Cutter damage detection method based on power detection

Eigenvalues ​​are extracted through Hall sensor and wavelet algorithm, combined with Savitzky-Golay filtering algorithm to eliminate noise, and a tool damage detection model is constructed, which solves the problems of tool monitoring data imbalance and noise interference in the existing technology, and realizes the unattended equipment The accurate inspection and timely alarm have improved production efficiency and processing quality.

CN119017137APending Publication Date: 2024-11-26QINGDAO YUNDING SOFTWARE DEV CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411120111.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-15
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The existing tool monitoring methods have problems such as data imbalance, noise interference and high cost, resulting in low accuracy of tool damage monitoring and unattended real-time detection.

Method used

The Hall sensor is used to collect the three-phase current signal of the spindle UVW, extract the characteristic values ​​through the DB4 wavelet algorithm, and use the Savitzky-Golay filtering algorithm to eliminate noise, and build a tool damage detection model to achieve real-time detection.

Benefits of technology

It improves the accuracy and stability of tool damage detection, reduces production costs, ensures processing quality and safe production, and realizes accurate detection and timely alarms of unattended equipment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119017137A_ABST
    Figure CN119017137A_ABST
Patent Text Reader

Abstract

The invention provides a tool damage detection method based on power detection, and belongs to the technical field of tool wear detection. Firstly, a three-phase current signal of a main shaft is collected by a Hall sensor, and equivalent direct current of main shaft operation is obtained through calculation; secondly, carrying out feature extraction by using a DB4 wavelet algorithm, and extracting a feature value, namely a damaged root-mean-square R, reflecting the current change in the machining process; performing curve smoothing processing on the extracted damaged root-mean-square R by adopting a filtering algorithm, eliminating noise and fluctuation in the data, retaining original characteristics and variation trend of the data, and obtaining fitted curve data as model input; fitting curve data obtained in real time are input into the constructed cutter damage detection model, and the output label 0 represents that the label 1 is normally output and represents that the cutter is damaged. According to the method, more accurate cutter damage detection on the cutter of the unattended equipment can be realized, the accuracy and the stability of the equipment are improved, the production efficiency of an enterprise can be improved, and the cost is reduced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of tool wear detection, and in particular relates to a tool breakage detection method based on power detection. Background Art

[0002] With the development of industrial automation and intelligence, automated production lines have been widely used in various fields. It is no longer one person guarding one device or one person guarding several devices, but one person is responsible for one or several production lines, and automatic processing is carried out through automation and information technology. In the process of machine tool processing, the condition of the tool is an important influencing factor in the process of machine tool processing. When the tool is damaged, it will affect the machine tool processing, resulting in a decrease in the qualified rate of the workpiece. What's worse, it may cause damage to the machine tool spindle and damage to the equipment. Therefore, how to effectively realize tool status monitoring is a key factor in improving production efficiency, reducing tool costs and processing quality.

[0003] At present, tool monitoring methods are generally divided into two categories, namely direct measurement and indirect measurement. The direct measurement method usually determines the current state of the tool by directly measuring the changes in the appearance and geometric shape of the blade. This method is simple and direct, but it cannot be measured during processing. It is necessary to wait until the current processing is completed before executing the corresponding measurement program. The indirect measurement method indirectly reflects the current tool status information by monitoring the changes in certain state signals of the tool during the processing process. Indirect monitoring methods include power signal detection, vibration signal detection, cutting force detection, and cutting temperature detection. The use of indirect monitoring methods will lead to uneven data. The time of each stage of the machine tool tool from the beginning of processing to the breakage and failure is uneven, which will lead to differences in the amount of data, which will affect the accuracy of the learning model and reduce the accuracy of tool breakage monitoring. In addition, the signals collected by the sensor not only have a lot of noise interference, but the excessive number of sensors collected will increase the cost and the difficulty of feature extraction. Summary of the invention

[0004] In view of the above problems, the present invention provides a tool breakage detection method based on power detection, which is characterized by comprising the following processes:

[0005] S1, using the Hall sensor to collect the spindle UVW three-phase current signal, and calculate the spindle three-phase AC current to obtain the equivalent DC current of the spindle operation;

