An on-line monitoring system for milling tool wear

Abstract

The application discloses a kind of milling cutter wear online monitoring systems, for monitoring the wear state of tool in milling process, to determine whether tool reaches service life, further decide whether to replace tool.The system includes machine vision module, signal processing module and identification and prediction module, wherein machine vision module obtains relief wear amount using machine vision method and as visual label;Signal processing module obtains acoustic emission signal characteristics using acoustic emission method and constructs AE feature vector;Identification and prediction module combines the visual label with AE feature vector, constructs ABC-BP-Elman-SVM / SVR combination model through algorithm, realizes tool wear state identification and wear amount and life prediction.

Description

Technical Field

[0001] This invention relates to the field of machining condition monitoring, and more specifically, to an online monitoring system for milling cutter wear. Background Technology

[0002] Tool wear is a common phenomenon in machining processes. If a tool is severely worn or broken and is still used for machining, the surface quality of the workpiece will be reduced, and the machined parts will not meet industry standards. Therefore, tool monitoring is essential. Tool condition monitoring technology can accurately and efficiently monitor tool wear during machining, which is of great value and significance for low-cost production, improving tool utilization, and increasing production efficiency.

[0003] Tool condition monitoring technology monitors tool wear using two methods: direct and indirect methods. The direct method, machine vision, directly captures images of the tool and processes them to obtain the wear level. This method is highly accurate and less affected by environmental factors, making it the most popular choice. The indirect method, using acoustic emission signals, is more sensitive to wear conditions than other monitoring signals, has simpler signal processing, and avoids significant mechanical noise interference. Therefore, using acoustic emission signals to monitor tool wear is highly feasible.

[0004] AI, combining machine learning and statistical knowledge, constructs probabilistic models that can accurately identify tool wear conditions and predict tool life. Therefore, applying AI to tool condition monitoring systems offers unique advantages. Direct methods can only obtain tool wear during machining intervals, making online monitoring difficult. Furthermore, direct methods require a large sample size; combining them with AI necessitates multiple machine stops for sample collection, increasing machine downtime and failing to meet actual production requirements. Indirect methods, while capable of online monitoring, require repeated tool disassembly and reassembly to label each signal with wear levels during model building, consuming significant time and manpower. Therefore, neither direct nor indirect methods alone can establish an efficient tool condition monitoring system. Summary of the Invention

[0005] The purpose of this invention is to achieve efficient monitoring of milling tool wear conditions and improve the efficiency and accuracy of tool wear condition monitoring. To achieve the above objective, this invention provides an online milling cutter wear monitoring system, which comprises a machine vision module, a signal processing module, and a recognition and prediction module. The machine vision module uses machine vision to obtain the flank wear amount and uses it as a visual label; the signal processing module uses acoustic emission method to obtain acoustic emission signal features and constructs an AE feature vector; the recognition and prediction module combines the visual label with the AE feature vector and constructs an ABC-BP-Elman-SVM / SVR combined model through an algorithm to achieve online monitoring of tool wear conditions. The machine vision module includes the entire process of acquiring tool images using an automatic image acquisition method based on structural similarity, image preprocessing, threshold segmentation of the preprocessed image using an adaptive threshold segmentation method based on feature points, and extraction of wear contours and wear amounts. The signal processing module includes the acquisition and processing of acoustic emission signals. The recognition and prediction module includes the process of model building and tool wear monitoring and prediction.

[0006] The present invention discloses an online milling cutter wear monitoring system, which employs an automatic image acquisition method based on structural similarity. The method is characterized by: selecting a tool image (std) with the flank face directly facing the CCD camera as the standard image; acquiring tool images (x) to be detected at different angles; comparing the brightness, contrast, and structural similarity of the image x with the standard image (std); calculating the image quality (SSIM) at different rotation angles (γ) and the image at γ = 0° based on the brightness, contrast, and structural similarity; determining a threshold for the image acquisition interval angle based on the SSIM value; and calculating the minimum interval angle based on the threshold interval angle. Based on minimum interval angle The optimal spindle speed is set based on the interval angle coefficient and actual working conditions, and images are acquired based on this speed.

