A numerical control machining arbitrary tooth number cutter vibration signal adaptive identification analysis method

Abstract

The application discloses a kind of numerical control processing arbitrary tooth number cutter vibration signal adaptive identification analysis method, belong to aviation manufacturing field, including the following steps: based on vibration sensor obtains cutter vibration data, parses vibration data and draws vibration data image v_image, design vibration image feature descriptor realizes feature extraction, based on the feature difference evaluation function designed in two-dimensional space feature distribution map obtains feature difference key point, according to adaptive regional threshold value judging mode and neural network identification mode obtains identification result, fusion result and comparison obtain signal output.The application is suitable for analyzing and identifying the wear amount, broken tooth, missing tooth, missing and other abnormalities or defects of cutter tooth, based on adaptive identification and result fusion analysis, can further improve the accuracy of tool abnormality identification, reduce false alarm rate and other technical indicators, reduce the possibility of workpiece scrap caused by tool abnormality, thereby avoiding or reducing economic loss.

Description

Technical Field

[0001] This invention relates to the field of aircraft technology, and more specifically to an adaptive identification and analysis method for vibration signals of CNC machining tools with arbitrary tooth counts. Background Technology

[0002] In the aerospace field, titanium alloy structural components are characterized by small batches and multiple categories. At the same time, titanium alloy components are characterized by high hardness, complex features, a variety of tool types required for the entire machining process, complex and variable machining trajectories, and great influence from complex time-varying working conditions.

[0003] If the time-frequency domain feature analysis is performed directly on the tool vibration data acquired by the sensor to determine whether there is a tool abnormality, such as abnormality diagnosis, machine tool emergency stop, and high load alarm, the accuracy of the judgment is usually low, with many false alarms and missed alarms. It cannot meet the requirements for monitoring and identifying tool abnormalities such as tooth breakage and wear during the machining of aircraft structural parts, especially titanium alloy structural parts.

[0004] Meanwhile, the existing mainstream time-frequency domain feature analysis methods cannot meet the monitoring needs because the vibration signal data obtained by the sensor has characteristics such as high frequency and significant changes in local time domain values. Analyzing existing methods for identifying and analyzing tool vibration signals, the transformation of high-frequency vibration signals, such as normalization, mean, variance, and entropy, is the core of vibration data feature extraction, and monitoring is completed based on designed judgment criteria. However, existing methods cannot adaptively analyze data with different changing characteristics, demonstrating poor universality. Summary of the Invention

[0005] The purpose of this invention is to provide an adaptive identification and analysis method for vibration signals of CNC machining tools with arbitrary tooth counts, enabling the monitoring of machining tools for titanium alloy parts under complex and variable working conditions. The method performs identification and analysis based on an adaptively changing mathematical model for each time series, thereby solving the problem of poor adaptability of existing methods. At the same time, it provides a new approach and reference method for online monitoring of CNC machining.

[0006] To achieve the above-mentioned technical effects, the technical solution of the present invention is as follows:

[0007] An adaptive identification and analysis method for vibration signals of CNC machining tools with arbitrary tooth counts includes the following steps:

[0008] The tool vibration data is acquired and stored in the form of a dictionary (dic). The key of the dictionary represents the time unit in seconds, and the corresponding value represents the data collected per unit time. The corresponding value includes the vibration amount of the vibration signal in the X-axis and Y-axis directions.

[0009] Parse the vibration data and plot the vibration data image v_image. Take the time interval data of length prior to the current time time_temp as the raw data for plotting the vibration image. Compare the maximum value on the X and Y axes of each key stored in the value within the corresponding time interval. Plot the response amplitude using max. value This indicates that the valid data extracted within the corresponding time period is represented by the string "data". length This indicates that the v_image image is drawn based on this data;

[0010] Design a vibration image feature descriptor to achieve statistical analysis of the feature difference distribution image across all angles, using the center position (num) as the criterion. c / 2,num r / 2) Start statistical analysis of the number of valid feature data points distributed on the structure tree corresponding to each offset angle. The object being described is the v_image image. Complete the statistical description analysis of the full-angle features of the v_image image recorded in the structure tree.

[0011] Based on statistical data, the feature data is plotted in a two-dimensional feature space;

[0012] Design a feature difference evaluation function and obtain key features plotted in a two-dimensional feature space. Based on the designed evaluation function, calculate and obtain key point feature data of feature differences for any vibration data in a two-dimensional spatial feature distribution image.

[0013] Design a key point feature data mapping method and implement data feature image drawing. The obtained key point feature data is used to back-map the corresponding feature positions to the data feature map drawn based on the original vibration data, and the image is drawn to obtain the corresponding intuitive result image.

[0014] The relative distribution characteristics of vibration data corresponding to different cutting teeth in the original image v_image can be obtained based on the corresponding regions in the image. The cutting tooth feature data are determined according to the number of effective pixels corresponding to different regions in image v_image. Let the number of effective pixels in the corresponding regions S1, S2, ..., Sn be denoted as size1, size2, ..., size n ;

[0015] An adaptive judgment method is designed so that for each image generated at each time interval, the corresponding size1, size2, ..., size can be calculated. n Numerical value;

[0016] If there is a situation where the relative area distribution is smaller than the maximum area, the recognition network needs to be activated; otherwise, the recognition result 1 is obtained directly. Recognition result 1 indicates that the corresponding tool does not have any abnormalities during this period.

[0017] Features other than vibration data points in the network sample are considered noise in the image and are removed accordingly.

[0018] If the output of the recognition network is ∈[1, n], the corresponding recognition result is recognition result 2;

[0019] The fusion recognition result 1 and recognition result 2 are compared with the corresponding number of cutting teeth in the CNC program to obtain the final recognition output signal.

[0020] Furthermore, based on data length If the size of the corresponding time period (time_temp) is less than the length, then no image needs to be drawn, and the state of the tool is not analyzed within the corresponding time period.

[0021] Furthermore, if the corresponding time `time_temp` is greater than or equal to the length `length`, then a vibration image is drawn. The vibration image is drawn as follows:

[0022] 1) Using the OpenCV image processing library, create a new image of size (num... c ,num r A single-channel image v_image, where num c ,num r These represent the number of columns and rows of the image, respectively. To ensure that the acquired vibration image is conducive to feature analysis, the number of rows and columns of the image should be equal.

