A tool tip wear detection method and device based on an improved L1PS model

Abstract

The application discloses a kind of based on the improved L1PS model's tool tip wear detection method and device;The method is as follows: one, tool bottom edge image is collected and binarized.Two, the initial bottom edge image and binary image are respectively carried out image blocking, obtain the single target original image and single target binary image corresponding to each blade tooth of bottom edge;The edge image of each single target image G '(x,y) is calculated respectively.Three, with each edge image as mask, for single target original image, edge extraction is carried out respectively, and wear edge image is obtained;Four, the blade tooth wear area is calculated.The application real-time acquisition tool bottom edge image, and the high-precision edge extraction of L1SP is carried out to the image collected, obtains the edge image before and after tool bottom edge wear, and then obtains wear area, improves tool bottom edge edge detection precision, reduces the time for processing tool bottom edge image.

Description

Technical Field

[0001] This invention belongs to the field of intelligent manufacturing technology, specifically relating to a tool tip wear detection method and device based on an improved L1PS model. Background Technology

[0002] With the continuous development of my country's aerospace industry, the demand for difficult-to-machine materials, complex structural parts, and thin-walled parts is constantly increasing, and the performance requirements are becoming more stringent. These parts are typically machined using milling operations. In particular, thin-walled aerospace aluminum alloy parts are machined using high-speed, low-feed, and low-depth-of-cut technology, requiring all relevant dimensions to be machined in a single clamping operation, while demanding high precision dimensional and positional tolerances.

[0003] Currently, most methods for detecting milling cutter wear in industrial production rely on empirical observation and offline measurement. Manual inspection is not only costly but also highly subjective and lacks the ability to quantify the degree of wear. In offline measurement, the cutter and tool holder need to be repeatedly clamped and positioned, which significantly reduces efficiency and introduces additional assembly errors and variations in wear mechanisms. Therefore, to meet the demands of high-precision manufacturing processes, it is necessary to develop online methods for detecting milling cutter wear and improve detection accuracy.

[0004] Currently, there are several mature methods for improving the accuracy of end mill bottom wear measurement. The paper "Monitoring of Milling Tool Wear Condition Based on Two-Layer Angle Kernel Limit Learning Machine using Acoustic Sensors" introduces a Tool Condition Monitoring (TCM) method, which combines the optimal characteristic parameters of acoustic sensor signals with a kernel limit learning machine. This method can be applied in practical applications; however, signal analysis cannot provide direct and accurate information about tool wear values, and the signal is easily affected by the industrial environment. The paper "A Method for Automatically and Accurately Detecting Tool Wear Using a Gray-Scale Image Probabilistic Algorithm Based on Bayesian Inference" establishes a probability function based on the correlation between gray-scale changes along the cutting edge and the relative positions of the original boundary and the wear boundary, and reconstructs the curved original tool boundary through Bayesian inference. This method has the advantages of being non-contact and numerically accurate, but suffers from three major problems: difficulty in image acquisition, inability to detect online, and algorithmic complexity.

[0005] Currently, there are relatively few devices and methods for real-time high-precision detection of the wear area of ​​the end mill tip using machine vision algorithms. In order to improve the accuracy of end mill edge detection and meet the needs of high-precision manufacturing processes, this paper proposes a method and application for high-precision wear measurement of the end mill tip based on an improved L1PS model. Summary of the Invention

[0006] In order to solve the above-mentioned technical problems, improve the accuracy of milling cutter edge detection, and meet the needs of high-precision manufacturing, this invention proposes a method and apparatus for detecting the wear area of ​​the tool tip by designing a CNC program and controlling the tool installation, acquiring images of the bottom edge of the tool in real time, and using an improved image processing algorithm.

[0007] In a first aspect, the present invention proposes a tool tip wear detection method based on an improved L1PS model, comprising the following steps:

[0008] Step 1: Acquire the bottom edge image of the tool as the initial bottom edge image F(x,y). Segment the initial bottom edge image F(x,y) to obtain the binarized image G(x,y) as follows:

[0009]

[0010] Where L is the maximum number of levels in the grayscale image. T is the optimal threshold; the optimal threshold T is the segmentation threshold that maximizes the inter-class variance E'(t).

[0011] Step 2: Divide the initial bottom edge image F(x,y) and the binarized image G(x,y) into blocks to obtain the single target original image F'(x,y) and single target binary image G'(x,y) corresponding to each tooth of the bottom edge; calculate the edge image P(x,y) of each single target image G'(x,y).