[0007] The present invention discloses an online milling cutter wear monitoring system, which employs an adaptive threshold segmentation method based on feature points. The method is characterized by: acquiring feature points during image cropping, drawing a vertical line through the wear region and background region from the feature points, calculating the grayscale difference between adjacent pixels on the vertical line, marking any point where the difference exceeds the upper limit of the difference value as a grayscale abrupt change, counting the number of pixel pairs at all grayscale abrupt changes, calculating a threshold based on the grayscale difference and grayscale value of the pixels at the grayscale abrupt changes, and performing threshold segmentation on the preprocessed image based on the grayscale value of each pixel and the threshold.

[0008] The present invention discloses an online milling cutter wear monitoring system, which constructs an ABC-BP-Elman-SVM / SVR combined model. Its features include: combining the visual labels with AE feature vectors to form a sample dataset and dividing it; establishing three tool wear monitoring sub-models: a BP neural network model, an Elman neural network model, and an SVM / SVR model; and allocating the weights of the three tool wear monitoring sub-models using an artificial bee colony algorithm, thereby establishing the ABC-BP-Elman-SVM / SVR combined model.

[0009] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects;

[0010] 1. This invention proposes an automatic image acquisition method based on structural similarity, which can acquire tool wear images that meet the standards, while ensuring the acquisition of back face images of all cutting tools and controlling the image quality within an acceptable range. This improves the automation and robustness of the tool monitoring system, reduces downtime, and increases machining efficiency and tool utilization.

[0011] 2. The adaptive threshold segmentation method based on feature points proposed in this invention only requires the calculation of 161 pixels to determine the threshold applicable to different images, which improves the generalization of threshold segmentation while taking into account the computation time cost.

[0012] 3. The ABC-BP-Elman-SVM / SVR combined model proposed in this invention continuously updates the weights of the sub-models and combines them through an optimization algorithm, thereby updating the identification and prediction values ​​of the combined model. The combined model provided by this invention has good classification and comprehensive prediction performance, and can accurately identify tool wear and predict tool wear and life. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a schematic diagram of the system structure in the embodiment;

[0015] Figure 2 This is a specific flowchart in the embodiment;

[0016] Figure 3 This is a graph showing the relationship between the tool rotation angle and the SSIM value in the embodiment;

[0017] Figure 4This is a diagram showing the image acquisition angles and similarities between images from different angles in the embodiment.

[0018] Figure 5 This is a grayscale distribution diagram of the vertical lines of feature points in the embodiment;

[0019] Figure 6 This is the generalization performance evaluation table for the ABC-BP-Elman-SVM model in the examples;

[0020] Figure 7 This is the generalization performance evaluation table of the ABC-BP-Elman-SVR wear prediction model in the examples;

[0021] Figure 8 This is the generalization performance evaluation table of the ABC-BP-Elman-SVR lifetime prediction model in the embodiments;

[0022] Figure 9 This is a graph showing the predicted tool wear and life of the ABC-BP-Elman-SVR model in the embodiment; Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] The purpose of this invention is to provide an online monitoring system for milling cutter wear.

[0025] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] Figure 1 This is a schematic diagram of the system structure according to an embodiment of the present invention, including a machine vision module 100, a signal processing module 200, and a recognition and prediction module 300.

[0027] An embodiment of the online milling cutter wear monitoring system of the present invention includes the following steps:

[0028] Step S1: The machine vision module uses machine vision to obtain the amount of wear on the back face and uses it as a visual label.

[0029] Step S2: The signal processing module uses the acoustic emission method to obtain the acoustic emission signal features and constructs the AE feature vector;

[0030] Step S3, the identification and prediction module, combines the visual labels with the AE feature vectors and constructs an ABC-BP-Elman-SVM / SVR combined model through an algorithm, thereby realizing the identification of tool wear status and the prediction of wear amount and life;

[0031] The following is a detailed explanation of each step:

[0032] Optionally, step S1 specifically includes:

[0033] Step S11: An automatic image acquisition method based on structural similarity is used to acquire tool images. The tool image with the back face facing the CCD camera is selected as the standard image std. Tool images to be detected x are acquired at different angles. The brightness, contrast, and structural similarity of the image to be detected x and the standard image std are compared. The brightness similarity l(std,x) between the standard image std and the image to be detected x is calculated from the gray mean of the standard image std and the image to be detected x. The contrast similarity c(std,x) between the standard image std and the image to be detected x is calculated from the standard deviation of the standard image std and the image to be detected x. The structural similarity s(std,x) between the standard image std and the image to be detected x is calculated from the covariance of the standard image std and the standard deviation of the image to be detected x, i.e., calculated according to the following formula:

[0034]

[0035]

[0036]

[0037] In the formula, μ std With μ x The grayscale mean values ​​of the standard image std and the image to be detected x, respectively, are σ. std With σ x σ represents the standard deviation of the standard image std and the image to be detected x, respectively. std-x Let l(std) be the covariance between the standard image std and the image to be detected x. C1 is a very small positive number, which is used to prevent errors in the calculation of l(std,x) and other indicators when the brightness or other parameters are too small. C2, C3 and C1 have the same value and function.