[0023] 2) Based on S speed and V fre Determine the radians and angles u corresponding to each hz data point. angle The center position of the image v_image (num c / 2,num r / 2) The image is plotted starting from the origin of the vibration data. The offset of each Hz data point along the corresponding angular direction is the maximum value of the X and Y axes in the corresponding vibration data. value If a point is drawn on (num c / 2,num r / 2) The position indicates that the corresponding vibration value is zero;

[0024] The corresponding rotational speed of the machine tool during the machining process is used S speed The unit is a positive integer; the sampling frequency corresponding to the vibration sensor is... V fre The unit is Hz, and the corresponding value is read from the CNC program;

[0025] 3) Using the time data at time (temp-length) as a correspondence to zero degrees, the center position (num) is the zero degree of the angle in the image v_image. c / 2,num r / 2) The horizontal direction to the right and the angle are combined and increase in a counterclockwise direction;

[0026] 4) In the image v_image, sequentially overlay and plot the maximum amplitude of vibration at the corresponding cumulative angle for each hz data point. value This continues until the data collected by the vibration sensor within the time range of time_temp-length is plotted.

[0027] 5) Perform scale unification processing on all data points in the image v_image to obtain a new unified scale map;

[0028] 6) After generating the vibration data image within the time range of time_temp-length corresponding to time_temp, the image corresponding to time_temp+1 needs to be generated, with the corresponding time range being (time_temp+1)-length. The images are generated sequentially to achieve online real-time tool status monitoring.

[0029] Furthermore, the vibration image feature descriptor is designed as follows:

[0030]

[0031] Where D(i) represents the feature descriptor structure tree; i represents the angle corresponding to the feature in the structure tree, i∈[0,360); flag represents a temporary statistical marker, each i initially corresponds to flag=0, and if the if condition is met, the corresponding flag=1; init represents initialization; h(X-axis, Y-axis, i) represents all pixels corresponding to the i in the structure tree; pixel gray This represents the color value of the non-background pixels in the drawn image v_image.

[0032] Furthermore, based on the statistical data, the feature data is plotted in a two-dimensional feature space. In the feature space, the corresponding horizontal coordinate represents the angle range value, the corresponding vertical coordinate represents the size of the feature data recorded in the structure tree, and the intersection of the horizontal and vertical coordinates is the origin.

[0033] Furthermore, the function of the feature difference evaluation function is to identify key point features in the curve of the distribution map, so as to achieve dimensionality reduction and abstract expression of the features. The feature difference evaluation function is expressed as:

[0034]

[0035] Where teech represents the number of teeth of the tool that needs to be identified; for indicates loop processing; Represents any numerical value; The symbol indicates the existence of a sign; f(X, Y) represents a function that satisfies the correspondence between statistical data points; the X-axis... beg This represents the initial x-coordinate value of the feature space.

[0036] Furthermore, the key point data mapping method in the design is as follows:

[0037] 1) Determine the angle value in the original image based on x1 in the initial (x1, y1) and create a new single-channel image pic0 with a pure white background and the same scale. Determine the initial angle value corresponding to the new image pic0 based on the initial angle selected when drawing the two-dimensional spatial feature distribution map. Draw a ray from the center of the image based on the angle value determined by x1. The angle of the ray is consistent with x1.

[0038] 2) When drawing the image based on the data features, draw a circle in the pic0 image with the radius corresponding to the normalized distributed data points in the image, and the center of the circle is the center of the image;

[0039] 3) Starting with the initial value of x1, find the feature data points x1 + int(360 / teech)*1, x1 + int(360 / teech)*2, x1 + int(360 / teech)*3, ..., x1 + int(360 / teech)*n respectively with a step size of int(360 / teech), and draw a ray in the image pic0 based on each corresponding point;

[0040] 4) Calculate the coordinates of the intersection points of all rays on the same circle in the image pic0, denoted as p1(x,y), p2(x,y), ... p n (x,y), based on the numerical value of the intersection point, the ray is truncated and the region where the circle is located in the image pic0 is divided into n equal parts S1, S2, ..., Sn.

[0041] Furthermore, the effective pixel count is calculated as follows:

[0042] In the image v_image, for any feature region Si, if the vibration value at a certain moment falls within the corresponding Si region, the number of valid pixels is incremented by one; if there are vibration data points at other moments that also fall within the corresponding position of the Si region, the value of the valid data point is not incremented by one; until all points in the Si region are counted, the sum of the position data that satisfy the relationship at all corresponding points is the number of valid pixels in the corresponding region Si, which is also the feature representation of the corresponding region.

[0043] Furthermore, the adaptive judgment design method is as follows: for size1, size2, ..., size... n Sort the areas in ascending order to find the largest value; divide each area by the largest value, size. max The corresponding ratio contains values ​​less than half of the maximum area, enabling adaptive judgment.

[0044] Furthermore, after noise removal, the sample image is binarized. The background of the processed image is black, and the grayscale corresponding to the regions with vibration data characteristics is white.

[0045] Furthermore, the recognition result 1 and recognition result 2 are fused to obtain the fused result. If the recognition network is not activated, the corresponding output of result 2 is 0, and the fused result is based on result 1. If the recognition network is activated, it indicates that the tool has potential risks and abnormalities, and the recognition result 2 is based on result 2.

[0046] Furthermore, the value of the number of teeth of the corresponding tool in the CNC machining program is read and compared with the fusion result, and a signal output result is obtained. If the corresponding values ​​are consistent, it means that the tool has no abnormality and machining can continue; if the corresponding values ​​are inconsistent, it means that the corresponding tool has an abnormality and a stop machining signal should be output.

[0047] Furthermore, the initial value of the horizontal axis corresponds to an angle value that is any angle within the entire angle range, and the corresponding number is either a positive integer or a decimal.

[0048] The advantages of this application are:

[0049] 1. This invention is applicable to analyzing and identifying anomalies or defects in cutting tool teeth, such as wear, broken teeth, missing teeth, and other abnormalities. Based on adaptive identification and result fusion analysis, it can further improve the accuracy of tool anomaly identification, reduce the false alarm rate, and lower the possibility of workpiece scrap due to tool anomalies, thereby avoiding or reducing economic losses. Furthermore, the method designed in this invention is highly applicable to tools with different numbers of teeth. Based on the design concept proposed in this invention, corresponding identification and analysis modules can be quickly developed, improving the method's adaptability to various scenarios and reducing development costs. Attached Figure Description

[0050] Figure 1 A method for adaptive identification and analysis of vibration signals from CNC machining tools with arbitrary tooth counts.

