Cutter state prediction system and method

Abstract

The embodiment of the invention provides a tool state prediction system and method, which can ensure the read-write efficiency of high-frequency data in a tool state prediction process so as to ensure the real-time performance of tool state prediction. The prediction system comprises a data acquisition layer used for acquiring vibration time sequence data of a cutter in the numerical control machine tool; the data storage layer comprises a time sequence data storage unit used for storing vibration time sequence data, a preprocessing data storage unit used for storing vibration time sequence preprocessing data, and a feature data storage unit used for storing vibration time sequence feature data; the data processing layer comprises a preprocessing unit which is used for preprocessing the vibration time sequence data to obtain vibration time sequence preprocessed data; the feature extraction unit is used for performing feature extraction on the vibration time sequence preprocessing data to obtain vibration time sequence feature data; and the state prediction unit is used for predicting and obtaining the current state of the cutter according to the vibration time sequence characteristic data.

Description

Technical Field

[0001] The present application relates to the field of machine tool processing technology, and in particular to a tool state prediction system and method. Background Art

[0002] CNC machine tools process materials through the tools installed on them. The tool status determines the yield of the materials processed by the CNC machine tools. Therefore, it is necessary to design a system that can predict the tool status in real time. Since dozens or hundreds of CNC machine tools are usually required in large workshops, a large amount of CNC equipment-related data will be generated in the entire workshop at all times, which will put forward great requirements on the real-time data storage, real-time data processing and real-time data prediction capabilities of the real-time prediction system.

[0003] Therefore, how to achieve efficient real-time prediction of tool status under high-frequency data is a technical problem that needs to be solved urgently. Summary of the invention

[0004] The embodiments of the present application provide a tool state prediction system and method, which can ensure the reading and writing efficiency of high-frequency data during the tool state prediction process, thereby ensuring the real-time nature of the tool state prediction.

[0005] In a first aspect, an embodiment of the present application provides a tool state prediction system, the prediction system comprising:

[0006] The data acquisition layer is used to collect the vibration time series data of the tool in the CNC machine tool;

[0007] The data storage layer includes a time series data storage unit, a preprocessing data storage unit and a feature data storage unit; wherein the time series data storage unit is used to store the vibration time series data, the preprocessing data storage unit is used to store the vibration time series preprocessing data, and the feature data storage unit is used to store the vibration time series feature data;

[0008] The data processing layer includes a preprocessing unit, a feature extraction unit and a state prediction unit; wherein the preprocessing unit is used to preprocess the vibration time series data to obtain the vibration time series preprocessed data; the feature extraction unit is used to extract features from the vibration time series preprocessed data to obtain the vibration time series feature data; the state prediction unit is used to predict and obtain the current state of the tool based on the vibration time series feature data.

[0009] In a second aspect, an embodiment of the present application provides a method for predicting a tool state, which is applied to the prediction system of the first aspect, and the method includes:

[0010] The preprocessing unit reads the vibration time series data in the time series data storage unit, preprocesses the vibration time series data to obtain the vibration time series preprocessing data, and stores the vibration time series preprocessing data in the preprocessing data storage unit;

[0011] The feature extraction unit reads the vibration time series preprocessing data in the preprocessing data storage unit, performs feature extraction on the vibration time series preprocessing data to obtain the vibration time series feature data, and stores the vibration time series feature data in the feature data storage unit;

[0012] The state prediction unit reads the vibration time series characteristic data in the characteristic data storage unit, and predicts and obtains the current state of the tool according to the vibration time series characteristic data.

[0013] In the embodiment of the present application, in the prediction system, the data acquisition layer is responsible for collecting the vibration time series data of the tool in the CNC machine tool, and the vibration time series data will be written into the time series data storage unit of the data storage layer, and then the preprocessing unit in the data processing layer will read the vibration time series data from the time series data storage unit for preprocessing to obtain the vibration time series preprocessing data. The vibration time series preprocessing data will be written into the preprocessing data storage unit of the data storage layer. Next, the feature extraction unit in the data processing layer will read the vibration time series preprocessing data from the preprocessing data storage unit for feature extraction to obtain vibration time series feature data. The vibration time series feature data will be written into the feature data storage unit of the data storage layer. Finally, the state prediction unit in the data processing layer will read the vibration time series feature data from the feature data storage unit to perform the tool state prediction task. In the above prediction system architecture, the time series vibration data, the vibration time series preprocessing data, and the vibration time series feature data are respectively stored in different storage units of the data storage layer, so that the above data located in different storage units at the same time can be read and written based on the maximum performance without reading and writing congestion, thereby improving the data transmission efficiency in the entire system, and then improving the real-time performance of tool state prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0015] Figure 1 An architecture diagram of a prediction system provided in an embodiment of the present application;

[0016] Figure 2A flowchart of a prediction method provided in an embodiment of the present application;

[0017] Figure 3 A schematic diagram of a flow chart of a feature extraction method provided in an embodiment of the present application;

[0018] Figure 4 A schematic diagram of a flow chart of another method for extracting a vibration feature matrix provided in an embodiment of the present application;

[0019] Figure 5 A schematic diagram of a flow chart of another method for extracting a vibration feature matrix provided in an embodiment of the present application;

[0020] Figure 6 A schematic diagram of a process flow of a tool state prediction method provided in an embodiment of the present application;

[0021] Figure 7 A schematic diagram of a process flow of a tool state prediction method provided in an embodiment of the present application;

[0022] Figure 8 An architecture diagram of another prediction system provided in an embodiment of the present application;

[0023] Fig. 9 A flowchart of a method for determining a current difference threshold provided in an embodiment of the present application;

[0024] Fig.10 A flowchart of a method for updating a current difference threshold value provided in an embodiment of the present application;

[0025] Fig.11 A flowchart of a method for updating a current difference threshold provided in an embodiment of the present application;

[0026] Fig.12 An architectural diagram of another prediction system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to better understand the technical solution of this specification, the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0028] It should be clear that the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this specification.

[0029] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit this specification. The singular forms of "a", "said" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0030] CNC machine tools process materials through the tools installed on them. The status of the tools determines the yield of the materials processed by the CNC machine tools. Therefore, it is necessary to design a system that can predict the status of the tools in real time.

[0031] In view of this, an embodiment of the present application provides a prediction system, in which multiple intermediate data used for tool state prediction are stored in different storage units respectively, so that the above-mentioned intermediate data located in different storage units at the same time can be read and written based on maximum performance without read and write congestion, thereby improving the transmission efficiency of the intermediate data in the entire system, and further improving the real-time performance of tool state prediction.

[0032] The technical solution protected by this application is described in detail below with reference to the accompanying drawings.