[0012] Step 3: Using each edge image P(x,y) as a mask, perform edge extraction on the original single-target image F'(x,y) to obtain the wear edge image S(x,y); the process of obtaining the wear edge image S(x,y) is as follows:

[0013]

[0014] Where j1 and j2 are the average gray values ​​of the foreground and background regions of the original single-target image F'(x,y), respectively; B(x,y) is the foreground region image; C1 is the convex set; λ, β, and v are positive parameters; and Ω is the bounded image domain. The optimized P(x,y) is used as the wear edge image S(x,y).

[0015] Step 4: Perform an XOR operation between the wear edge image S(x,y) and the edge image P(x,y) to obtain the wear area S of the cutting edge.

[0016] Preferably, for any segmentation threshold t, the expression for the corresponding inter-class variance E'(t) is:

[0017]

[0018] in, The weights are: ω0 and ω1 are the cumulative probabilities of classes D0 and D1, respectively; μ0 and μ1 are the average gray values ​​of classes D0 and D1, respectively; μ is the weight of the initial bottom edge image F(x,y). The threshold t divides all pixels of the initial bottom edge image F(x,y) into class D0 (gray values ​​less than or equal to threshold t) and class D1 (gray values ​​greater than threshold t); μ0 and μ1 are the average gray values ​​of classes D0 and D1, respectively; μ1 is the weight of the bottom edge image F(x,y). t Let F(x,y) be the global average gray value of the initial bottom edge image.

[0019] As a preferred option, in step one, the weights The expression is:

[0020]

[0021] Where p(t+i) is the probability of grayscale value t+i appearing.

[0022] Preferably, in step one, the expressions for the cumulative probabilities ω0 and ω1 are:

[0023]

[0024] Where p(k) is the probability of gray value k appearing.

[0025] Preferably, the expression for the probability p(k) of the grayscale value k appearing is:

[0026]

[0027] Where M is the number of rows in the image matrix; N is the number of columns in the image matrix.

[0028] As a preferred embodiment, the method for obtaining the single-target image G'(x,y) in step two is as follows:

[0029] The Hough transform is used to detect the longest line in a binarized image G(x,y). The centroid of the smallest bounding rectangle of the foreground region of the binarized image G(x,y) is taken as the rotation point (a,b). The longest line is rotated n-1 times around the rotation point (a,b). Each rotation... Degree; where n is the number of teeth on the bottom edge. A zero matrix Zero(max(a1,a2),max(b1,b2)) is constructed using the maximum value among the coordinates of the two vertices (a1,b1) and (a2,b2) of the longest straight line. The image of the largest connected component is preserved, resulting in a single-target image G'(x,y).

[0030] Preferably, in step four, after obtaining the wear area S, the wear area S is compared with a preset wear limit threshold; if the wear area S is greater than the wear limit threshold, the tool is replaced.

[0031] Secondly, this invention proposes a tool tip wear detection device based on an improved L1PS model to implement the aforementioned tool tip wear detection method based on the improved L1PS model. The tool tip wear detection device includes a control terminal and a data acquisition module. The data acquisition module is installed on the control system module and is used to acquire images of the bottom edge of the tool and transmit them to the control terminal. The control terminal processes the bottom edge images of the tool to obtain the wear area of ​​the tool.

[0032] Preferably, the data acquisition module includes a fixed stage and a camera and a microscope mounted on the fixed stage; the microscope is mounted on the lens of the camera.

[0033] Preferably, the microscope is equipped with a ring light source and a CNC linear cold light source on the outer side of the tube to improve the problem of uneven light intensity in the bottom edge image of the cutting tool.

[0034] The beneficial effects of this invention are:

[0035] 1. This invention acquires real-time images of the bottom edge of a cutting tool and performs multi-target segmentation and high-precision edge extraction using L1SP on the acquired images to obtain edge images of the bottom edge of the cutting tool before and after wear, thereby obtaining the wear area and improving the accuracy of edge detection of the bottom edge of the cutting tool.

[0036] 2. This invention generates multiple single-target images corresponding to the blades from the bottom edge images of multiple targets and extracts them separately. Compared with the simple L1SP algorithm that directly processes the original bottom edge image, this invention significantly reduces the required processing time. Under the same conditions, the actual processing time of the simple L1SP algorithm is about 21 minutes, while this invention only requires 3.2 seconds.