[0051] Figure 2 Explanation of the zero degree and the direction of increase of angle in image v_image.

[0052] Figure 3 All data points in the image v_image are scaled to obtain a scaled map.

[0053] Figure 4 Based on statistical data, feature data are plotted in a two-dimensional feature space.

[0054] Figure 5 The evaluation function based on the design obtains key points of feature difference in a two-dimensional spatial feature distribution image.

[0055] Figure 6 Divide the region containing the circle in image pic0 into n equal parts S1, S2, ..., Sn. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0057] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0058] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0059] In the description of this application, it should be noted that the terms "upper," "vertical," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product is in use, or the orientation or positional relationship commonly understood by those skilled in the art. They are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on this application. In addition, the terms "first," "second," etc., are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0060] In the description of this application, it should also be noted that, unless otherwise expressly specified and limited, the terms "set," "install," and "connect" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0061] Example 1

[0062] An adaptive identification and analysis method for vibration signals of CNC machining tools with arbitrary tooth counts includes the following steps:

[0063] The tool vibration data is acquired and stored in the form of a dictionary (dic). The key of the dictionary represents the time unit in seconds, and the corresponding value represents the data collected per unit time. The corresponding value includes the vibration amount of the vibration signal in the X-axis and Y-axis directions.

[0064] Parse the vibration data and plot the vibration data image v_image. If the unit time for plotting the vibration image is time_temp, the corresponding time is taken to the second and milliseconds are omitted. The time interval data of length length before the current time time_temp is taken as the raw data for plotting the vibration image. Compare the maximum value of each key stored in the value on the X and Y axes within the corresponding time interval. Plot the response amplitude using max. value This indicates that the valid data extracted within the corresponding time period is represented by the string "data". length This indicates that the v_image image is drawn based on this data;

[0065] Design a vibration image feature descriptor to achieve statistical analysis of the feature difference distribution image across all angles, using the center position (num) as the criterion. c / 2,num r / 2) Start statistical analysis of the number of valid feature data points distributed on the structure tree corresponding to each offset angle. The object being described is the v_image image. Complete the statistical description analysis of the full-angle features of the v_image image recorded in the structure tree.

[0066] Based on statistical data, the feature data is plotted in a two-dimensional feature space;

[0067] Design a feature difference evaluation function and obtain key features plotted in a two-dimensional feature space. Based on the designed evaluation function, calculate and obtain key point feature data of feature differences for any vibration data in a two-dimensional spatial feature distribution image. The corresponding data points are denoted as (x1, y1), (x2, y2), (x3, y3), ..., (x... n y n The value of n is determined by the number of teeth corresponding to the cutting tool;

[0068] Design a key point feature data mapping method and implement data feature image drawing. The obtained key point feature data is used to back-map the corresponding feature positions to the data feature map drawn based on the original vibration data, and the image is drawn to obtain the corresponding intuitive result image.

[0069] Based on the corresponding region in image pic0, the relative distribution characteristics of vibration data corresponding to different cutting teeth in the original image v_image can be obtained. The cutting tooth feature data are determined according to the number of effective pixels corresponding to different regions in image v_image. Let the number of effective pixels in the corresponding regions S1, S2, ..., Sn be denoted as size1, size2, ..., size... n ;

[0070] An adaptive judgment method is designed so that for each image generated at each time interval, the corresponding size1, size2, ..., size can be calculated. n Numerical value;

[0071] If there is a situation where the relative area distribution is smaller than the maximum area, the recognition network needs to be activated; otherwise, the recognition result 1 is obtained directly. Recognition result 1 indicates that the corresponding tool does not have any abnormalities during this period.

[0072] Features other than vibration data points in the network sample are considered noise in the image. Since the differences between the features of the vibration data images are small, the corresponding noise should be removed in order to highlight the features of the vibration data points and improve the recognition ability of the classification network.

[0073] If the output of the recognition network is ∈[1, n], the corresponding recognition result is recognition result 2;

[0074] The fusion recognition result 1 and recognition result 2 are compared with the corresponding number of cutting teeth in the CNC program to obtain the final recognition output signal.

[0075] Example 2

[0076] An adaptive identification and analysis method for vibration signals of CNC machining tools with arbitrary tooth counts, the method comprising:

[0077] Based on a vibration sensor installed near the tool on the machine tool spindle, vibration data of the tool is acquired during the machining process. The vibration data mainly consists of the X and Y displacements of the sensor at a given time interval.

[0078] The corresponding rotational speed of the machine tool during the machining process is used S speed The unit is a positive integer; the sampling frequency corresponding to the vibration sensor is... V fre The unit is Hz, and the corresponding value is read from the CNC program.

[0079] The data collected by the sensor is stored in the form of a dictionary (dic). The key in the dictionary represents the time unit in seconds, and the corresponding value represents the data collected per unit time. The corresponding value includes the vibration amount of the vibration signal in the X and Y directions.

[0080] Parse the data stored by the vibration sensor and draw the vibration data image v_image. If the unit time for drawing the vibration image is time_temp, the corresponding time is taken in seconds and milliseconds are omitted.

[0081] The data spanning the current time (time_temp) backwards by a length is used as the raw data for plotting the vibration image. The maximum X and Y values ​​stored in the value for each key within the corresponding time period are compared, and the response amplitude is plotted using max. value This indicates that the valid data extracted within the corresponding time period is represented by the string "data". length This indicates that the v_image image is drawn based on this data.

[0082] Based on data length If the corresponding time, time_temp, is less than the length (i.e., a short period of time before the vibration sensor is initially started), then no image needs to be drawn, and the state of the tool is not analyzed within the corresponding time period.

[0083] If the corresponding time time_temp ≥ length (i.e., a period of time has elapsed since the vibration sensor was initially started), then a vibration image is drawn. The vibration image is drawn as follows:

[0084] Using the OpenCV image processing library, create a new image of size (num). c ,num r A single-channel image v_image, where num c ,num r These represent the number of columns and rows of the image, respectively. To ensure that the acquired vibration image is conducive to feature analysis, the number of rows and columns of the image should be equal.