[0006] S2, using DB4 wavelet algorithm for feature extraction; decomposing and reconstructing the equivalent DC current signal, the signal is decomposed layer by layer into wavelet coefficients of different frequency components, and then the original signal is restored through the reconstruction process, so as to extract the characteristic value reflecting the current change during the machining process, that is, the damage root mean square R;

[0007] S3, using Savitzky-Golay filtering algorithm to smooth the curve of the damage root mean square R extracted during the processing, eliminating the noise and fluctuation in the data, retaining the original characteristics and change trend of the data, and obtaining the fitting curve data as the model input;

[0008] S4, input the fitting curve data obtained in real time into the constructed tool breakage detection model; the tool breakage detection model obtains the upper and lower edge curve values ​​based on the tool processing state reference curve self-learning and tool breakage curve learning, and then combines the breakage coefficient to form the final upper and lower threshold curves. If the real-time value falls within the range of the upper and lower thresholds, it will output label 0, indicating normal; once the value exceeds this range, it will output label 1, indicating that the tool is broken.

[0009] Preferably, the specific process of S2 is:

[0010] First, the continuous wavelet transform of the equivalent DC current x(t) of the machine tool spindle is defined by convolution as:

[0011]

[0012] Where t is the displacement; a is the scale factor; b is the translation parameter; is the mother wavelet; after scaling and translating the mother wavelet, a wavelet sequence is obtained:

[0013]

[0014] For any signal f(t)∈L 2 (R), the continuous wavelet transform is:

[0015]

[0016] The reconstruction formula is:

[0017]

[0018] The damage root mean square R is extracted as the characteristic value reflecting the current change during the machining process:

[0019]

[0020] Preferably, the specific process of S3 is:

[0021] First, a sliding window with R(i) as the center and containing 2M+1 data points is set. The size of the window can be adjusted according to the characteristics of the data and processing requirements;

[0022] Then, the data in the window is used as a local data segment, and a p-order polynomial q(n) is constructed to fit the array:

[0023]

[0024] Among them, a0, a1, …, a m is the fitting coefficient, and the residual C is obtained through least square fitting:

[0025]

[0026] When the residual C is the smallest, the filtering effect is best. During the fitting process, a best fitting curve Q(t) will be found. This curve can reflect the characteristics of the local data segment to the greatest extent. This curve will be used as the input of the model to determine whether the tool is damaged.

[0027] Preferably, the tool breakage detection model in S4 is based on the collected raw data, and features are extracted through the DB4 wavelet algorithm and the extracted breakage root mean square R is smoothed by the Savitzky-Golay filtering algorithm to obtain standard model input data; the raw data is collected through a Hall sensor, the collection frequency is set to 10Khz, and the collection is carried out after the spindle speed tends to be stable, and the three-phase current signal is read at one time through the IO interrupt.

[0028] Preferably, the calculation of the three-phase AC current of the main shaft in S1 is a root mean square calculation:

[0029]

[0030] In the formula, I U ,I V ,I W Represent the current values ​​of the three-phase currents respectively.

[0031] Preferably, the self-learning of the tool processing state reference curve of the tool breakage detection model is to learn and store the first three processing processes when the machine tool is running normally, generate three learning curves Q1(t), Q2(t), and Q3(t), use them as inputs of the tool breakage detection model, and count the upper limit value and the lower limit value of each time point respectively, and finally form two curves:

[0032]

[0033]

[0034] Where: [ti] represents a certain time point of data collection during the tool processing process; [up] represents all the upper limit points during the tool processing process at the time [ti], and conversely [down] represents all the lower limit points during the tool processing process at the time [ti]; and They represent the values ​​of the upper and lower curves obtained after three learnings at time [ti] respectively;

[0035] According to the upper and lower edge curve values ​​obtained by learning, combined with the breakage coefficient, the final upper and lower threshold curves are formed, that is, the tool breakage curve:

[0036]

[0037]

[0038] Where δ and ω are damage coefficients. The selection of coefficients is based on engineering experience, processing requirements, and the distribution of current data. By adjusting the coefficients, it is ensured that the upper and lower threshold curves are neither too loose to effectively monitor abnormal currents nor too strict to frequently trigger false alarms. The upper and lower threshold curves finally outputted can clearly define the normal fluctuation range of power during the processing process.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] The present invention provides a tool breakage detection method based on power detection, which can realize more accurate tool breakage detection for tools of unattended equipment and issue alarms in time, thereby improving the accuracy and stability of the equipment and the qualified rate of workpieces processed by the equipment, and can help enterprises improve production efficiency, reduce costs, and ensure processing quality and safe production.