[0033] See also Figure 1 , is an architecture diagram of a prediction system provided in an embodiment of the present application. Figure 1 As shown, the prediction system includes:

[0034] The data acquisition layer 10 is used to collect the vibration time series data of the tool in the CNC machine tool;

[0035] The data storage layer 20 includes a time series data storage unit 201, a preprocessing data storage unit 202 and a feature data storage unit 203; wherein the time series data storage unit 201 is used to store vibration time series data, the preprocessing data storage unit 202 is used to store vibration time series preprocessing data, and the feature data storage unit 203 is used to store vibration time series feature data;

[0036] The data processing layer 30 includes a preprocessing unit 301, a feature extraction unit 302 and a state prediction unit 303; wherein the preprocessing unit 301 is used to preprocess the vibration time series data to obtain vibration time series preprocessed data; the feature extraction unit 302 is used to extract features from the vibration time series preprocessed data to obtain vibration time series feature data; the state prediction unit 303 is used to predict and obtain the current state of the tool based on the vibration time series feature data.

[0037] See also Figure 2 , which is a flow chart of a prediction method provided in an embodiment of the present application. Based on the above prediction system, the tool status prediction method is carried out in the following manner.

[0038] Step 1: the preprocessing unit 301 reads the vibration time series data in the time series data storage unit 201 , preprocesses the vibration time series data to obtain vibration time series preprocessed data, and stores the vibration time series preprocessed data in the preprocessed data storage unit 202 .

[0039] Step 2, the feature extraction unit 302 reads the vibration time series preprocessing data in the preprocessing data storage unit 202, performs feature extraction on the vibration time series preprocessing data to obtain vibration time series feature data, and stores the vibration time series feature data in the feature data storage unit 203.

[0040] Step three, the state prediction unit 303 reads the vibration time series characteristic data in the characteristic data storage unit 203, and predicts and obtains the current state of the tool according to the vibration time series characteristic data.

[0041] The entire prediction system is described in detail below.

[0042] In the embodiment of the present application, a plurality of CNC machine tools are arranged in the processing workshop, and a vibration sensor is installed on the main shaft of each CNC machine tool, and the vibration sensor can collect the vibration time series data generated by the tool after the CNC machine tool is started based on the preset frequency. For example, the preset frequency is 4Khz. It should be understood that the vibration sensor collects three-axis vibration time series data, namely, the vibration time series data of the tool on the X-axis, the vibration time series data of the Y-axis, and the vibration time series data of the Z-axis, which means that 4K vibration data will be collected every 1s for each axis in the above-mentioned X-axis / Y-axis / Z-axis, then 12K vibration data will be collected for the three axes in 1s, and then superimposed on 4K timestamps, totaling 16K data.

[0043] The data acquisition layer 10 may include multiple Gateway integrated gateways, each of which is connected to at least one vibration sensor for acquiring vibration time series data of at least one vibration sensor. As a possible embodiment, each Gateway integrated gateway is connected to two vibration sensors, that is, each Gateway integrated gateway can simultaneously acquire vibration time series data acquired by two vibration sensors.

[0044] For each Gateway integrated gateway in the data acquisition layer 10, after receiving the start processing signal corresponding to a certain CNC machine tool, the vibration data and the corresponding timestamp are obtained from the corresponding vibration sensor. Each time the Gateway integrated gateway receives a vibration data, it will update the timestamp accordingly and push the received vibration data into the vibration data queue. If the data acquisition layer 10 determines that the timestamp change exceeds the preset time length, for example, the preset time length is 1s, the vibration data in the vibration data queue will be sent to the data storage layer 20 as a segmented data, thereby improving the data transmission efficiency between the data acquisition layer 10 and the data storage layer 20. At the same time, the above-mentioned vibration data in the vibration data queue will be cleared to facilitate the continued storage of new vibration data until the data acquisition layer 10 receives the processing end signal.

[0045] For the vibration time series data from each Gateway integrated gateway, Kafka and Flink are correspondingly set in the data storage layer 20 for one-to-one processing. For example, for the vibration time series data from the first Gateway integrated gateway, Kafka topic1 and Flink job1 are set for targeted processing. By analogy, for the vibration time series data from the mth Gateway integrated gateway, Kafka topic m and Flink job m are set for targeted processing. Among them, Kafka is a distributed message queue system that acts as a "buffer pool" for vibration time series data, which can temporarily store a large amount of vibration time series data from the corresponding Gateway integrated gateway. Flink is a powerful distributed stream processing framework that can process real-time data streams obtained from Kafka. The specific processing content can be designed according to actual needs, and this application does not impose any special restrictions on this. The data processed by Flink will be written to the time series data storage unit 201 in the data storage layer 20.

[0046] Then the preprocessing unit 301 in the data processing layer 30 reads the vibration time series data from the time series data storage unit 201 for preprocessing, and obtains vibration time series preprocessing data. As a possible implementation, the preprocessing unit 301 reads the previous vibration time series data and the current vibration time series data generated by the current tool of the numerical control machine tool during the processing from the time series data storage unit 201, and then the previous vibration time series data and the current vibration time series data are preprocessed to obtain the previous vibration time series preprocessing data and the current vibration time series preprocessing data. Since the process of preprocessing the previous vibration time series data and the current vibration time series data is the same, the following is an example of preprocessing the current vibration time series data.

[0047] Since the process of using a CNC machine tool to process the current material may involve switching different tools to execute different processing program segments in sequence, for example, in the process of processing the current material, it involves [T1 tool processing, tool change 1, T2 tool processing, tool change 2, T1 tool processing], and the processing start time of different processing program segments is different, therefore, the preprocessing unit 301 will use the vibration data with the same machine number, the same tool number, and the same processing start time as the same segmentation data in the process of processing the current material, and then filter out at least one segmentation data (i.e., the current vibration time series data) corresponding to the tool currently required to be detected by the tool number, and filter out the data located in the non-processing program segment in each segmentation data of the at least one segmentation data, and finally, merge the above-mentioned segmentation data after the screening operation, that is, merge the vibration time series data generated by the same machine, the same tool, and different processing start times after filtering out the non-processing program segments when processing the above-mentioned current material, so as to obtain the current vibration time series preprocessing data corresponding to the current tool processing the current material. Similarly, the previous vibration time series preprocessing data corresponding to the current tool processing the previous material can be obtained.

[0048] The vibration time series preprocessing data obtained by the preprocessing unit 301 after preprocessing will be written into the preprocessing data storage unit 202 in the data storage layer 20. Then, the feature extraction unit 302 in the data processing layer 30 will read the vibration time series preprocessing data from the preprocessing data storage unit 202 for feature extraction to obtain vibration time series feature data.

[0049] See also Figure 3 , is a flow chart of a feature extraction method provided in an embodiment of the present application. The flow chart of the method is described as follows:

[0050] Step S1: According to the previous vibration timing preprocessing data and the current vibration timing preprocessing data, respectively obtain the vibration characteristic matrix corresponding to the previous vibration timing preprocessing data and the current vibration timing preprocessing data.