[0085] based on S speed and V fre Determine the radians and angles u corresponding to each hz data point. angle The center position of the image v_image (num c / 2,num r / 2) The image is plotted starting from the origin of the vibration data. The offset of each hz data point along the corresponding angular direction is the maximum value of X and Y in the corresponding vibration data. value If a point is drawn on (num c / 2,num r / 2) The position indicates that the corresponding vibration value is zero;

[0086] Using the time data at time (temp-length) as a correspondence to zero degrees, the center position in the image v_image is the zero degree angle (num). c / 2,num r / 2) The horizontal direction to the right and the angle are combined and increase in a counterclockwise direction;

[0087] In the image v_image, the maximum amplitude of each hz data point at its corresponding accumulated angle ((time_temp-length) corresponds to zero degrees) is plotted sequentially. value This continues until the data collected by the vibration sensor within the time range of time_temp-length is plotted.

[0088] Since the images v_image generated in different time periods have the same size, but the maximum offset of the corresponding time_temp-length time range varies greatly, it is necessary to perform scale unification processing on all data points in the image v_image to obtain a new unified scale map. That is, the trend of vibration data in each image is the relative offset of the features within the time_temp-length time range.

[0089] After generating the vibration data image within the time range of time_temp-length corresponding to time_temp, the image corresponding to time_temp+1 needs to be generated, with the corresponding time range being (time_temp+1)-length. The images are generated sequentially to achieve online real-time tool status monitoring.

[0090] Design vibration image feature descriptors: using the center position (num) c / 2,num r / 2) Begin statistical analysis of the number of valid feature data points distributed on the tree structure corresponding to each offset angle. The corresponding descriptor can be expressed as:

[0091]

[0092] Where D(i) represents the feature descriptor structure tree; i represents the angle corresponding to the feature in the structure tree, i∈[0,360); flag represents a temporary statistical marker, each i initially corresponds to flag=0, and if the if condition is met, the corresponding flag=1; init represents initialization; h(x,y,i) represents all pixels corresponding to the i in the structure tree; pixel gray This represents the color value of the non-background pixels in the drawn image v_image.

[0093] Based on the vibration image feature descriptor, the statistical distribution of feature differences across all angles of the image is realized, recorded in the structure tree, and the features are described.

[0094] Based on statistical data, the feature data is plotted in a two-dimensional feature space. In the feature space, the corresponding horizontal axis represents the numerical range of angles, and the corresponding vertical axis represents the size of the feature data recorded in the structure tree. The intersection of the horizontal and vertical axes is the origin.

[0095] The initial value of the horizontal axis corresponds to an angle that can be any angle value within the entire angle range. The corresponding number can be a positive integer or a decimal. Regardless of the choice, it will not affect the features obtained by the evaluation function.

[0096] Design a feature difference evaluation function to obtain key features from the distribution map. The evaluation function aims to identify key point features from the curves in the distribution map, thereby achieving dimensionality reduction and abstract representation of the features. The corresponding evaluation function can be expressed as:

[0097]

[0098] Where teech represents the number of teeth of the tool that needs to be identified; for indicates loop processing; Represents any numerical value; The symbol indicates the existence of a sign; f(x, y) represents a function that satisfies the correspondence between statistical data points; x beg This represents the initial x-coordinate value in the feature space. The entire formula can be further interpreted as representing, for any chosen initial x-coordinate value in the feature space, the region within the globally equally spaced x-coordinate range where the numerical features are maximized.

[0099] The entire formula can be further interpreted as, in the feature space, for any selected initial abscissa value, the region within the globally equally spaced abscissa range where the numerical features have the greatest feature representation.

[0100] Based on the designed evaluation function, for any vibration data's two-dimensional spatial feature distribution image, the key point feature data of feature differences can be calculated and obtained. Let the corresponding data points be denoted as (x1, y1), (x2, y2), (x3, y3), ..., (x... n y n The value of n is determined by the number of teeth corresponding to the cutting tool.

[0101] Based on key point feature data, the corresponding feature locations are back-mapped to a data feature map drawn from the original vibration data, and the image is then plotted to obtain the corresponding intuitive result image. The designed key point feature data mapping method is as follows:

[0102] Based on the initial (x1, y1) x1, determine the angle value in the original image, and create a new single-channel image pic0 with a pure white background and the same scale. Based on the initial angle selected when drawing the two-dimensional spatial feature distribution map, determine the initial angle value corresponding to the new image pic0. Based on the angle value determined by x1, draw a ray from the center of the image, and the angle of the ray is consistent with x1.

[0103] When drawing images based on data features, circles are drawn in the pic0 image with the radii corresponding to the normalized distributed data points in the image, and the center of the corresponding circle is the center of the image.

[0104] Starting with the initial value of x1, find the feature data points x1 + int(360 / teech)*1, x1 + int(360 / teech)*2, x1 + int(360 / teech)*3, ..., x1 + int(360 / teech)*n respectively with a step size of int(360 / teech), and draw a ray in the image pic0 based on each corresponding point;

[0105] Calculate the coordinates of the intersection points of all rays on the same circle in the image pic0, labeled p1(x,y), p2(x,y), ... p n (x,y), based on the numerical value of the intersection point, the ray is truncated and the region where the circle is located in the image pic0 is divided into n equal parts S1, S2, ..., Sn.

[0106] The relative distribution characteristics of vibration data corresponding to different cutting teeth in the original image v_image can be obtained based on the corresponding region in image pic0. The cutting tooth feature data are determined based on the number of effective pixels corresponding to different regions in image v_image. The calculation method for the number of effective pixels is as follows:

[0107] In the image v_image, for any feature region Si, if the vibration value at a certain moment falls within the corresponding Si region, the number of effective pixels is incremented by one.

[0108] If vibration data points at other times also fall at the corresponding positions in the corresponding Si region, the value of the valid data points will not be incremented by one.

[0109] The process continues until all points in region Si are counted. The sum of the position data that satisfy the relationship at all corresponding points is the effective number of pixels in region Si, which is also the feature representation of the corresponding region.

[0110] Let the number of valid pixels in the corresponding regions S1, S2, ..., Sn be denoted as size1, size2, ..., size n .

[0111] An adaptive judgment method is designed so that for each image generated at each time interval, the corresponding size1, size2, ..., size can be calculated. n The numerical values ​​are different for each image, so a fixed threshold would significantly impact recognition accuracy. The adaptive judgment design involves adjusting the threshold values ​​for size1, size2, ..., size... n Sort the areas in ascending order to find the largest value; divide each area by the largest value, size. max The corresponding ratio contains values ​​less than half of the maximum area, enabling adaptive judgment.