[0041] The design uses the DB4 wavelet algorithm to decompose and reconstruct the effective current value signal. In this process, the signal is decomposed layer by layer into wavelet coefficients of different frequency components, and then the original signal is restored through the reconstruction process. In this way, the characteristic values ​​reflecting the current changes during the machining process are extracted. These characteristic values ​​not only contain detailed information on power changes, but also provide strong data support for subsequent analysis of tool breakage.

[0042] The Savitzky-Golay filtering algorithm is used to smooth the damage root mean square R extracted during the processing process. The local data segment is fitted by the least squares method, and the value of each data point is estimated using the fitted function. This method can not only eliminate the noise and fluctuations in the data, but also retain the original characteristics and change trends of the data. Through the continuity and smoothness of the fitting function, the data curve can be smoothed and the reliability of the data can be improved.

[0043] Finally, through the design of a tool breakage detection model TBD, the upper and lower threshold curves obtained by the final output can clearly define the normal fluctuation range of power during the processing process. When the real-time power data exceeds this range, it can be judged that the processing process may have abnormal conditions, so that corresponding measures can be taken in time to adjust and optimize. In this way, not only can the stability of the processing process be improved, but also the production cost can be reduced and the product quality can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a flowchart of the core logic of the present invention;

[0045] Figure 2 It is a schematic diagram of the structure of the tool breakage detection model of the present invention;

[0046] Figure 3 Schematic diagram of a power curve after feature extraction in an embodiment of the present invention;

[0047] Figure 4 Schematic diagram of a power curve after smoothing in an embodiment of the present invention;

[0048] Figure 5 A schematic diagram of reference curve self-learning in an embodiment of the present invention;

[0049] Figure 6 Schematic diagram of a power curve during a normal machining process in an embodiment of the present invention;

[0050] Figure 7 Schematic diagram of power curve of tool breakage during machining in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The present invention is further described below in conjunction with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] In this embodiment, a tool breakage detection method based on power detection is implemented as follows: Figure 1 As shown:

[0053] Power sensor: Selecting a suitable power sensor can monitor the power consumption of the tool in real time during the machining process. Usually, a sensor with high sensitivity and fast response speed is used, such as a Hall effect sensor or a load current sensor.

[0054] Signal acquisition and processing: The data collected by the power sensor is converted into a digital signal using an analog-to-digital converter, and then the signal is filtered, amplified and processed to extract characteristic parameters for subsequent analysis.

[0055] Feature extraction and analysis: The collected power data is extracted and analyzed with the help of signal processing algorithms to identify features related to tool breakage, such as power fluctuation frequency, amplitude changes, etc.

[0056] State recognition and judgment algorithm: Develop an algorithm for identifying and judging the damaged state of the tool, establish a model based on the characteristic parameters, realize automatic recognition of the tool state, and judge whether the tool is in a normal or damaged state.

[0057] Real-time monitoring and feedback system: Establish a real-time monitoring system that can continuously monitor the tool status and provide timely feedback, such as prompting the operator or automatic control system through sound, light or alarm.

[0058] Data processing and storage: Processing and storing the monitored data can achieve data traceability and analysis, providing a basis for equipment performance optimization and fault prevention.

[0059] 1. Data Collection

[0060] For the collection and calculation of current signals, in order to ensure the comprehensiveness and accuracy of the collected data, the collection frequency is set to as high as 10Khz. Such high-frequency collection and real-time calculation not only provide powerful data support for the machine tool processing process, but also provide the possibility to timely discover and deal with potential problems.

[0061] The main control of the acquisition system uses an M33 core microcontroller + 20K LUT fpga, and an external 16MRAM is used to cache real-time data. M33 and fpga use FMC high-speed parallel port bus communication. The analog acquisition chip uses 16-bit high-speed synchronous sampling AD7606 to connect to fpga through the parallel port. After the fpga acquires the analog signal, it triggers M33 to read the data through the IO interrupt, and then puts it into the ring Buff to wait for the host computer to obtain the data.