[0038] Calculate the image quality acquired at different rotation angles γ and the SSIM value of the image when γ = 0°, from... Figure 3The threshold [-8°, 8°] for the image acquisition interval angle can be determined. The SSIM value is a simplified calculation of luminance similarity l(std,x), contrast similarity c(std,x), and structural similarity s(std,x), calculated according to the following formula:

[0039] SSIM(std,x) = [l(std,x)] a ×[c(std,x)] b ×[s(std,x)] c

[0040] In the formula, l(std,x) is the brightness similarity, c(std,x) is the contrast similarity, s(std,x) is the structural similarity, and a, b, and c are the positive exponential parameters of each index.

[0041] Based on the aforementioned interval angle threshold [-8°, 8°], and taking an image acquisition time of t = 4s and the number of images acquired z = 28, the minimum interval angle for image acquisition can be obtained. Based on minimum interval angle The optimal spindle speed n = 130.18 rpm was set based on the interval angle coefficient and actual working conditions. Images were acquired based on this speed, and the three images with the highest SSIM values ​​were selected as the original images. The interval angle coefficient k should satisfy the following constraints:

[0042] k belongs to the set K = {k|0} <k<28,k∈Z};

[0043] In image acquisition, if the current image acquisition angle is the same as the previously acquired angle, the k value is discarded;

[0044] After acquiring the first image from the original position, at an angle As the basic angular unit, after 27 increments, it becomes After the angle is changed, the angle of image acquisition should include the following: Figure 4 The angles of all the arrows shown;

[0045] Step S12: Crop the original image to obtain the region of interest image of the blade tip, and perform noise reduction and enhancement to obtain the preprocessed image;

[0046] Step S13: An adaptive thresholding method based on feature points is used to perform thresholding segmentation on the preprocessed image, such as... Figure 5 As shown, feature points are obtained during the image cropping process, and a vertical line is drawn through the wear area and the background area. The grayscale difference between adjacent pixels on the vertical line is calculated using the following formula:

[0047] g(x,x+1)=|f(x+1)-f(x)|

[0048] In the formula, g(x,x+1) is the gray level difference between adjacent pixels, and f(x+1) and f(x) are the gray level values ​​of adjacent pixels.

[0049] When the difference between adjacent pixels on the vertical line exceeds the upper limit of the difference value, it is marked as a gray-level abrupt change. The number of pixel pairs at all such gray-level abrupt changes is counted. A threshold is calculated based on the gray-level difference and gray-level value of the pixels at the gray-level abrupt changes. Threshold segmentation is then performed on the preprocessed image based on the gray-level value of each pixel and the threshold. The threshold is calculated using the following formula:

[0050]

[0051] In the formula, T is the threshold, k is the number of pixel pairs at the gray-level abrupt change, and g(x) k ,x k +1) represents the grayscale difference of the pixels at the grayscale abrupt change, f(x) k ) and f(x k +1) represents the gray value of the pixel at the gray-scale abrupt change.

[0052] The segmented wear image is optimized for wear regions and edges are extracted to obtain the tool wear profile.

[0053] Step S14: Perform image angle correction on the tool wear contour, determine the calculated edge of the corrected image and divide the region, perform pixel equivalent calibration on the divided wear region and extract the tool wear amount by scanning the number of pixels, and use the tool wear amount as a visual label.

[0054] Step S2, specifically includes the following steps:

[0055] Step S21: Acquire acoustic emission signals under the same milling conditions as the machine vision method described above;

[0056] Step S22: Preprocess the acoustic emission signal to remove null values, outliers and baseline offsets in the acoustic emission signal to obtain the preprocessed acoustic emission signal.