[0112] If there is a situation where the relative area distribution is smaller than the maximum area, the recognition network needs to be activated; otherwise, the recognition result 1 is obtained directly. Recognition result 1 indicates that the corresponding tool does not have any abnormalities during this period.

[0113] The design identifies the network, and the overall architecture of the network adopts the feature extraction of convolutional neural network and uses softmax to classify vibration images. The number of classification categories is n, which is the same as the number of teeth of the cutting tool. During the training of the model, the weight parameters of the network are changed, and the trained weight parameters are loaded during prediction, so the parameters of the entire network remain unchanged.

[0114] The training phase of the model requires transformation processing of the image sample data, mainly including the following enhancement methods: horizontal flipping augmenters.Fliplr(1.0), with the corresponding probability set to 0.9;

[0115] Vertically flip `augmenters.Flipud(1.0)` with a probability of 1.0; rotate the image by (0, 360) degrees using `augmenters.Affine(rotate=(0, 360))`.

[0116] The affine transformation has a coefficient of variation in the X-axis direction ranging from (0.5, 1.3) and a coefficient of variation in the Y-axis direction ranging from (0.5, 1.3), defined as augmenters.Affine(scale={"x": (0.5, 1.0), "y": (0.5, 1.0)}).

[0117] Image overall scale transformation processing (0.5, 1.5) augmenters.Resize((0.5, 1.5)).

[0118] One or two enhancement strategies are randomly selected from the above methods to enhance the samples. The implementation method is: augmenters.SomeOf((1,2), which increases the number of samples while ensuring that there are differences in features between the samples.

[0119] In the denoising process of the enhanced sample images, all features other than the sample vibration data points are regarded as noise in the image. Since the differences between the vibration data image features are small, the corresponding noise should be removed in order to highlight the vibration data point features and improve the recognition ability of the classification network.

[0120] After noise removal, the sample image is binarized. The background of the processed image is black, and the grayscale corresponding to the regions with vibration data characteristics is white.

[0121] The obtained binary image is input into the network for training, and the corresponding weight file is obtained. Based on the weight file obtained during the training process, the weight parameters with stronger recognition and generalization ability are selected, and the recognition network of the prediction sample image is initialized based on the corresponding parameters.

[0122] For the prediction stage, if the recognition network needs to be started after the regional threshold judgment process, the obtained image v_image is binarized and de-noising processed and directly input into the recognition network to complete the prediction of the result. The corresponding output result is ∈[1,n], and the corresponding recognition result is recognition result 2.

[0123] The recognition result 1 and recognition result 2 are fused to obtain the fused result. If the recognition network is not activated, the corresponding output of result 2 is 0, and the fused result is based on result 1; if the recognition network is activated, it indicates that the tool has potential risks and abnormalities, and the recognition result 2 is based on result 2.

[0124] The signal output is obtained by comparing the value of the number of teeth of the corresponding tool in the CNC machining program with the fusion result. If the corresponding values ​​are consistent, it means that the tool has not been damaged or worn and can continue machining. If the corresponding values ​​are inconsistent, it means that the corresponding tool has been abnormal and a stop machining signal should be output to complete the entire identification and analysis process.

[0125] Example 3

[0126] The program of this invention is based on a 64-bit Windows 7 system, with an Intel(R) Xeon(R) W-22233.60GHz processor, 32GB of RAM, and a frequency of 3200MHz. The graphics card used for deep learning training and detection is an Nvidia-P2200 (5GB VRAM), with driver version 441.66 and CUDA version 10.2. The hard disk capacity is 256GB. The software implementation platform is based on Visual Studio 2019, with C++ as the corresponding programming language. The vibration sensor operates at a frequency of 1-10000Hz, and the corresponding image processing library is OpenCV 4.5.5, based on the Release x64 platform. The deep learning platform is based on the PyTorch framework, with Python version 3.7.1, PyTorch version 1.5.0, Torchvision version 0.6.0, sample annotation using folders, and the matrix operation tool library NumPy version 1.21.4 with SDK version 10.0.22000.0. The main steps of an adaptive identification and analysis method for vibration signals of CNC machining tools with arbitrary tooth count are as follows: Figure 1 As shown.

[0127] 1. Based on a vibration sensor installed near the tool on the machine tool spindle, vibration data of the tool during machining is acquired. The vibration data mainly consists of the X and Y displacements of the sensor at corresponding unit moments, changing over time. The corresponding rotational speed of the machine tool during machining is... S speed The unit is a positive integer; the sampling frequency corresponding to the vibration sensor is... V fre The unit is Hz, and the corresponding value is read from the CNC program.

[0128] 2. The data collected by the sensor is stored in the form of a dictionary (dic). The key of the dictionary represents the time unit in seconds, and the corresponding value represents the data collected per unit time. The corresponding value includes the vibration amount of the vibration signal in the X and Y directions.

[0129] 3. Parse the vibration data and plot the vibration data image v_image. If the unit time for plotting the vibration image is time_temp, the corresponding time is taken to the second, omitting the millisecond. Take the time interval data of length length preceding the current time time_temp as the raw data for plotting the vibration image. Compare the maximum value of X and Y in the value stored for each key within the corresponding time interval, and plot the response amplitude using max. valueThis indicates that the valid data extracted within the corresponding time period is represented by the string "data". length This indicates that the v_image image is drawn based on this data.

[0130] 4. Based on data length If the corresponding time, time_temp, is less than the length (i.e., a short period of time before the vibration sensor is initially started), then no image needs to be drawn, and the state of the tool is not analyzed within the corresponding time period.

[0131] 5. If the corresponding time time_temp ≥ length (i.e., a period of time has elapsed since the initial start-up of the vibration sensor), then a vibration image is plotted. The vibration data image is plotted as follows:

[0132] 1) Using the OpenCV image processing library, create a new image of size (num... c ,num r A single-channel image v_image, where num c ,num r These represent the number of columns and rows of the image, respectively. To ensure that the acquired vibration image is conducive to feature analysis, the number of rows and columns of the image should be equal.