[0037] 3. This invention does not require stopping the machine when obtaining the wear area of ​​the bottom edge of the tool, and can automatically and efficiently detect the wear amount of the bottom edge of the tool, reducing the time cost of detecting the wear area of ​​the bottom edge of the tool. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of the blade tip wear detection device used in this invention.

[0039] Figure 2 This is a flowchart of the present invention.

[0040] Figure 3 This is the binarized image obtained in step two of this invention.

[0041] Figure 4 This is a schematic diagram of the single-target original image and single-target binary image processing process after segmentation in Embodiment 1 of the present invention.

[0042] Figure 5 This is a schematic diagram of the single-target binary image extracted in Comparative Example 1 of the present invention. Detailed Implementation

[0043] The present invention will be further described below with reference to the accompanying drawings.

[0044] like Figure 1 As shown, a tool tip wear detection device based on an improved L1PS model includes a control terminal, a control system module, and a data acquisition module. The control system module is used for milling operations and includes a machine tool 1, a worktable 2, a cutting tool 6, and a workpiece 9 mounted on the machine tool 1; the worktable 2 is used to move the data acquisition module in the horizontal direction; the cutting tool 6 is mounted above the worktable 2 for machining the workpiece 9. The data acquisition module includes a fixed stage 3 and a microscope 5, an industrial camera 4, a ring light source 7, and a CNC linear cold light source 8 mounted on the fixed stage 3. The fixed stage 3 is fixed to the worktable 2 by bolts. The connection between the microscope 5 and the fixed stage 3 is made of shock-absorbing resin to reduce image blurring caused by machine tool vibration. The microscope 5 is used to magnify the bottom edge image of the cutting tool 6 for better observation and detection of the wear area of ​​the bottom edge. The industrial camera 4 is mounted directly below the eyepiece of the microscope 5 to capture the magnified image of the bottom edge of the cutting tool 6. The ring light source 7 is mounted on the objective lens side of the microscope 5 to reduce the influence of ambient light. The CNC linear cold light source 8 is mounted on the side of the microscope 5 to supplement the light intensity of the bottom edge image of the cutting tool 6 through the inside of the microscope 5. The combination of the ring light source 7 and the CNC linear cold light source 8 adjusts and improves the problem of uneven light intensity in the bottom edge image of the cutting tool 6. The control terminal is used to adjust the devices in the control system module and the data acquisition module to acquire the bottom edge image of the cutting tool 6 and process the acquired image to obtain the wear area of ​​the cutting tool 6.

[0045] like Figure 2 As shown, the working principle of the tool tip wear detection device based on the improved L1PS model is as follows:

[0046] Step 1: Parameter preset.

[0047] Manually move the microscope 5 directly below the tool 6, and use the industrial camera 4 to pre-capture the bottom edge image of the tool 6 after magnification by the microscope 5. Feed back the pre-captured image to the control terminal. The control terminal adjusts the height and focal length of the microscope 5, as well as the illumination intensity of the combination of the ring light source 7 and the CNC linear cold light source 8, according to the size and average gray value of the pre-captured image.

[0048] Step 2: CNC programming.

[0049] The control unit uses a system module to reset the spindle speed and move the tool position, and sets a fixed machining time rule. During image acquisition, the spindle stops rotating, the feed rate is 0, a 5-second pause is specified, and the microscope 5 is moved directly below the tool 6. The industrial camera 4 acquires an image of the bottom edge of the tool 6, magnified by the microscope 5. After image acquisition, the spindle resumes rotation, the workpiece machining module returns to its original position, and the remaining machining program continues. The industrial camera 4 transmits the acquired image to the control unit for processing.

[0050] Step 3: Multi-target segmentation.

[0051] The image undergoes initial segmentation. This embodiment achieves better segmentation results when the target and the image background have significantly different grayscale levels. The principle behind image segmentation in this step is as follows:

[0052] The grayscale value range of the image is 0 to L-1; where L is the maximum level in the grayscale image, with a value of 2. 8 The expression for the probability p(k) of grayscale value k appearing is:

[0053]

[0054] Where F(x,y) is the initial bottom edge image; M is the number of rows in the image matrix; and N is the number of columns in the image matrix.