[0051] In the embodiment of the present application, the vibration feature matrix includes vibration time domain features and vibration frequency domain features. The vibration time domain features and vibration frequency domain features are described in detail below.

[0052] The vibration time domain features include: amplitude feature index, statistical feature index, waveform feature index and impact feature index. Among them, the amplitude feature index includes: peak value, valley value, peak-to-peak value; the statistical feature index includes: mean value, variance, standard deviation, root mean square value; the waveform feature index includes: kurtosis, skewness and waveform index; the impact feature index includes: peak index, pulse index and marginal coefficient. The following is a detailed introduction to the above thirteen vibration time domain features.

[0053] Vibration time domain feature 1:

[0054] Peak value: The maximum value of the signal, reflecting the maximum amplitude of tool vibration. The formula for the peak value is shown in (1):

[0055] f td1 =max(s(t))(1)

[0056] Among them, f td1 represents the peak value, and s(t) represents the vibration amplitude of any axis at the tth time stamp.

[0057] Vibration time domain feature 2:

[0058] Valley value: The minimum value of the signal, reflecting the minimum amplitude of tool vibration. The formula for the valley value is shown in (2):

[0059] f td2 =min(s(t))(2)

[0060] Among them, f td2 Indicates valley value.

[0061] Vibration time domain feature three:

[0062] Peak-to-peak value: The difference between the peak value and the valley value, which reflects the amplitude range of tool vibration.

[0063] The formula for the peak-to-peak value is shown in (3):

[0064] f td3 =f td1 -f td2 (3)

[0065] Among them, f td3 Indicates peak-to-peak value.

[0066] Vibration time domain feature 4:

[0067] Mean: represents the average amplitude of the signal, reflecting the average level of tool vibration. The formula for the mean is shown in (4):

[0068]

[0069] Among them, f td4 represents the mean, and T represents the total number of timestamps.

[0070] Vibration time domain feature 5:

[0071] Variance: measures the degree to which the signal deviates from the mean, reflecting the stability of tool vibration. The calculation formula of variance is shown in (5):

[0072]

[0073] Among them, ftd5 Represents variance.

[0074] Vibration time domain feature six:

[0075] Standard deviation: The square root of the variance, which is also an indicator of tool vibration stability. The calculation formula of standard deviation is shown in (6):

[0076]

[0077] Among them, f td6 Represents standard deviation.

[0078] Vibration time domain feature seven:

[0079] RMS value: The RMS value of a signal is often used to measure the energy of tool vibration. The RMS value calculation formula is shown in (7):

[0080]

[0081] Among them, f td7 Represents the root mean square value.

[0082] Vibration time domain feature eight:

[0083] Kurtosis: The concentration of signal distribution, describing the concentration near the tool vibration peak. The calculation formula of Kurtosis is shown in (8):

[0084]

[0085] Among them, f td8 Represents kurtosis.

[0086] Vibration time domain feature nine:

[0087] Skewness: The asymmetry of the signal distribution, describing the degree of deviation of the tool vibration. The calculation formula of skewness is shown in (9):

[0088]

[0089] Among them, f td9 Represents skewness.

[0090] Vibration time domain feature 10:

[0091] Waveform index: The ratio between the root mean square value of the signal and the absolute value of the average value of the signal. The waveform index can reflect the characteristics of the tool during processing as follows:

[0092] First, the asymmetry and nonlinearity of the signal waveform: The waveform index can reveal the symmetry and nonlinear characteristics of the signal waveform. If the signal is a completely symmetrical sine wave, then its waveform factor is close to 1. For the actual tool vibration signal, due to the possible asymmetric impact and irregular fluctuations, the waveform index will be greater than 1, and the larger the ratio, the more irregular the signal waveform.

[0093] Second, the complexity of tool vibration: A larger value of the waveform index usually means that the tool vibration signal contains more complex components, such as impact, noise or non-periodic components. This may be related to tool wear, uneven load during cutting or the dynamic characteristics of the machine tool.

[0094] Third, signal energy distribution: The waveform factor is also related to the distribution of signal energy. A higher waveform factor may indicate that the signal energy is concentrated at certain moments or frequencies, which may affect the cutting efficiency and processing quality of the tool.

[0095] The calculation formula of the waveform indicator is shown in formula (10):

[0096]

[0097] Among them, f td10 Indicates a waveform indicator.

[0098] Vibration time domain feature 11:

[0099] Peak index: reflects the impact component in tool vibration. If the peak index is large, it means that there is a large impact component in tool vibration, which may be related to tool wear, breakage or instability during processing. The calculation formula of the peak index is shown in (11):

[0100]

[0101] Among them, f td11 Indicates peak index.

[0102] Vibration time domain feature 12:

[0103] Pulse index: A relative measure of the maximum amplitude or intensity of a signal and the overall energy level or intensity. It reflects the relationship between the extreme value and the average value of the signal in amplitude or intensity. The pulse index can reflect the following characteristics of the tool during processing:

[0104] Tool wear status: As the tool gradually wears, parameters such as the maximum cutting force and maximum vibration amplitude during cutting may change. At the same time, the overall vibration or cutting force level may also be affected. Therefore, this ratio can be used as a sensitive indicator to monitor the wear status of the tool. When the ratio increases, it may indicate that tool wear has caused the maximum amplitude or intensity of the cutting process to increase relative to the overall energy level.

[0105] Stability of the cutting process: The stability of the cutting process is critical to the quality of tool processing. The RMS value of the signal reflects the overall vibration or cutting force level during the cutting process, while the maximum value of the absolute value of the signal reflects the extreme conditions during the cutting process. Therefore, this ratio can reflect the stability of the cutting process. If the ratio fluctuates greatly, it means that there are large amplitude or intensity changes in the cutting process, which may indicate that the cutting process is not stable enough.

[0106] Changes in cutting conditions: Changes in cutting conditions such as cutting speed, feed rate, cutting depth, etc. will affect the maximum cutting force, maximum vibration amplitude, and overall vibration or cutting force level during the cutting process. Therefore, by monitoring the changes in this ratio, the impact of changes in cutting conditions on the tool processing process can be discovered in a timely manner.

[0107] The calculation formula of the pulse index is shown in (12):

[0108]

[0109] Among them, f td12 Indicates the pulse indicator.

[0110] Vibration time domain feature 13:

[0111] Marginal coefficient: In the context of tool processing, the marginal coefficient reflects the relationship between the maximum amplitude in the tool vibration signal and the discrete degree of vibration. A higher marginal coefficient indicates that there are larger peaks in the tool vibration, which may mean that there are larger shocks or instabilities in the processing process relative to the average level of vibration. This indicator can be used to evaluate the wear state of the tool or abnormal conditions in the processing process because it can reveal the relationship between the extreme values ​​in the vibration signal and the overall vibration level. The calculation formula of the marginal coefficient is shown in (13):

[0112]

[0113] Among them, f td13 represents the marginal coefficient.