[0133] 2) Based on S speed and V fre Determine the radians and angles u corresponding to each hz data point. angle The center position of the image v_image (num c / 2,num r / 2) The image is plotted starting from the origin of the vibration data. The offset of each hz data point along the corresponding angular direction is the maximum value of X and Y in the corresponding vibration data. value If a point is drawn on (num c / 2,num r / 2) The position indicates that the corresponding vibration value is zero;

[0134] 3) Using the time data at time (temp-length) as a correspondence to zero degrees, the center position (num) is the zero degree of the angle in the image v_image. c / 2,num r / 2) The horizontal direction to the right, combined with the angle, increases counterclockwise, such as... Figure 2 As shown;

[0135] 4) In the image v_image, sequentially overlay the maximum amplitude of each hz data point at its corresponding accumulated angle ((time_temp-length) corresponds to zero degrees). value This continues until the data collected by the vibration sensor within the time range of time_temp-length is plotted.

[0136] 5) Since the generated images v_image have the same size across different time periods, but the maximum offset within the corresponding time_temp-length range varies significantly, it is necessary to perform scale unification processing on all data points in image v_image to obtain a new unified scale map. That is, the trend of vibration data in each image represents the relative offset of features within the time_temp-length range. The corresponding image is shown below. Figure 3 As shown;

[0137] 6) After generating the vibration data image within the time range of time_temp-length corresponding to time_temp, the image corresponding to time_temp+1 needs to be generated, with the corresponding time range being (time_temp+1)-length. The images are generated sequentially to achieve online real-time tool status monitoring.

[0138] 6. Design vibration image feature descriptors: using the center position (num) c / 2,num r / 2) Begin statistical analysis of the number of valid feature data points distributed on the tree structure corresponding to each offset angle. The corresponding descriptor can be expressed as:

[0139]

[0140] Where D(i) represents the feature descriptor structure tree; i represents the angle corresponding to the feature in the structure tree, i∈[0,360); flag represents a temporary statistical marker, each i initially corresponds to flag=0, and if the if condition is met, the corresponding flag=1; init represents initialization; h(x,y,i) represents all pixels corresponding to the i in the structure tree; pixel gray This represents the color value of the non-background pixels in the drawn image v_image.

[0141] 7. Design a vibration image feature descriptor to realize the statistical analysis of the feature difference distribution image from all angles, record it in a structure tree, and describe the features.

[0142] 8. Based on statistical feature data, plot the feature data in a two-dimensional feature space, such as... Figure 4As shown in the diagram. In the feature space, the x-coordinate represents the numerical range of angles, and the y-coordinate represents the size of the feature data recorded in the structure tree. The intersection of the x-coordinate and y-coordinate is the origin. The initial value of the x-coordinate corresponds to any angle value within the entire angle range, and the corresponding number can be a positive integer or a decimal. Regardless of the choice, it will not affect the features obtained by the evaluation function.

[0143] 9. Design a feature difference evaluation function to obtain key features from the distribution map. The evaluation function identifies key point features from the curves in the distribution map, thereby achieving dimensionality reduction and abstract representation of the features. The corresponding evaluation function can be expressed as:

[0144]

[0145] Where teech represents the number of teeth of the tool that needs to be identified; for indicates loop processing; Represents any numerical value; The symbol indicates the existence of a sign; f(x, y) represents a function that satisfies the correspondence between statistical data points; x beg This represents the initial x-coordinate value in the feature space. The entire formula can be further interpreted as representing, for any chosen initial x-coordinate value in the feature space, the region within the globally equally spaced x-coordinate range where the numerical features are maximized.

[0146] 10. Based on the designed feature difference evaluation function, for any vibration data's two-dimensional spatial feature distribution image, the key point feature data of feature differences can be calculated and obtained. The corresponding recognition calculation effect is as follows: Figure 5 As shown, let the corresponding data points be denoted as (x1, y1), (x2, y2), (x3, y3), ..., (x... n y n The value of n is determined by the number of teeth corresponding to the cutting tool.

[0147] 11. Based on the key point feature data, the corresponding feature positions are back-mapped to a data feature map drawn based on the original vibration data, and the image is then plotted to obtain the corresponding intuitive result image. The designed key point feature data mapping method is as follows:

[0148] 1) Determine the angle value in the original image based on x1 in the initial (x1, y1) and create a new single-channel image pic0 with a pure white background and the same scale. Determine the initial angle value corresponding to the new image pic0 based on the initial angle selected when drawing the two-dimensional spatial feature distribution map. Draw a ray from the center of the image based on the angle value determined by x1. The angle of the ray is consistent with x1.

[0149] 2) When drawing the image based on the data features, draw a circle in the pic0 image with the radius corresponding to the normalized distributed data points in the image, and the center of the circle is the center of the image;

[0150] 3) Starting with the initial value of x1, find the feature data points x1 + int(360 / teech)*1, x1 + int(360 / teech)*2, x1 + int(360 / teech)*3, ..., x1 + int(360 / teech)*n respectively with a step size of int(360 / teech), and draw a ray in the image pic0 based on each corresponding point;

[0151] 4) Calculate the coordinates of the intersection points of all rays on the same circle in the image pic0, denoted as p1(x,y), p2(x,y), ... p n (x,y), based on the numerical values ​​of the intersection points, the ray is truncated, and the region containing the circle in image pic0 is divided into n equal parts S1, S2, ..., Sn, as follows. Figure 6 As shown.

[0152] 12. Based on Figure 6 The relative distribution characteristics of vibration data corresponding to different blades in the original image v_image can be obtained by corresponding regions. The blade feature data are determined based on the number of effective pixels corresponding to different regions in the image v_image. The number of effective pixels is calculated as follows:

[0153] In the image v_image, for any feature region Si, if the vibration value at a certain moment falls within the corresponding Si region, the number of valid pixels is incremented by one. If vibration data points at other moments also fall within the corresponding positions of the Si region, the number of valid data points is not incremented. This process continues until all points in the Si region have been counted. The sum of the position data satisfying the relationship at all corresponding points is the number of valid pixels in the corresponding region Si, which is also the feature representation of the corresponding region. Let the number of valid pixels in the corresponding regions S1, S2, ..., Sn be denoted as size1, size2, ..., size... n .

[0154] 13. Design an adaptive judgment method so that for each time interval of generated image, the corresponding size1, size2, ..., size can be calculated. n The numerical values ​​are different for each image, so a fixed threshold would significantly impact recognition accuracy. The adaptive judgment design involves adjusting the threshold values ​​for size1, size2, ..., size... n Sort the areas in ascending order to find the largest value; divide each area by the largest value, size. maxThe corresponding ratio contains values ​​less than half of the maximum area, enabling adaptive judgment.