[0062] In terms of power supply, each power supply uses isolated power supply, and different analog areas and digital grounds are connected by magnetic beads.

[0063] Software: AD sampling information reads 3 current signals at a time through IO interruption, and calculates the effective value of the current in real time. The collected original value and the calculated effective value are stored in RAMbuff waiting for the host computer to read. In order to ensure the stability of communication and the accuracy of data, the system will use a custom protocol to communicate with the host computer to transmit the collected data.

[0064] 2. Data Processing

[0065] Create a UDP socket to receive data, listen to the specified UDP port, and wait for data from the sender. The received data usually exists in the form of a byte stream, which needs to be parsed. According to the protocol specification of the message, the received byte stream is parsed bit by bit to extract the current value of the three-phase current.

[0066] The Hall sensor is used to collect the spindle UVW three-phase current signal. There is a fixed phase difference between the three-phase currents. When the spindle speed tends to be stable, the three-phase current amplitude also tends to be stable. The root mean square calculation of the spindle three-phase AC current is performed to obtain the equivalent DC current of the spindle operation:

[0067]

[0068] In the formula, I U ,I V ,I W Represent the current values ​​of the three-phase currents respectively.

[0069] Monitor the value of DI to determine whether it is currently in the processing state. When DI is 1, combine the parsed information into a record. Write this record in CSV format to a file named by the specified rule. After completing data reception, parsing and storage, close the UDP socket and release related resources.

[0070] 3. Feature Extraction

[0071] Wavelet analysis transforms the signal by selecting appropriate basis functions, extracting the characteristic components of the basis functions at multiple time-frequency scales, and thus describing the changing characteristics of the signal at multiple time-frequency scales. The signal components at high-frequency scales reflect the detailed characteristics of the signal, while the signal components at low-frequency scales reflect the slowly changing characteristics of the signal. In view of the fact that the power signal of the machine tool spindle has multiple frequency components mixed and mainly concentrated in the low-frequency part, the multi-resolution wavelet transform can analyze and extract the signal components reflecting the normal machining process, observe the changing characteristics of the signal from the signal components of multiple machining characteristic frequency bands, and use relevant algorithms to judge the status of the tool.

[0072] The DB4 wavelet algorithm in the Daubechies wavelet family is used to deeply analyze the power changes during the processing process. This algorithm is widely used in the field of signal processing due to its excellent localization characteristics and multi-resolution analysis capabilities.

[0073] First, the continuous wavelet transform of the machine tool spindle current x(t) is defined by convolution as:

[0074]

[0075] Where t is the displacement; a is the scale factor; b is the translation parameter; is the mother wavelet. After scaling and translating the mother wavelet, a wavelet sequence can be obtained:

[0076]

[0077] For any signal f(t)∈L 2 (R), the continuous wavelet transform is:

[0078]

[0079] The reconstruction formula is:

[0080]

[0081] Extract the damage root mean square R as data feature:

[0082]

[0083] The DB4 wavelet algorithm is used to decompose and reconstruct the effective value signal. In this process, the signal is decomposed layer by layer into wavelet coefficients of different frequency components, and then the original signal is restored through the reconstruction process. In this way, the characteristic values ​​reflecting the current changes during the machining process are extracted. These characteristic values ​​not only contain detailed information on power changes, but also provide strong data support for subsequent analysis of tool breakage. Figure 3 Schematic diagram of power curve after feature extraction in this embodiment.

[0084] 4. Data Smoothing

[0085] The Savitzky-Golay filtering algorithm is used to smooth the curve of the damage root mean square R extracted during the machining process. The algorithm fits the local data segment through the least squares method, and then uses the fitted function to estimate the value of each data point. This method can not only eliminate the noise and fluctuations in the data, but also retain the original characteristics and change trends of the data.

[0086] When applying the Savitzky-Golay filtering algorithm, first set a sliding window centered at R(i) and containing 2M+1 data points. The size of the window can be adjusted according to the characteristics of the data and processing requirements.

