares the AI prediction value derived through the analysis step with the data collected while applying it to the actual process to correct the tool. The time point is predicted, and the tool correction time point is detected immediately before the point at which the derived AI prediction value rapidly decreases while comparing the derived AI prediction value with the pattern of the spindle load value appearing in the actual process.

According to the present invention, the correction time of the tool is predicted by referring to the data learned through AI modeling for the tool for the machine tool and automatically alarmed to notify the field worker of the correct correction time so as not to miss the correction time of the tool It works.

In particular, the present invention allows decision-making to be made by an automatic system, not managed by the operator's experience or the naked eye, and moreover, the tool's correction point, which may vary depending on the actual state of the tool, can be predicted through real-time modeling. Therefore, there is a remarkable effect of alarming a more accurate correction time than before.

As a result, the present invention can improve process management of machine tools, thereby improving productivity, and, above all, zeroing out quality problems to maximize precision.

1 is a view showing a calibration processor of a tool for a conventional machine tool.
2 is a graph showing the outer diameter of a workpiece and the wear amount of a tool according to the product quantity as a result of operation by a conventional process.
3 is a diagram showing a calibration alarm device for a tool for a machine tool according to an embodiment of the present invention.
Figure 4 is a detailed view of Figure 3;
FIG. 5 is a graph showing spindle speed and load collected by the data collection unit of FIG. 3 .
FIG. 6 is a graph visualized by normalizing the spindle speed and load in the data normalization unit of FIG. 4 .
7 is a diagram showing the RNN basic structure in the AI modeling unit of FIG. 4 .
8 is a diagram showing the LSTM basic structure in the AI modeling unit of FIG. 4.
9 is a graph illustrating predicted values learned through the AI modeling unit of FIG. 4 .
10 is a diagram showing a correction time point detected from an actual tool load versus an AI prediction value.
11 is a flowchart for explaining a calibration alarm method for a tool for a machine tool according to an embodiment of the present invention.

Hereinafter, specific embodiments of the present invention will be described in detail with reference to the accompanying drawings.

3 is a view showing a correction alarm device for a tool for a machine tool according to an embodiment of the present invention, and FIG. 4 is a detailed view of FIG. 3 .

First, referring to FIG. 3 , the correction alarm device for a tool for a machine tool according to an embodiment of the present invention includes a data collection unit 120, a data analysis unit 130, a tool calibration point detection unit 140, and a notification unit 150. includes

The data collection unit 120 collects data on a spindle load value generated by the operation of the CNC controller 110 and the quantity of products processed by the operation of the CNC controller 110 .

Here, the CNC control unit 110 means a functional unit involved in the overall operation of the machine tool in the machining process using the machine tool. The CNC controller 110 refers to control for operating a lathe for seating a workpiece, a spindle and a motor for rotating the lathe, and a tool (processing TIP) for processing a workpiece seated on the lathe.

That is, when the machine tool operates, data on the spindle load generated by the operation of the machine tool and the number of finished products can be obtained from the CNC control unit 110 by computerized numerical control (CNC). .

The data analysis unit 130 analyzes the data collected through the data collection unit 120 to derive an AI prediction value for predicting the amount of tool wear.

In particular, the data analysis unit 130 according to an embodiment of the present invention is characterized by predicting the amount of tool wear by learning data on the spindle load value and the quantity of products produced by AI (Artificial Intelligence) algorithm. To this end, the data analysis unit 130 includes modeling learned by the AI algorithm. Its configuration will be described again in detail in FIG. 4 .

The tool correction time point detection unit 140 detects a time point at which the predicted value output through the data analysis unit 130 rapidly decreases as the tool correction time point. At this time, the tool correction time point detection unit 140 may extract the tool correction time point while comparing the data collected while applying to the actual process and the predicted value output through the data analysis unit 130 (see T in FIG. 10 ).

For example, the tool correction time detection unit 140 compares the prediction value output through the data analysis unit with the data collected while applying to the actual process, immediately before the AI prediction value rapidly decreases, and at the same time, the spindle load value rapidly decreases. The point in time can be detected as the tool correction point in time.