[0155] 14. If there is a situation where the relative area distribution is smaller than the maximum area, the recognition network needs to be activated; otherwise, the recognition result 1 is obtained directly. Recognition result 1 indicates that the corresponding tool does not have any abnormalities during this period.

[0156] 15. Design a recognition network. The overall architecture of the network adopts convolutional neural network to extract features and uses softmax to classify vibration images. The number of classification categories is n, which is the same as the number of teeth of the cutting tool. During model training, the weight parameters of the network are changed, and during prediction, the trained weight parameters are loaded, so the overall network parameters remain unchanged.

[0157] 16. During the training phase of the model, the sample data of the image needs to be transformed, mainly including the following enhancement methods: horizontal flip augmenters.Fliplr(1.0), with a probability of 0.9; vertical flip augmenters.Flipud(1.0), with a probability of 1.0; image rotation range (0,360) degrees augmenters.Affine(rotate=(0,360)); affine transformation coefficients in the X-axis direction are in the range (0.5,1.3), and coefficients in the Y-axis direction are in the range (0.5,1.3), augmenters.Affine(scale={"x": (0.5, 1.0), "y": (0.5, 1.0)}); overall image scale transformation processing (0.5,1.5) augmenters.Resize((0.5,1.5)). One or two enhancement strategies are randomly selected from the above methods to enhance the samples. The implementation method is: augmenters.SomeOf((1,2), which increases the number of samples while ensuring that there are differences in features between the samples.

[0158] 17. Denoising of enhanced sample images: In this invention, all features other than the sample vibration data points are regarded as noise in the image. Since the differences between the vibration data image features are small, in order to highlight the vibration data point features and improve the recognition ability of the classification network, the corresponding noise should be removed.

[0159] 18. After noise removal, the sample image is binarized. The background of the processed image is black, and the grayscale corresponding to the regions with vibration data characteristics is white.

[0160] 19. Input the obtained binary image into the network for training and obtain the corresponding weight file. Based on the weight file obtained during the training process, select the weight parameters with stronger recognition and generalization capabilities, and initialize the recognition network of the prediction sample image based on the corresponding parameters.

[0161] 20. For the prediction stage, if the recognition network needs to be started after the regional threshold judgment process, the obtained image v_image is binarized and de-noising processed and directly input into the recognition network to complete the prediction of the result. The corresponding output result is ∈[1, n], and the corresponding recognition result is recognition result 2.

[0162] 21. Fuse recognition result 1 and recognition result 2 to obtain the fused result. If the recognition network is not activated, the corresponding output of result 2 will be 0, and the fused result will be based on result 1; if the recognition network is activated, it indicates that the tool has potential risks and abnormalities, and recognition result 2 will be based on the fused result.

[0163] 22. Read the value of the number of teeth of the corresponding tool in the CNC machining program and compare it with the fusion result, and obtain the signal output result. If the corresponding values ​​are consistent, it means that the tool has not been damaged or worn and can continue to be processed. If the corresponding values ​​are inconsistent, it means that the corresponding tool has been abnormal and a stop processing signal should be output to complete the entire identification and analysis process.