[0114] The vibration frequency domain features include: power spectrum feature indicators, statistical feature indicators, and waveform feature indicators. Among them, the power spectrum feature indicators include: power spectrum density peak, peak frequency; statistical feature indicators include: average power spectrum density, power spectrum density variance, frequency standard deviation, frequency root mean square; waveform feature indicators include: spectrum kurtosis, spectrum skewness, frequency skewness, frequency kurtosis, center of gravity frequency, D factor, E factor, G factor. The following is a detailed introduction to the above fourteen vibration frequency domain features.

[0115] Vibration frequency domain feature 1:

[0116] Average power spectral density: The average value of the power spectral density in the entire frequency domain reflects the overall energy distribution of tool vibration. The calculation formula of the average power spectral density is shown in (14):

[0117]

[0118] Among them, f tfd1 represents the average power spectral density, p(k) represents the power spectral density of the kth frequency point, and K represents the total number of frequency points.

[0119] Vibration frequency domain feature 2:

[0120] Power spectral density variance: The degree of dispersion of power spectral density reflects the difference in energy distribution of different frequency components. The calculation formula of power spectral density variance is shown in (15):

[0121]

[0122] Among them, f tfd2 represents the power spectral density variance.

[0123] Vibration frequency domain feature three:

[0124] Spectral skewness: The asymmetry of the power spectrum density distribution reflects the degree of skewness of the tool vibration spectrum. The calculation formula of frequency deviation is shown in (16):

[0125]

[0126] Among them, f tfd3 Represents the spectral skewness.

[0127] Vibration frequency domain feature 4:

[0128] Spectral kurtosis: The sharpness of the power spectrum density distribution reflects the peak concentration of the tool vibration spectrum. The calculation formula of the spectrum skewness is shown in (17):

[0129]

[0130] Among them, ftfd4 represents the spectral kurtosis.

[0131] Vibration frequency domain feature 5:

[0132] Center of gravity frequency: The geometric center of the power spectrum, reflecting the main concentration area of ​​the tool vibration energy. The calculation formula of the center of gravity frequency is shown in (18):

[0133]

[0134] Among them, f tfd5 represents the center of gravity frequency, f (k) Represents the frequency of the kth frequency point.

[0135] Vibration frequency domain feature six:

[0136] Frequency standard deviation: The weighted standard deviation of the spectrum, which describes the width of the spectrum distribution of the signal. The calculation formula of the frequency standard deviation is shown in (19):

[0137]

[0138] Among them, f tfd6 Represents the frequency standard deviation.

[0139] Vibration frequency domain feature seven:

[0140] Frequency root mean square: The weighted root mean square value of the spectrum, which describes the energy center of the signal's spectrum distribution. The calculation formula for frequency root mean square is shown in (20):

[0141]

[0142] Among them, f tfd7 Represents the root mean square frequency.

[0143] Vibration frequency domain feature eight:

[0144] D factor: It is a dimensionless number used to describe the shape of the spectrum. It describes the distribution characteristics of the signal spectrum, especially the sharpness of the spectrum and the influence of high-frequency components. The calculation formula of D factor is shown in (21):

[0145]

[0146] Among them, f tfd8 Denotes the D factor.

[0147] Vibration frequency domain feature nine:

[0148] E factor: It is a dimensionless number used to describe the shape of the spectrum. It reflects the influence of the energy center and high-frequency components of the spectrum. The calculation formula of E factor is shown in (22):

[0149]

[0150] Among them, f tfd9 Indicates the E factor.

[0151] Vibration frequency domain feature 10:

[0152] G factor: It is a dimensionless number that can be used to standardize the spectral width of the signal so that it is independent of the center of gravity frequency. It reflects the relative width or variability of the spectral distribution. The calculation formula of G factor is shown in (23):

[0153]

[0154] Among them, f tfd10 Indicates the G factor.

[0155] Vibration frequency domain feature 11:

[0156] Frequency skewness: It represents the weighted skewness of the signal spectrum and describes the asymmetry of the spectrum distribution. The calculation formula of frequency skewness is shown in (24):

[0157]

[0158] Among them, f tfd11 Represents frequency skewness.

[0159] Vibration frequency domain feature 12:

[0160] Frequency kurtosis: describes the sharpness of the signal spectrum, that is, whether the spectrum distribution has a high peak. High kurtosis indicates that there are significant sharp components in the spectrum, which may correspond to shocks or transient events in the signal. The calculation formula of frequency kurtosis is shown in (25):

[0161]

[0162] Among them, f tfd12 Represents frequency kurtosis.

[0163] Vibration frequency domain feature 13:

[0164] Power spectral density peak: The maximum value of the power spectral density reflects the power of the most significant frequency component. The calculation formula for the power spectral density peak is as shown in (26):

[0165] f tfd13 =max(p(k))(26)

[0166] Among them, f tfd13 Represents the peak value of the power spectral density.

[0167] Vibration frequency domain feature 14:

[0168] Peak frequency: The frequency point with the largest power spectrum density indicates the most significant frequency component in the tool vibration. The calculation formula for peak frequency is shown in (27):

[0169] f tfd14 =argmax(p(k))(27)

[0170] Among them, f tfd14 Indicates the peak frequency.

[0171] In some embodiments, considering that the vibration frequency of the current tool during the machining process may be mainly concentrated in a specific frequency band, therefore, in the embodiment of the present application, the feature extraction unit 302 can extract the vibration feature matrix only for the vibration timing data within the machining frequency band, thereby improving the accuracy of the extracted vibration feature matrix.

[0172] See also Figure 4 , which is a flow chart of another method for extracting a vibration feature matrix provided in an embodiment of the present application. Step S1 can be specifically implemented through sub-steps S11 to S12:

[0173] Step S11: filtering the previous vibration time series preprocessing data and the current vibration time series preprocessing data respectively based on the first target frequency band.

[0174] Step S12: based on the filtered previous vibration time series preprocessing data and the current vibration time series preprocessing data, respectively extract the vibration time domain features and vibration frequency domain features corresponding to the previous vibration time series preprocessing data and the current vibration time series preprocessing data.

[0175] In the above embodiment, the first target frequency band can be considered as the working frequency band of the current tool in the processing stage. For example, the first target frequency band can be [50Hz, 1950Hz]. Then the feature extraction unit 302 uses the first target frequency band to filter the previous vibration timing preprocessing data and the current vibration timing preprocessing data corresponding to the last two processing processes, which is beneficial to screen out other frequency bands irrelevant to the processing stage, so that the time-frequency domain features extracted for the processing stage based on the previous vibration timing preprocessing data and the current vibration timing preprocessing are more accurate.

[0176] In some embodiments, considering that the rotation speed and number of blades of the tool itself during operation may also affect the distribution of its effective vibration signal, therefore, in an embodiment of the present application, the feature extraction unit 302 can extract frequency domain features within the frequency band where the effective signals of tool vibration are concentrated, thereby improving the accuracy of the extracted frequency domain features.