The notification unit 150 displays and notifies the tool correction time point detected by the tool correction point detection unit 140 through a visual or auditory method. The visual method is to display a guide message requesting tool calibration on the screen, which can also include a method of displaying it on the screen of the machine tool or a separate screen, or by transmitting a message to the operator's terminal if necessary. there is. The auditory method includes a method of expressing through sound reproduction or the like.

With this configuration, the tool calibration alarm device for a machine tool according to an embodiment of the present invention detects a tool calibration time in real time through modeling in response to an actual process, and informs the operator of an accurate calibration time. Then, when tool calibration is performed by a worker or by an automatic system, the CNC control unit 110 updates and operates the correction data, and the data collection unit 130 collects operating data based on the correction data again.

FIG. 4 is a detailed view of the data analysis unit 130 of FIG. 3 .

The data analysis unit 130 includes a data design unit 131, a data purification unit 132, and an AI modeling unit 133.

The data design unit 131 defines data collected by the data collection unit and combines, integrates, or separates the collected data to facilitate analysis.

The data design unit 131 must first clearly define source data. For example, the spindle load value is the load generated by the equipment that rotates the tool, the product quantity means the production quantity per hour, and the spindle speed is the number of revolutions per minute of the equipment that rotates the tool. can be defined

The data refiner 132 reduces and integrates only the actual processing section in which actual processing is performed from the data defined in the data design unit 131 to normalize the data.

If the original data is used as it is, the composition of the data is inconsistent, making it impossible to perform efficient analysis in the analysis process. Therefore, data cleaning is a process of normalizing data so that data analysis can be performed efficiently.

The normalization process can be divided into the following three steps.

First, reduce and integrate data. The original data is a data set in which actual machining is performed when the workpiece and the tool come into contact with each other, and a data set in which operations unrelated to machining, such as loading, unloading, and cleaning, are mixed together. Here, data sets between tasks unrelated to machining are discarded and only data closely related to actual machining are reduced.

For example, referring to FIG. 5 , the graph of FIG. 5 is data obtained by collecting spindle load value (load) and spindle speed (speed) during actual machining. The spindle load value shows almost no change and then rises slightly within the Q circle. According to the data of spindle speed, machining starts when the spindle speed suddenly decreases, and machining ends when the spindle speed suddenly increases.

Therefore, in the first step of the normalization process, only the actual machining section (P) in which actual machining is performed can be reduced and integrated and refined as shown in FIG. 6 (a).

Second, it corrects outliers and noise data by converting the reduced or integrated data.

In this process, each data is normalized using the Z-Score Normalization formula, data that are too far from the mean (eg z≥3) are removed, and data that looks like noise but are considered important are removed. Normalize through calibration.

Third, in order to utilize efficient analysis, each normalized data is transformed by applying the sliding window technique once again. The sliding window technique is a useful technique for analyzing and learning time series data.

Wholeset X={x1, x2, x3, … , xi}, the subset s={x1, x2, x3, ... , xn}, the sliding window is composed of the same fixed number of elements and stored. At this time, if there is newly arriving data after the memory is full during the saving process, the oldest information is removed from the window, and the new data information is saved by adding the latest element.

FIG. 6 is a graph visualized by normalizing the spindle speed and load in the data normalization unit of FIG. 4 .

Looking at the graph (c) of FIG. 6, the pattern of the spindle load value slightly increases over time, then decreases significantly in a short time at an arbitrary point in time, and then gradually increases slightly again. It is learned so that it can be recognized in the unit 133.

Referring back to FIG. 4 , the AI modeling unit 133 predicts the amount of tool wear by learning normalized data through AI modeling through the data refinement unit. At this time, the AI modeling unit 133 learns using an algorithm using a long short-term memory (LSTM) technique.

7 is a diagram showing the RNN basic structure of AI modeling, and FIG. 8 is a diagram showing the LSTM basic structure of AI modeling.

LSTM is one of the RNN techniques, but LSTM has the ability to preserve long-term memory, making up for the shortcomings of RNN techniques with short-term memory preservation ability.