[0087] Then, the data in the window is used as a local data segment, and a p-order polynomial q(n) is constructed to fit the array:

[0088]

[0089] Among them, a0, a1, …, a mis the fitting coefficient, and the residual C is obtained through least square fitting:

[0090]

[0091] When the residual C is the smallest, the filtering effect is best. During the fitting process, a best fitting curve Q(t) will be found. This curve can reflect the characteristics of the local data segment to the greatest extent. This curve will be used as the input of the model to determine whether the tool is damaged.

[0092] Once the fitting is completed, the obtained fitting function can be used to estimate the value of each data point. This process is actually an interpolation and reconstruction of the original data. Through the continuity and smoothness of the fitting function, the data curve can be smoothed and the reliability of the data can be improved. Figure 4 FIG. 4 is a schematic diagram of a power curve after smoothing in this embodiment.

[0093] 5. Learning Model

[0094] The tool breakage detection model (TBD) designed by the present invention is used to monitor the abnormal fluctuation of the machine tool spindle input power when the machine tool tool is broken during the processing. The model includes two modules: tool processing state reference curve self-learning and tool breakage curve learning. The model structure principle is as follows Figure 2 shown.

[0095] Machine tool condition monitoring requires learning of normal processing curves. Generally speaking, for a specific process and a fixed processing tool, the power curve of the processing process should be exactly the same without considering tool wear and external interference. However, in the actual processing process, as time goes by, the number of machine tool processing times continues to accumulate, and the tool will be damaged to a certain extent. The specific damage will be reflected in the change of the power signal.

[0096] The self-learning of the tool processing state reference curve is to learn and store the first three processing processes when the machine tool is running normally (a processing process refers to the processing of a workpiece by different tools under a program number; for example, a workpiece is first processed by tool No. 1 and then by tool No. 2, and then the processing is completed. This is a processing process and a learning process). Three learning curves Q1(t), Q2(t), and Q3(t) are generated. They are used as the input of the tool breakage detection model, and the upper and lower limit values ​​at each time point are counted respectively, and finally two curves are formed:

[0097]

[0098]

[0099] Where: [ti] represents a certain time point of data collection during the tool processing process; [up] represents all the upper limit points during the tool processing process at the time [ti], and conversely [down] represents all the lower limit points during the tool processing process at the time [ti]; and They respectively represent the values ​​of the upper and lower curves obtained after three learnings at time [ti].

[0100] According to the upper and lower edge curve values ​​obtained by learning, combined with the breakage coefficient, the final upper and lower threshold curves are formed, that is, the tool breakage curve:

[0101]

[0102]

[0103] Where δ and ω are damage coefficients, and the selection of coefficients is usually based on engineering experience, processing requirements, and the distribution of current data. By adjusting the coefficients, it can be ensured that the upper and lower threshold curves are neither too loose to effectively monitor abnormal currents nor too strict to frequently trigger false alarms. Figure 5 Schematic diagram of reference curve self-learning in this embodiment.

[0104] The upper and lower threshold curves obtained by the final output can clearly define the normal fluctuation range of power during the processing. When the real-time power data exceeds this range, it can be judged that there may be abnormal conditions in the processing process, so that corresponding measures can be taken in time to adjust and optimize. In this way, not only can the stability of the processing process be improved, but also the production cost can be reduced and the product quality can be improved.

[0105] 6. Working Mode

[0106] In this embodiment, after the learning mode ends and a suitable threshold curve is generated, the system switches to the working mode, imports the working file data to judge the tool status, and labels each data point. In this process, if a value falls within the range of the upper and lower thresholds, it will output label 0, indicating normal; once the value exceeds this range, it will output label 1, indicating that the tool is damaged. Figure 6 and Figure 7 As shown in FIG. 1 , they are respectively a schematic diagram of a power curve of a normal machining process and a schematic diagram of a power curve of a tool damaged during machining in this embodiment.

[0107] After completing the data labeling work, each learning value, the corresponding upper and lower thresholds, and the labels are sorted together to form a complete data set. Finally, in order to ensure the accuracy and traceability of the data, this data set is stored in the format of a CSV file for subsequent analysis and processing.