[0177] See also Figure 5, which is a flow chart of another method for extracting a vibration feature matrix provided in an embodiment of the present application. Step S12 can be specifically implemented through sub-steps S121 to S124:

[0178] Step S121: Determine the fundamental frequency of the time-frequency domain conversion according to the current set rotation speed of the tool.

[0179] In the embodiment of the present application, taking the current set speed of the tool as 6000 r / min as an example, the base frequency of the time-frequency domain conversion is 6000 / 60=100 Hz.

[0180] Step S122: Determine the maximum multiple of the time-frequency domain conversion according to the number of blades and the fundamental frequency of the current tool.

[0181] In the embodiment of the present application, taking the current tool having four blades as an example, if the base frequency is 100 Hz, then the maximum multiple of the time-frequency domain conversion is 100*4=400 Hz.

[0182] Step S123: determining at least two second target frequency bands for time-frequency domain conversion according to the base frequency and the maximum multiple frequency.

[0183] In the embodiment of the present application, if the base frequency is 100 Hz and the maximum multiplier is 400 Hz, then the second target frequency bands can be directly determined to be [50 Hz, 250 Hz] and [250 Hz, 450 Hz] respectively.

[0184] Of course, in another possible embodiment, a plurality of intermediate multiples can be determined between the base frequency and the maximum multiple, and then at least two second target frequencies for time-frequency domain conversion can be determined based on the base frequency, the plurality of intermediate multiples and the maximum multiple. Continuing with the example of a base frequency of 100 Hz and a maximum multiple of 400 Hz, the plurality of intermediate multiples that can be determined are 200 Hz and 300 Hz, respectively. On this basis, the second target frequency bands determined are [50 Hz, 150 Hz], [150 Hz, 250 Hz], [250 Hz, 350 Hz], and [350 Hz, 450 Hz].

[0185] It is worth noting that each second target frequency band needs to cover the base frequency, the middle multiple frequency or the maximum multiple frequency.

[0186] Step S124: performing time-frequency domain conversion on the filtered previous vibration timing preprocessing data and the current vibration timing preprocessing data, respectively, to obtain vibration frequency domain class features corresponding to each second target frequency band in the previous vibration timing preprocessing data and the current vibration timing preprocessing data.

[0187] In an embodiment of the present application, after determining multiple second target frequency bands in which effective signals of tool vibration are concentrated based on the set rotation speed and number of blades of the current tool, when performing time-frequency domain conversion, the vibration frequency characteristics of the previous vibration timing preprocessing data and the current vibration timing preprocessing data in each second target frequency band can be obtained respectively, without the need to obtain the vibration frequency domain characteristics of other frequency bands except the second target frequency band, thereby improving the accuracy of the extracted vibration frequency domain characteristics.

[0188] The above-mentioned vibration time series characteristic data will be written into the characteristic data storage unit 203 of the data storage layer 20. Finally, the state prediction unit 303 of the data processing layer 30 will read the vibration time series characteristic data from the characteristic data storage unit 203 to predict the tool state of the current tool.

[0189] In the above prediction system architecture, the time series vibration data, vibration time series preprocessing data, and vibration time series characteristic data are respectively stored in different storage units of the data storage layer 20. For example, the different storage units in the data layer 20 can be considered as part or all of the redis data storage system. These storage units can share the data storage space function to maximize the storage space utilization, so that the above data located in different storage units at the same time can be read and written based on the maximum performance without read and write congestion, thereby improving the data transmission efficiency in the entire system, and then improving the real-time performance of tool state prediction.

[0190] It is worth noting that the survival time can also be set specifically for the data in different storage units in the data storage layer 20. When the set survival time is reached, the corresponding data in the corresponding storage unit will be deleted, thereby saving the memory space of the corresponding storage unit.

[0191] The following is a detailed description of how the state prediction unit 303 predicts the tool state of the current tool.

[0192] See also Figure 6 , is a schematic diagram of a process flow of a tool state prediction method provided in an embodiment of the present application. The process flow of the method is described as follows:

[0193] Step S2: According to the vibration feature matrix corresponding to the previous vibration time series preprocessing data and the current vibration time series preprocessing data, the current feature difference between the previous machining process and the current machining process of the current tool is obtained.

[0194] In the embodiment of the present application, the state prediction unit 303 can obtain the current feature difference between the two recent machining processes of the current tool by calculating the Euclidean distance between the vibration feature matrix corresponding to the previous vibration time series preprocessing data and the current vibration time series data. Of course, the above-mentioned current feature difference can also be obtained by other methods, and the present application does not impose any special restrictions on this.

[0195] It is worth noting that before calculating the characteristic similarity, the vibration time domain features and the vibration frequency domain features in the vibration feature matrix can be standardized respectively.

[0196] Step S3: Based on the comparison result of the feature difference this time and the current difference threshold, determine the current state of the current tool.

[0197] In the embodiment of the present application, the current characteristic difference between the current vibration timing preprocessing data and the previous vibration timing preprocessing data reflects the degree of change of the current vibration timing preprocessing data compared with the previous vibration timing preprocessing data. The smaller the current characteristic difference, the smaller the degree of change of the current vibration timing preprocessing data compared with the previous vibration timing preprocessing data. On the contrary, the larger the current characteristic difference, the larger the degree of change of the current vibration timing preprocessing data compared with the previous vibration timing preprocessing data. By comparing the above-mentioned current characteristic difference with the current difference threshold, the current tool state corresponding to the current vibration timing preprocessing data is determined.

[0198] See also Figure 7 , is a flow chart of a tool state prediction method provided in an embodiment of the present application. Step S3 can be implemented by executing sub-steps S31 to S33:

[0199] Step S31: determine whether the current feature difference is greater than the current difference threshold; wherein the current feature difference is the Euclidean distance between the vibration feature matrices corresponding to the previous vibration time series preprocessing data and the current vibration time series preprocessing data.

[0200] In the embodiment of the present application, the Euclidean distance is used to evaluate the difference between the vibration feature matrix corresponding to the current vibration timing preprocessing data and the previous vibration timing preprocessing data. Therefore, the Euclidean distance between the vibration feature matrix corresponding to the current vibration timing preprocessing data and the previous vibration timing preprocessing data can be used to compare the set difference threshold.

[0201] Step S32: If it is greater than the current difference threshold, it is determined that the current tool is in an abnormal state.

[0202] In the embodiment of the present application, if the Euclidean distance between the current vibration timing preprocessing data and the previous vibration timing preprocessing data is large, it indicates that the current vibration timing preprocessing data has a greater degree of change than the previous vibration timing preprocessing data.

[0203] On this basis, if the tool state corresponding to the previous vibration timing preprocessing data is normal, then it can be inferred that the tool state corresponding to the current vibration timing preprocessing data is abnormal. If the tool state corresponding to the previous vibration timing preprocessing data is abnormal, then it can be inferred that the tool state corresponding to the current vibration timing preprocessing data is still abnormal.