[0057] Step S23: Extract feature values ​​from the preprocessed acoustic emission signal, extracting time-domain, frequency-domain, and time-frequency-domain features of the acoustic emission signal respectively. The time-domain features include: mean, root mean square, maximum amplitude, peak-to-peak value, standard deviation, root sign square, kurtosis factor, impulse factor, peak factor, and margin factor. The frequency-domain features include: frequency centroid and frequency variance. The time-frequency-domain features include the energy of 16 frequency bands extracted by wavelet packet transform and the effective values ​​of 11 components extracted by empirical mode decomposition, for a total of 27 time-frequency-domain features extracted.

[0058] Step S24: Select and fuse the time domain, frequency domain, and time-frequency domain features, filter out features with weak correlation and redundancy, obtain the remaining 30 features, fuse the remaining 30 features, and construct the AE feature vector from the fused 24 feature values.

[0059] Step S3 specifically includes the following steps:

[0060] Step S31: Combine the visual labels with the AE feature vectors to form a sample dataset and divide it. Establish three tool wear monitoring sub-models: a BP neural network model, an Elman neural network model, and an SVM / SVR model. Assign weights to the three sub-models using an artificial bee colony algorithm, thus establishing an ABC-BP-Elman-SVM / SVR combined model. Verify and test the accuracy of the combined model. The accuracy requirements for the ABC-BP-Elman-SVM model include: recall, precision, and harmonic mean. Figure 6 As shown. The required indicators for the prediction accuracy of the ABC-BP-Elman-SVR model include: Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and Spearman Rank Correlation Coefficient (SROCC). The generalization performance evaluation of the ABC-BP-Elman-SVR wear prediction model is as follows: Figure 7 As shown, the generalization performance evaluation of the ABC-BP-Elman-SVR lifetime prediction model is as follows: Figure 8 As shown;

[0061] Step S32: The tool wear condition is identified, tool wear amount is predicted, and tool life is predicted using the ABC-BP-Elman-SVM / SVR combined model. The tool wear amount and life prediction of the ABC-BP-Elman-SVR model are as follows: Figure 9 As shown.

[0062] The embodiments described in this specification are merely examples of implementations of the inventive concept. The scope of protection of this invention should not be considered as limited to the specific forms described in the embodiments. The scope of protection of this invention also extends to equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. An online monitoring system for milling cutter wear, characterized in that: The online end mill wear monitoring system comprises a machine vision module, a signal processing module, and a recognition and prediction module. The machine vision module uses machine vision to obtain the flank wear amount and uses it as a visual label. The signal processing module uses acoustic emission (AE) to obtain acoustic emission signal features and constructs an AE feature vector. The recognition and prediction module combines the visual label with the AE feature vector to form a sample dataset, divides it, and establishes three tool wear monitoring sub-models: a BP neural network model, an Elman neural network model, and an SVM / SVR model. The weights of the three sub-models are allocated using an artificial bee colony algorithm to construct an ABC-BP-Elman-SVM / SVR combined model. The machine vision module uses an automatic image acquisition method based on structural similarity for tool image acquisition and preprocessing. It then uses an adaptive threshold segmentation method based on feature points to perform threshold segmentation on the preprocessed image, as well as wear contour extraction and wear amount extraction. The signal processing module is used to implement the acoustic emission signal acquisition and processing process. The recognition and prediction module is used to build the model and implement the tool wear monitoring and prediction process. The automatic image acquisition method based on structural similarity includes the following steps: selecting a tool image with the back face facing the CCD camera as the standard image std; acquiring tool images x to be detected at different angles; comparing the brightness, contrast, and structural similarity of the image x to be detected with the standard image std; calculating the SSIM value of the images acquired at different rotation angles γ and the SSIM value of the image acquired when γ=0° based on the brightness, contrast, and structural similarity; determining the threshold of the image acquisition interval angle based on the SSIM value; and calculating the minimum interval angle based on the threshold of the interval angle. Based on minimum interval angle The optimal spindle speed is set based on the interval angle coefficient and actual working conditions, and images are acquired based on this speed.

2. The online milling cutter wear monitoring system according to claim 1, characterized in that, The adaptive thresholding method based on feature points includes the following steps: acquiring feature points during image cropping, drawing a vertical line through the wear area and background area, calculating the gray-level difference between adjacent pixels on the vertical line, marking the point where the difference between adjacent pixels on the vertical line exceeds the upper limit of the difference value, counting the number of pixel pairs at all gray-level abrupt changes, calculating the threshold based on the gray-level difference and gray-level value of the pixels at the gray-level abrupt changes, and performing thresholding segmentation on the preprocessed image based on the gray-level value of each pixel and the threshold.

Images