[0108] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0109] Although the above describes the specific implementation methods of the present invention, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A tool breakage detection method based on power detection, It is characterized in that The process includes: S1, using the Hall sensor to collect the spindle UVW three-phase current signal, and calculate the spindle three-phase AC current to obtain the equivalent DC current of the spindle operation; S2, feature extraction using DB4 wavelet algorithm; The equivalent DC current signal is decomposed and reconstructed. The signal is decomposed layer by layer into wavelet coefficients of different frequency components. Then the original signal is restored through the reconstruction process to extract the characteristic value reflecting the current change during the machining process, that is, the damage root mean square R. S3, using Savitzky-Golay filtering algorithm to smooth the curve of the damage root mean square R extracted during the processing, eliminating the noise and fluctuation in the data, retaining the original characteristics and change trend of the data, and obtaining the fitting curve data as the model input; S4, input the fitting curve data obtained in real time into the constructed tool breakage detection model; the tool breakage detection model obtains the upper and lower edge curve values ​​based on the tool processing state reference curve self-learning and tool breakage curve learning, and then combines the breakage coefficient to form the final upper and lower threshold curves. If the real-time value falls within the range of the upper and lower thresholds, it will output label 0, indicating normal; once the value exceeds this range, it will output label 1, indicating that the tool is broken.

2. A tool breakage detection method based on power detection as claimed in claim 1, It is characterized in that The specific process of S2 is: First, the continuous wavelet transform of the equivalent DC current x(t) of the machine tool spindle is defined by convolution as: Where t is the displacement; a is the scale factor; b is the translation parameter; is the mother wavelet; after scaling and translating the mother wavelet, a wavelet sequence is obtained: For any signal f(t)∈L 2 (R), the continuous wavelet transform is: The reconstruction formula is: The damage root mean square R is extracted as the characteristic value reflecting the current change during the machining process:

3. A tool breakage detection method based on power detection as claimed in claim 1, It is characterized in that The specific process of S3 is as follows: First, a sliding window with R(i) as the center and containing 2M+1 data points is set. The size of the window can be adjusted according to the characteristics of the data and processing requirements; Then, the data in the window is used as a local data segment, and a p-order polynomial q(n) is constructed to fit the array: Among them, a 0 ,a 1 ,…,a m is the fitting coefficient, and the residual C is obtained through least square fitting: When the residual C is the smallest, the filtering effect is best. During the fitting process, a best fitting curve Q(t) will be found. This curve can reflect the characteristics of the local data segment to the greatest extent. This curve will be used as the input of the model to determine whether the tool is damaged.

4. A tool breakage detection method based on power detection as claimed in claim 1, It is characterized in that The tool breakage detection model in S4 is based on the collected raw data, and features are extracted through the DB4 wavelet algorithm and the extracted breakage root mean square R is smoothed by the Savitzky-Golay filtering algorithm to obtain standard model input data; the raw data is collected through a Hall sensor, the collection frequency is set to 10Khz, and the data is collected after the spindle speed tends to be stable, and the three-phase current signal is read at one time through the IO interrupt.

5. A tool breakage detection method based on power detection as claimed in claim 1, Features: The calculation of the three-phase AC current of the main shaft in S1 is a root mean square calculation: In the formula, I U ,I V ,I W Represent the current values ​​of the three-phase currents respectively.

6. A tool breakage detection method based on power detection as claimed in claim 1, Features: The tool breakage detection model and the tool processing state reference curve self-learning are to learn and store the first three processing processes when the machine tool is running normally, and generate three learning curves Q 1 (t), Q 2 (t), Q 3 (t), and use them as the input of the tool breakage detection model. The upper and lower limits of each time point are counted respectively, and finally two curves are formed: Where: [ti] represents a certain time point of data collection during the tool processing process; [up] represents all the upper limit points during the tool processing process at the time [ti], and conversely [down] represents all the lower limit points during the tool processing process at the time [ti]; and They represent the values ​​of the upper and lower curves obtained after three learnings at time [ti] respectively; According to the upper and lower edge curve values ​​obtained by learning, combined with the breakage coefficient, the final upper and lower threshold curves are formed, that is, the tool breakage curve: Where δ and ω are damage coefficients. The selection of coefficients is based on engineering experience, processing requirements, and the distribution of current data. By adjusting the coefficients, it is ensured that the upper and lower threshold curves are neither too loose to effectively monitor abnormal currents nor too strict to frequently trigger false alarms. The upper and lower threshold curves finally outputted can clearly define the normal fluctuation range of power during the processing process.