[0204] Step S33: If it is not greater than the current difference threshold, it is determined that the current tool is in a normal state.

[0205] In the embodiment of the present application, if the Euclidean distance between the current vibration timing preprocessing data and the previous vibration timing preprocessing data is small, it indicates that the current vibration timing preprocessing data has a smaller degree of change than the previous vibration timing preprocessing data.

[0206] On this basis, if the tool state corresponding to the previous vibration time series preprocessing data is normal, then it can be inferred that the tool state corresponding to the current vibration time series preprocessing data is normal. If the tool state corresponding to the previous vibration time series data is abnormal, then it can be inferred that the tool state corresponding to the current vibration time series preprocessing data is still normal.

[0207] How to determine the current difference threshold and the initial difference threshold is described in detail below.

[0208] See also Figure 8 , is an architecture diagram of another prediction system provided in an embodiment of the present application. The prediction system also includes:

[0209] The database layer 40 is used to store the data of the data storage layer 20;

[0210] The model automatic deployment layer 50 includes a continuous integration unit 501 and a continuous delivery unit 502; wherein the continuous integration unit 501 is used to train and obtain a tool state prediction model according to the data in the database layer; and the continuous delivery unit 502 is used to deploy the tool state prediction model in the data processing layer 30.

[0211] In the embodiment of the present application, data in different storage units in the data storage layer 20 can be written into the database layer 40, thereby realizing the persistent storage of the above data.

[0212] For example, vibration time series data can be stored in the influxDB database in the database layer 40; vibration time series preprocessing data can be stored in the PostgreSQL database in the database layer 40; vibration time series feature data can be stored in the PostgreSQL database in the database layer 40; and tool state prediction results can be stored in the PostgreSQL database in the database layer 40.

[0213] The continuous integration unit 501 in the model automatic deployment layer 50 can read data from the database layer 40 for offline training of the tool state prediction model. It should be understood that the current difference threshold used for tool state detection in the above embodiment can be considered as an internal parameter of the trained tool state prediction model. Therefore, the following describes how to determine the current difference threshold.

[0214] See also Fig. 9 , is a flow chart of a method for determining a current difference threshold provided in an embodiment of the present application. After responding to the model training instruction of the continuous integration unit 501, the flow of the method is described as follows:

[0215] Step S41: calling the database layer 40 to obtain the time series data of all vibrations and tool status data generated by the tool during the historical machining process of the CNC machine tool.

[0216] In an embodiment of the present application, the continuous integration unit 501 can obtain the vibration timing data generated by the tool within a preset time period. Then, the detection frequency, tool parameters and alarm information of the tool setting instrument in history can be obtained. The vibration timing data are grouped according to the number of vibration timing data generated by the tool during the processing and the detection frequency of the tool setting instrument. For example, the detection frequency is 5, that is, a tool inspection is performed every time 5 materials are processed. If the number of vibration timing data obtained is 10, then the 10 vibration timing data can be divided into 2 groups. For each group of vibration timing data, the tool setting instrument will perform a tool inspection after generating the last vibration data of the group. If a broken tool is detected, an alarm message will be output.

[0217] For the vibration time series data of the nth group, if there is no corresponding alarm information after the last vibration time series data, the tool state corresponding to the vibration time series data of the nth group is determined to be normal. If there is an alarm information after the last vibration time series data, and the tool parameters indicate that the tool has not been replaced, the tool state corresponding to the vibration time series data of the nth group is determined to be suspected abnormal; if there is an alarm information after the last vibration time series data, and the tool parameters indicate that the tool has been replaced, the tool state corresponding to the last vibration time series data is determined to be abnormal, and the tool states corresponding to the other vibration time series data in the vibration time series data of the nth group except the last vibration time series data are determined to be suspected abnormal.

[0218] Step S42: calling the preprocessing unit 301 to preprocess the previous vibration time series data to obtain the previous vibration time series preprocessing data.

[0219] In the embodiment of the present application, the continuous integration unit 501 can call the preprocessing unit 301 to preprocess the previous vibration time series data, and the previous vibration time series preprocessed data are finally stored in the database layer 40. The specific preprocessing process can be found in the above records and will not be repeated here.

[0220] Step S43: calling the feature extraction unit 302 to obtain the vibration feature matrices corresponding to the vibration time series preprocessing data.

[0221] In the embodiment of the present application, the continuous integration unit 501 can call the feature extraction unit 302 to extract the characteristics of the previous vibration time series data, and obtain the corresponding vibration feature matrix, wherein the vibration feature matrix includes vibration time domain features and vibration frequency domain features. The specific vibration time domain features and vibration frequency domain features include which features, which can be referred to the content recorded above, and will not be repeated here. The previous vibration feature matrices are finally stored in the database layer 40.

[0222] It is worth noting that the status of the tool in the vibration time series data can be a normal state, an abnormal state, or a suspected abnormal state, which can be selected according to actual needs.

[0223] Step S44: calling the state prediction unit 303 to obtain the feature differences corresponding to two adjacent times in the historical processing of the CNC machine tool according to the previous vibration feature matrices, and constructing a previous feature difference set.

[0224] In the embodiment of the present application, the characteristic difference can be obtained by calculating the Euclidean distance between the vibration characteristic matrices corresponding to the adjacent vibration time series preprocessing data, thereby constructing a set of characteristic difference degrees. Of course, the characteristic difference can also be obtained by other methods, and the present application does not impose any special restrictions on this.

[0225] It is worth noting that before calculating the feature difference, the vibration time domain features and vibration frequency domain features in the vibration feature matrix can be standardized separately. Since the main purpose of this application is to detect abnormal tool status, during the standardization process, only the vibration feature matrix corresponding to the normal tool status is standardized.

[0226] Step S45: With the purpose of enabling the initial difference threshold to distinguish between the abnormal state and the normal state of the tool, the initial difference threshold is iteratively adjusted according to previous feature difference sets and tool states until the iteration is completed to obtain the initial difference threshold; the initial difference threshold is used as the initial value of the current difference threshold.

[0227] Two possible threshold iterative adjustment implementation schemes are provided here.

[0228] For the first feature difference in the previous feature difference sets, if the first feature difference is less than the initial value of the initial difference threshold, the predicted tool state is determined to be normal; since the real tool state is normal at this time, and the predicted tool state is also normal, it indicates that the initial value of the initial difference threshold is set appropriately, and the initial difference threshold should be controlled to remain unchanged or updated in a decreasing direction. If the first feature difference is greater than the initial value of the initial difference threshold, the predicted tool state is determined to be abnormal; since the real tool state is normal at this time, and the predicted tool state is abnormal, it indicates that the initial value of the initial difference threshold is set too small, and the initial difference threshold should be controlled to be updated in an increasing direction. The subsequent iterative process is analogous to this until the latest initial difference threshold is obtained.