Claims

1. An adaptive identification and analysis method for vibration signals of CNC machining tools with arbitrary tooth counts, characterized in that: Includes the following steps: The tool vibration data is acquired and stored in the form of a dictionary (dic). The key of the dictionary represents the time unit in seconds, and the corresponding value represents the data collected per unit time. The corresponding value includes the vibration amount of the vibration signal in the X-axis and Y-axis directions. Parse the vibration data and plot the vibration data image v_image. Take the time interval data of length prior to the current time time_temp as the raw data for plotting the vibration image. Compare the maximum value on the X and Y axes of each key stored in the value within the corresponding time interval. Plot the response amplitude using max. value This indicates that the valid data extracted within the corresponding time period is represented by the string "data". length This indicates that the v_image image is drawn based on this data; Design a vibration image feature descriptor to achieve statistical analysis of the feature difference distribution image across all angles, using the center position (num) as the criterion. c / 2,num r / 2) Start statistical analysis of the number of valid feature data points distributed on the structure tree corresponding to each offset angle. The object being described is the v_image image. Complete the statistical description analysis of the full-angle features of the v_image image recorded in the structure tree. Based on statistical data, the feature data is plotted in a two-dimensional feature space; Design a feature difference evaluation function and obtain key features plotted in a two-dimensional feature space. Based on the designed evaluation function, calculate and obtain key point feature data of feature differences for any vibration data in a two-dimensional spatial feature distribution image. Design a key point feature data mapping method and implement data feature image drawing. The obtained key point feature data is used to back-map the corresponding feature positions to the data feature map drawn based on the original vibration data, and the image is drawn to obtain the corresponding intuitive result image. The relative distribution characteristics of vibration data corresponding to different cutting teeth in the original image v_image can be obtained based on the corresponding regions in the image. The cutting tooth feature data are determined according to the number of effective pixels corresponding to different regions in image v_image. Let the number of effective pixels in the corresponding regions S1, S2, ..., Sn be denoted as size1, size2, ..., size n ; An adaptive judgment method is designed so that for each image generated at each time interval, the corresponding size1, size2, ..., size can be calculated. n Numerical value; If there is a situation where the relative area distribution is smaller than the maximum area, the recognition network needs to be activated; otherwise, the recognition result 1 is obtained directly. Recognition result 1 indicates that the corresponding tool does not have any abnormalities during this period. Features other than vibration data points in the network sample are considered noise in the image and are removed accordingly. If the output of the recognition network is ∈[1, n], the corresponding recognition result is recognition result 2; The recognition result 1 and recognition result 2 are fused and compared with the corresponding number of cutting teeth in the CNC program to obtain the final recognition output signal; If the corresponding time `time_temp` is greater than or equal to the length `length`, then a vibration image is drawn. The vibration image is drawn as follows: 1) Using the OpenCV image processing library, create a new image of size (num... c ,num r A single-channel image v_image, where num c ,num r These represent the number of columns and rows of the image, respectively. To ensure that the acquired vibration image is conducive to feature analysis, the number of rows and columns of the image should be equal. 2) Based on S speed and V fre Determine the corresponding radians and angles u for each hz data point. angle The center position of the image v_image (num c / 2,num r / 2) The image is plotted starting from the origin of the vibration data. The offset of each Hz data point along the corresponding angular direction is the maximum value of the X and Y axes in the corresponding vibration data. value If a point is drawn on (num c / 2,num r / 2) The position indicates that the corresponding vibration value is zero; The corresponding rotational speed of the machine tool during the machining process is used S speed The unit is a positive integer; the sampling frequency corresponding to the vibration sensor is... V fre The unit is Hz, and the corresponding value is read from the CNC program; 3) Using the time_temp-length data as a reference to zero degrees, the zero-degree angle in the image v_image is taken as the center position (num). c / 2,num r / 2) The horizontal direction to the right and the angle are combined and increase in a counterclockwise direction; 4) In the image v_image, sequentially overlay the maximum amplitude of vibration at the corresponding cumulative angle for each hz data point. value This continues until the data collected by the vibration sensor within the time range of time_temp-length is plotted. 5) Perform scale unification processing on all data points in the image v_image to obtain a new unified scale map; 6) After generating the vibration data image within the time range of time_temp-length corresponding to time_temp, the image corresponding to time_temp+1 needs to be generated, with the corresponding time range being (time_temp+1)-length. The images are generated sequentially to achieve online real-time tool status monitoring. The vibration image feature descriptor is designed as follows: Where D(i) represents the feature descriptor structure tree; i represents the angle corresponding to the feature in the structure tree, i∈[0,360); flag represents a temporary statistical marker, each i initially corresponds to flag=0, and if the if condition is met, the corresponding flag=1; init represents initialization; h(X-axis, Y-axis, i) represents all pixels corresponding to the i in the structure tree; pixel gray This represents the color value of the non-background pixels in the drawn image v_image; The function of feature difference evaluation is to identify key point features in the curve of the distribution map, so as to achieve dimensionality reduction and abstract expression of the features. The feature difference evaluation function is expressed as: Where teech represents the number of teeth of the tool that needs to be identified; for indicates loop processing; Represents any numerical value; The symbol indicates the existence of a sign; f(X, Y) represents a function that satisfies the correspondence between statistical data points; the X-axis... beg This represents the initial x-coordinate values ​​in the feature space; The key data mapping method in the design is as follows: 1) Determine the angle value in the original image based on x1 in the initial (x1, y1) and create a new single-channel image pic0 with a pure white background and the same scale. Determine the initial angle value corresponding to the new image pic0 based on the initial angle selected when drawing the two-dimensional spatial feature distribution map. Draw a ray from the center of the image based on the angle value determined by x1. The angle of the ray is consistent with x1. 2) When drawing the image based on the data features, draw a circle in the pic0 image with the radius corresponding to the normalized distributed data points in the image, and the center of the circle is the center of the image; 3) Starting with the initial value of x1, find the feature data points x1 + int(360 / teech)*1, x1 + int(360 / teech)*2, x1 + int(360 / teech)*3, ..., x1 + int(360 / teech)*n respectively with a step size of int(360 / teech), and draw a ray in the image pic0 based on each corresponding point; 4) Calculate the coordinates of the intersection points of all rays on the same circle in the image pic0, denoted as p1(x,y), p2(x,y), ... p n (x,y), based on the numerical values ​​of the intersection points, the ray is truncated, and the region containing the circle in the image pic0 is divided into n equal parts S1, S2, ..., Sn; The adaptive judgment design method is as follows: for size1, size2, ..., size n Sort the areas in ascending order to find the largest value; divide each area by the largest value, size. max The corresponding ratio contains values ​​less than half of the maximum area, enabling adaptive judgment.

2. The adaptive identification and analysis method for vibration signals of CNC machining tools with arbitrary tooth counts according to claim 1, characterized in that: Based on data length If the size of the corresponding time period (time_temp) is less than the length, then no image needs to be drawn, and the state of the tool is not analyzed within the corresponding time period.

3. The adaptive identification and analysis method for vibration signals of CNC machining tools with arbitrary tooth counts according to claim 1, characterized in that: Based on statistical data, the feature data is plotted in a two-dimensional feature space. The corresponding horizontal coordinate in the feature space represents the angle range value, and the corresponding vertical coordinate represents the size of the feature data recorded in the structure tree. The intersection of the horizontal and vertical coordinates is the origin.

4. The adaptive identification and analysis method for vibration signals of CNC machining tools with arbitrary tooth count as described in claim 1, characterized in that: The effective pixel count is calculated as follows: In the image v_image, for any feature region Si, if the vibration value at a certain moment falls within the corresponding Si region, the number of valid pixels is incremented by one; if there are vibration data points at other moments that also fall within the corresponding position of the Si region, the value of the valid data point is not incremented by one; until all points in the Si region are counted, the sum of the position data that satisfy the relationship at all corresponding points is the number of valid pixels in the corresponding region Si, which is also the feature representation of the corresponding region.

5. The adaptive identification and analysis method for vibration signals of CNC machining tools with arbitrary number of teeth according to claim 1, characterized in that: After noise removal, the sample image is binarized. The background of the processed image is black, and the grayscale corresponding to the regions with vibration data characteristics is white.

6. The adaptive identification and analysis method for vibration signals of CNC machining tools with arbitrary number of teeth according to claim 1, characterized in that: The recognition result 1 and recognition result 2 are fused to obtain the fused result. If the recognition network is not activated, the corresponding output of result 2 is 0. The fused result is based on result 1. If the identification network has been activated, it indicates that the tool has potential risks and abnormalities. The identification result 2 shall prevail.

7. The adaptive identification and analysis method for vibration signals of CNC machining tools with arbitrary number of teeth according to claim 1, characterized in that: The value of the number of teeth of the corresponding tool in the CNC machining program is read and compared with the fusion result, and a signal output result is obtained. If the corresponding values ​​are consistent, it means that there is no abnormality in the tool and machining can continue; if the corresponding values ​​are inconsistent, it means that the corresponding tool has an abnormality and a stop machining signal should be output.

8. The adaptive identification and analysis method for vibration signals of CNC machining tools with arbitrary number of teeth according to claim 1, characterized in that: The initial value of the horizontal axis corresponds to an angle that is any angle value within the entire angle range, and the corresponding number is either a positive integer or a decimal.

Images