A typical RNN technique processes inputs (x1, x2, ..., xt) in a sequential manner in learning as shown in FIG. The sequential method means that the information of the previous step is received when calculating the output (y1, y2, ..., yt) of the current step. In this way, it is connected like a chain so that the input of the current stage carries its own information throughout the next stage. However, in learning, since the RNN technique has a gradient vanishing (convergence to 0) problem and a gradient exploding problem, a mechanism to forget old data and remember the latest data as time passes, as shown in FIG. has been introduced

In the embodiment of the present invention, the activation function was changed and applied without using the LSTM technique as it is. That is, the Relu function was used instead of the sigmoid function used in the basic LSTM. If the median value is not 0, not only does it take a long time to learn, but it is inefficient because the learning data diverges like a zigzag because the sign is determined by the input data.

Therefore, the LSTM model based on the Relu function according to an embodiment of the present invention can be expressed by the following formula.

The above formula can be solved as follows.

At this time, ct and ht can be obtained by the following formula.

Therefore, by applying the above LSTM model, the pattern of tool compensation management is predicted with learning data and verified with test data.

The AI modeling unit 133 according to an embodiment of the present invention defined the number of nodes of the LSTM model as 8 and divided the given data into the training set and the test set in a ratio of 8: 2 to perform 500 learnings (see the table below). ).

9 is a graph illustrating predicted values learned through the AI modeling unit of FIG. 4 .

The blue graph is a predicted value derived through learning by the AI modeling unit 133, and the orange graph is a value representing a tool calibration test set. It can be seen that the patterns of the two graphs are actually similar and the performance of the AI algorithm is excellent. That is, the predicted AI value derived through the learning of the AI modeling unit 133 corresponds similarly to the pattern of the spindle load value appearing in the actual process.

10 is a diagram showing a correction time point extracted from actual tool load versus AI prediction value.

That is, FIG. 10 is a graph contrasted with CNC spindle load value data additionally collected while applying an algorithm through AI modeling to an actual process. The orange graph is the actual data, and the blue data is the AI prediction value derived through AI modeling. Although the two graphs fluctuate up and down, it can be seen that the AI predicted value (blue) corresponds similarly to the pattern of the actual data (orange). That is, it can be confirmed that the pattern of the real data is well followed.

T indicated in FIG. 10 is a time point at which the predicted AI value rapidly decreases, and becomes a tool correction time point. The tool correction time point is detected through the tool correction time point detection unit (140 in FIG. 3) described above, and an alarm for notifying this is output from the notification unit (150 in FIG. 3).

As such, since the spindle load value generated during the actual process appears irregularly as if it fluctuates up and down as shown in FIG. 10, the embodiment of the present invention applied the AI model to find the pattern of the spindle load value, By comparing the pattern of the derived AI predicted value and the spindle load value, the tool correction point was found just before the point at which the derived AI predicted value rapidly decreased.

Next, with reference to FIG. 11, a correction alarm method for a tool for a machine tool according to an embodiment of the present invention will be described.

First, in the data collection step (S100), the data collection unit of the device according to the embodiment of the present invention collects data on the spindle load value and the quantity of products generated by the operation of the CNC control unit.

In the next data analysis step, the data analysis unit of the device analyzes the data collected through the data collection step and predicts the tool wear amount by learning by an algorithm.

This data analysis step specifically includes a data design step (S110), a data refinement step (S120), and an AI modeling step (S130).

Specifically, in the data design step (S110), the data design unit of the device defines the data collected in the data collection step (S100) and combines, integrates, or separates the collected data to facilitate analysis.

As the next data refinement step (S120), the data refinement unit of the device normalizes by reducing and integrating only the actual processing section in which actual processing is performed from the data defined in the data design step (S110). The normalization process can apply the three operations described above.

In the next AI modeling step (S130), the AI modeling unit of the device learns the data normalized through the data refinement step (S120) by AI modeling to predict the amount of tool wear.

In the next tool correction point detection step ( S140 ), the tool correction point detection unit of the device detects a point in time when the predicted value derived through the AI modeling step ( S130 ) rapidly decreases as the tool correction point in time.

At the time of detection, the tool calibration point detection unit of the device compares the predicted value derived through the AI modeling step (S130) with the data collected while applying it to the actual process, and at the same time, the predicted value rapidly decreases, and at the same time, the spindle load value rapidly increases. The point of decrease is detected as the tool compensation point.