[0229] The second solution is to use the current initial difference threshold to judge the tool state for n groups of vibration time series data, obtain the judgment index (such as accuracy, false positive rate, etc.) corresponding to the n groups of vibration time series data, and then iteratively adjust the initial difference threshold according to the judgment index.

[0230] The continuous delivery unit 502 in the model automatic deployment layer 50 is used to deploy the latest tool state prediction model to the data processing layer, so that the tool state prediction model in the data processing layer 30 can be continuously updated, thereby ensuring the accuracy of tool state detection.

[0231] In some embodiments, considering that the tool itself will be constantly worn during the machining process, the current difference threshold used to determine the tool state should also be adaptively updated to improve the accuracy of the tool state determination.

[0232] See also Fig.10 , is a flow chart of a method for updating a current difference threshold provided in an embodiment of the present application. In response to the update threshold instruction of the continuous delivery unit 502, the flow of the method is described as follows:

[0233] S51: calling the database layer 40 to obtain multiple vibration time series data of the processing process of a previously set number of times.

[0234] S52: calling the preprocessing unit 301 to preprocess the multiple vibration time series data to obtain the multiple vibration time series preprocessing data. The multiple vibration time series preprocessing data is finally stored in the database layer 40.

[0235] S53: calling the feature extraction unit 302 to extract features from the multiple vibration time series preprocessing data, and obtaining the multiple vibration feature matrix corresponding to the multiple vibration time series preprocessing data. The multiple vibration feature matrix is ​​finally stored in the database layer 40.

[0236] S54: calling the state prediction unit 303 to obtain the feature difference between two adjacent processings of the CNC machine tool in the previously set number of processings according to the multiple vibration feature matrix. The feature difference between two adjacent processings of the previously set number of processings is finally stored in the database layer 40.

[0237] S55: updating the current difference threshold according to the feature difference between two adjacent processings of the previously set number of times.

[0238] In an embodiment of the present application, due to the differences in workpieces during the processing of the current tool, the current difference threshold may not be able to distinguish the state of the current tool well. Therefore, after each set number of processing times, the current difference threshold will be updated to improve the accuracy of tool state judgment.

[0239] The following describes in detail how to update the current difference threshold.

[0240] See also Fig.11 , is a flowchart of a method for updating the current difference threshold provided in an embodiment of the present application. Step S55 can be implemented by executing sub-steps S551 to S554:

[0241] S551: Obtain the feature differences corresponding to two adjacent processings in the previous several processing processes, and construct a similarity set.

[0242] S552: Obtain a threshold to be updated according to the mean and variance of the feature differences corresponding to the similarity set and the size parameters of the current tool.

[0243] S553: ​​Determine whether the threshold to be updated is greater than the initial difference threshold.

[0244] S554: If it is greater than, the threshold to be updated is used as the current difference threshold; if it is not greater than, the initial difference threshold is used as the current difference threshold.

[0245] In the embodiment of the present application, for several previous machining processes, the overall similarity mean and overall similarity variance corresponding to multiple feature differences can be determined based on the feature difference between the vibration time series preprocessing data corresponding to two adjacent machining processes. Then, the threshold to be updated is determined based on the overall similarity mean, the overall similarity variance and the size parameters of the current tool. The calculation formula for the threshold to be updated is shown in (28):

[0246] y=μ+cl*σ(28)

[0247] Wherein, y represents the threshold to be updated, μ represents the mean of the characteristic difference corresponding to the difference set, σ represents the variance of the characteristic difference corresponding to the difference set, and cl represents the size parameter of the tool. Of course, in the above embodiment, the standard deviation can also be used to replace the variance, and this application does not impose any special restrictions on this.

[0248] If the threshold to be updated is greater than the initial difference threshold obtained by previous training, the threshold to be updated is used as the new current difference threshold. It should be understood that the judgment mechanism for tool abnormality in this application is that the current feature difference is greater than the current difference threshold. In this case, when the threshold to be updated is greater than the initial difference threshold obtained by previous training, the threshold to be updated is used as the new current difference threshold. This means that the current feature difference needs to be greater than the threshold to be updated before it is judged as a tool abnormality. Here, a larger current difference threshold is used as the standard for judging tool abnormality, which changes the detection adaptability and better detects the tool status.

[0249] If the threshold to be updated is not greater than the initial difference threshold obtained by previous training, the initial difference threshold is used as the new current difference threshold, that is, the lower limit value of the threshold is used to detect the tool state.

[0250] In some embodiments, after the state prediction unit 303 in the data processing layer 30 outputs the tool state prediction result, if the tool state prediction result is that the tool is abnormal, then the tool needs to be replaced in time.

[0251] See also Fig.12 , is an architecture diagram of another prediction system provided in an embodiment of the present application, such as Fig.12 As shown, as a possible implementation, the prediction system may further include an application layer 60, which includes an automatic tool changing control unit 601, which is used to generate and issue automatic tool changing instructions according to the current state of the tool to control the tool changing robot to perform tool changing operations, that is, to control the tool changing robot to automatically change tools, thereby improving the efficiency and timeliness of tool changing.

[0252] In some embodiments, it is considered that a user may need to view the collected vibration time series data.

[0253] As a possible implementation, the application layer 60 also includes a front-end UI unit 602, which is used for visualizing the data of the prediction system. For example, the vibration time series data will be simply processed (reducing the collection frequency, averaging, median and other statistical methods), and the data will be presented in three axes + timestamp. Of course, other intermediate data can also be displayed, and this application does not impose any special restrictions on this.

[0254] The above description is only a preferred embodiment of this specification and is not intended to limit this specification. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of this specification should be included in the scope of protection of this specification.

Claims

1. A tool state prediction system, characterized in that: The prediction system comprises: The data acquisition layer is used to collect the vibration time series data of the tool in the CNC machine tool; The data storage layer includes a time series data storage unit, a preprocessing data storage unit and a feature data storage unit; wherein the time series data storage unit is used to store the vibration time series data, the preprocessing data storage unit is used to store the vibration time series preprocessing data, and the feature data storage unit is used to store the vibration time series feature data; The data processing layer includes a preprocessing unit, a feature extraction unit and a state prediction unit; wherein the preprocessing unit is used to preprocess the vibration time series data to obtain the vibration time series preprocessed data; the feature extraction unit is used to extract features from the vibration time series preprocessed data to obtain the vibration time series feature data; the state prediction unit is used to predict and obtain the current state of the tool based on the vibration time series feature data.

2. The tool state prediction system according to claim 1, characterized in that: The prediction system also includes: A database layer, used for storing data of the data storage layer; The model automatic deployment layer includes a continuous integration unit and a continuous delivery unit; wherein the continuous integration unit is used to train and obtain a tool state prediction model based on the data in the database layer; and the continuous delivery unit is used to deploy the tool state prediction model in the data processing layer.

3. The tool state prediction system according to claim 1, characterized in that: The prediction system also includes: The application layer includes a front-end UI unit and an automatic tool-changing control unit; wherein the automatic tool-changing control unit is used to generate and issue automatic tool-changing instructions according to the current state of the tool to control the tool-changing robot to perform tool-changing operations; the front-end UI unit is used to visualize the data of the prediction system.

4. A method for predicting tool status, characterized in that: The method applied to the tool state prediction system according to any one of claims 1 to 3 comprises: The preprocessing unit reads the vibration time series data in the time series data storage unit, preprocesses the vibration time series data to obtain the vibration time series preprocessing data, and stores the vibration time series preprocessing data in the preprocessing data storage unit; The feature extraction unit reads the vibration time series preprocessing data in the preprocessing data storage unit, performs feature extraction on the vibration time series preprocessing data to obtain the vibration time series feature data, and stores the vibration time series feature data in the feature data storage unit; The state prediction unit reads the vibration time series characteristic data in the characteristic data storage unit, and predicts and acquires the current state of the tool according to the vibration time series characteristic data.

5. The method for predicting tool status according to claim 4, characterized in that: The method further comprises: The preprocessing unit responds to the update threshold instruction of the continuous delivery unit, reads the multiple vibration time series data in the database layer, preprocesses the multiple vibration time series data to obtain the multiple vibration time series preprocessed data, and stores the multiple vibration time series preprocessed data to the database layer; The feature extraction unit reads the multiple vibration time series preprocessing data in the database layer, performs feature extraction on the multiple vibration time series preprocessing data to obtain a multiple vibration feature matrix corresponding to the multiple vibration time series preprocessing data, and stores the multiple vibration feature matrix in the database layer; The state prediction unit reads the multiple vibration feature matrix in the database layer, and obtains the feature difference corresponding to two adjacent times of the machining process of the CNC machine tool for a previously set number of times according to the multiple vibration feature matrix; The continuous delivery unit updates the current difference threshold according to the feature differences corresponding to two adjacent processings of the previously set number of times.

6. The method for predicting tool status according to claim 5, characterized in that: The updating of the current difference threshold according to the feature difference between two adjacent times of the processing of the previously set number of times includes: Obtain the feature differences between two adjacent processings in the previous several times and construct a similarity set; Obtaining a threshold to be updated according to the mean and variance of the feature differences corresponding to the similarity set and the size parameters of the current tool; Determine whether the threshold to be updated is greater than the initial difference threshold; If it is greater than, the threshold to be updated is used as the current difference threshold; If not, the initial difference threshold is used as the current difference threshold.

7. The method for predicting tool status according to claim 6, characterized in that: The method further comprises: The preprocessing unit responds to the model training instruction of the continuous integration unit, reads the previous vibration time series data and tool status data generated by the tool during the historical machining process of the CNC machine tool in the database layer, preprocesses the previous vibration time series data to obtain previous vibration time series preprocessing data, and stores the previous vibration time series preprocessing data in the database layer; The feature extraction unit reads the previous vibration time series preprocessing data in the database layer, extracts the previous vibration feature matrices corresponding to the previous vibration time series preprocessing data, and stores the previous vibration feature matrices in the database layer; The state prediction unit reads the previous vibration feature matrix in the database layer, obtains the feature difference corresponding to two adjacent times in the historical processing of the CNC machine tool according to the previous vibration feature matrix, and constructs a previous feature difference set; The continuous integration unit aims to enable the initial difference threshold to distinguish between the abnormal state and the normal state of the tool. According to the previous feature difference sets and the tool state, the initial difference threshold is iteratively adjusted until the iteration is completed to obtain the initial difference threshold; wherein the initial value of the current difference threshold is the initial difference threshold.

8. The method for predicting tool status according to claim 4, characterized in that: The step of extracting features from the vibration time series preprocessed data to obtain the vibration time series feature data comprises: Preprocessing the previous vibration time series data and the current vibration time series data generated by the current tool of the CNC machine tool during the machining process to obtain the previous vibration time series preprocessing data and the current vibration time series preprocessing data; wherein the vibration time series data includes the previous vibration time series data and the current vibration time series data; The step of the feature extraction unit acquiring the vibration time series feature data comprises: According to the previous vibration time sequence preprocessing data and the current vibration time sequence preprocessing data, respectively obtaining the vibration characteristic matrix corresponding to the previous vibration time sequence preprocessing data and the current vibration time sequence preprocessing data; wherein the vibration time sequence characteristic data includes the vibration characteristic matrix; The predicting and acquiring the current state of the tool according to the vibration time series characteristic data includes: According to the vibration feature matrix corresponding to the previous vibration time series preprocessing data and the current vibration time series preprocessing data, obtaining the current feature difference between the previous machining process and the current machining process of the current tool; Based on the comparison result of the current feature difference and the current difference threshold, the current state of the current tool is determined.

9. The method for predicting tool status according to claim 8, characterized in that: The step of obtaining the vibration characteristic matrix corresponding to the previous vibration time series preprocessing data and the current vibration time series preprocessing data includes: Based on the first target frequency band, the previous vibration time series preprocessing data and the current vibration time series preprocessing data are filtered respectively; Based on the filtered previous vibration time series preprocessing data and the current vibration time series preprocessing data, vibration time domain features and vibration frequency domain features corresponding to the previous vibration time series preprocessing data and the current vibration time series preprocessing data are extracted respectively.

10. The method for predicting tool status according to claim 9, characterized in that: The method of extracting vibration time domain features and vibration frequency domain features corresponding to the previous vibration time sequence preprocessing data and the current vibration time sequence preprocessing data based on the filtered previous vibration time sequence preprocessing data and the current vibration time sequence preprocessing data respectively includes: Determining the fundamental frequency of the time-frequency domain conversion according to the set rotation speed of the current tool; Determining the maximum frequency multiple of the time-frequency domain conversion according to the number of blades of the current tool and the fundamental frequency; Determine at least two second target frequency bands for time-frequency domain conversion according to the base frequency and the maximum multiple frequency; The filtered previous vibration timing preprocessing data and the current vibration timing preprocessing data are respectively converted into time-frequency domains, and the vibration frequency domain class features corresponding to each second target frequency band in the previous vibration timing preprocessing data and the current vibration timing preprocessing data are respectively obtained.

11. The method for predicting tool status according to claim 8, characterized in that: The determining the current state of the current tool based on the comparison result of the current feature difference and the current difference threshold comprises: Determine whether the current feature difference is greater than the current difference threshold; wherein the current feature difference is the Euclidean distance between the vibration feature matrices corresponding to the previous vibration time series preprocessing data and the current vibration time series preprocessing data; If it is greater than the current difference threshold, it is determined that the current tool is in an abnormal state; If it is not greater than the current difference threshold, it is determined that the current tool is in a normal state.

Images