In the next alarm step (S150), the notification unit of the device displays and informs the tool correction time point detected through the tool correction point detection step (S140) through a visual or auditory method.

The above description is merely illustrative of the present invention, and various modifications may be made by those skilled in the art without departing from the technical spirit of the present invention. Therefore, the embodiments disclosed in the specification of the present invention are not intended to limit the present invention. The scope of the present invention should be construed by the claims below, and various equivalents and modifications that can replace them at the time of this application should be construed as being included in the scope of the present invention.

120: data collection unit 130: data analysis unit
131: data design unit 132: data refinement unit
133: AI modeling unit 140: tool correction point detection unit
150: notification unit

Claims (8)

a data collection unit that collects data on a spindle load value generated by the operation of the CNC control unit and the quantity of products processed by the operation of the CNC control unit;
a data analysis unit learning the data collected through the data collection unit by an AI algorithm to derive an AI prediction value for predicting tool wear;
a tool correction point detector for detecting a tool correction point while comparing the predicted AI value derived through the data analysis unit with data collected while applying the data to an actual process; and
Including; a notification unit for displaying and notifying the tool correction point detected by the tool correction point detection unit in a visual or auditory manner,
The AI prediction value derived through the data analysis unit appears in correspondence with the pattern of the spindle load value appearing in the actual process,
The tool correction point detection unit,
The tool correction time is predicted by comparing the AI prediction value derived through the data analysis unit with data collected while applying it to the actual process, and the derived AI prediction value is compared with the pattern of the spindle load value appearing in the actual process. A correction alarm device for a tool for a machine tool, characterized in that for detecting the tool correction point just before the point at which the predicted AI value rapidly decreases. delete delete According to claim 1,
The data analysis unit,
a data design unit that defines the data collected by the data collection unit and combines, integrates, or separates the collected data to facilitate analysis;
a data refinement unit for normalizing by reducing and integrating only the actual processing section in which actual processing is performed from the data defined in the data design unit;
an AI modeling unit learning the data normalized through the data refinement unit through AI modeling to derive an AI prediction value for predicting tool wear;
Calibration alarm device for a tool for a machine tool comprising a. According to claim 4,
The AI modeling unit,
A calibration alarm device for a tool for a machine tool, characterized in that it is learned using an algorithm using a long short-term memory (LSTM) technique. As a method of a calibration alarm device for a tool for a machine tool,
Collecting, by the data collection unit of the device, data related to a spindle load value generated by an operation of the CNC control unit and a quantity of products processed by the operation of the CNC control unit;
Deriving, by the data analysis unit of the device, an AI prediction value for predicting tool wear by learning the data collected through the data collection step by an AI algorithm;
Detecting, by a tool correction point detection unit of the device, a tool correction point of time while comparing the AI prediction value derived through the analyzing step with data collected while applying it to an actual process; and
Including; displaying and notifying, by a notification unit of the device, the tool correction time point detected through the step of detecting the tool correction point by a visual or auditory method,
The AI prediction value derived through the data analysis unit appears in correspondence with the pattern of the spindle load value appearing in the actual process,
The step of detecting the tool correction time point,
The AI prediction value derived through the analysis step is compared with data collected while applying to the actual process to predict the tool correction time, and while comparing the derived AI prediction value with the pattern of the spindle load value appearing in the actual process, the A correction alarm method for a tool for a machine tool, characterized in that the tool correction point is detected immediately before the point at which the derived AI prediction value rapidly decreases. According to claim 6,
The step of predicting the amount of tool wear,
a data design step of defining the data collected in the data collection step by the data design unit of the device and combining, integrating, or separating the collected data to facilitate analysis;
a data refinement step of normalizing by reducing and integrating only the actual processing section in which actual processing is performed from the data defined in the data design step by the data refining unit of the device; and
An AI modeling step of deriving an AI prediction value for predicting tool wear amount by learning the data normalized through the data refinement step by AI modeling by the AI modeling unit of the device;
Compensation alarm method of a tool for a machine tool comprising a. delete

Images