[0055] The image is divided into two classes, D0 and D1, using a floating threshold t. Class D0 consists of pixels with gray values ​​in the range [0, t], and class D1 consists of pixels with gray values ​​in the range [t+1, L-1]. The cumulative probabilities ω0 and ω1 of classes D0 and D1 are expressed as follows:

[0056] ω0=∑ 0≤k≤t p(k)

[0057] ω1=∑ t<k≤L-1 p(k)

[0058] ω0+ω1=1

[0059] The expressions for the average gray values ​​μ0 and μ1 of class D0 and class D1 are as follows:

[0060] μ0=∑ 0≤k≤t kp(k) / ω0

[0061] μ1=∑ t<k≤L-1 kp(k) / ω1

[0062] global average value μ of the image t The expression is:

[0063] μ t =ω0μ0+ω1μ1

[0064] The expression for the inter-class variance E between classes D0 and D1 is:

[0065] E=ω0(μ0-μ t ) 2 +ω1(μ1-μ t ) 2

[0066] Assign a weight The expression for obtaining the updated inter-class variance E'(t) is:

[0067]

[0068] in, For the weights, the expression is: p(t+i) is the probability of grayscale value t+i appearing.

[0069] Based on the inter-class variance E' corresponding to different thresholds t, the optimal threshold T is obtained as follows:

[0070] T = Argmax 0≤t≤L-1 (E'(t))

[0071] Where Argmax(·) is a function for finding the independent variable corresponding to the maximum dependent variable;

[0072] Binarize the image based on the optimal threshold.

[0073]

[0074] Where G(x,y) is the binarized image after segmentation, and the result is as follows: Figure 3 As shown.

[0075] Step 4: Convert multiple objectives into a single objective.

[0076] After initial segmentation, the binarized image G(x,y) is divided into blocks, transforming the multi-objective task into a single-objective task. The Hough transform is used to detect the longest straight line in the binarized image G(x,y), which represents the fitted edge of a blade. The centroid of the smallest bounding rectangle of the foreground region of the binarized image G(x,y) is taken as the rotation point (a,b), and the longest straight line is rotated n-1 times around the rotation point (a,b). Each rotation... The degree can be used as the basis for image segmentation; where n is the number of bottom edges of the tool 6. For example, a two-edge tool will rotate the detected straight line 180 / 2 = 90 degrees once, a three-edge tool will rotate the detected straight line 60 degrees twice, and a four-edge tool will rotate the detected straight line 45 degrees three times. Taking a two-bladed cutting tool as an example, the image is segmented by creating a new zero matrix Zero(max(a1,a2),max(b1,b2)) based on the maximum value of the coordinates of the two vertices (a1,b1) and (a2,b2) of the straight line. Elements in the binarized image G(x,y) that are smaller than the vertex coordinates are replaced into the zero matrix, and the area of ​​the connected components is calculated. Only the image of the largest connected component is retained to obtain the single-target binary image G'(x,y), avoiding the accidental removal of other bottom edge images by the single-target binary image G'(x,y) during the straight line rotation. The edge image P(x,y) of the single-target binary image G'(x,y) is calculated. The above process is repeated to obtain the single-target binary image G'(x,y) of the other blade, and the edge image P(x,y) of the single-target binary image G'(x,y) is calculated.

[0077] Using the same method, the initial bottom edge image F(x,y) is divided into image blocks to obtain the single-target original image F'(x,y) corresponding to each tooth of the bottom edge.

[0078] Step 5: High-precision edge extraction.

[0079] like Figure 4 As shown, in complex tool structures, the global thresholding method suffers from poor segmentation accuracy and segmentation errors due to tool wear. The L1PS algorithm, by additionally considering the offset field that causes uneven grayscale, demonstrates excellent edge segmentation performance. Using the edge image P(x,y) obtained in step three as a mask, L1PS is used to perform detailed segmentation of a single edge in the initial bottom edge image F(x,y), while simultaneously performing parallel computation based on the number of cutting edges. Irregular boundaries and shapes in the image are processed using the level set method to obtain an accurate wear edge image S(x,y). L1PS utilizes L... 1 Manhattan norm (MAN) is used to effectively handle outliers and noise in image decomposition theory. It effectively solves the inhomogeneity of images and demonstrates robustness to noise and outliers, suitable for smooth transitions between worn and unworn areas of the cutting edge.

[0080] Using each edge image P(x,y) as a mask, edge extraction is performed on the original image F'(x,y) of a single target to obtain the wear edge image S(x,y); the process of obtaining the wear edge image S(x,y) is as follows:

[0081] Using Lipschitz boundaries and setting For a bounded image domain, the image domain Ω of the original single-target image F'(x,y) is divided into a foreground region Ω1 and a background region Ω2, where Ω1∩Ω2=φ and Ω1∪Ω2=Ω. L1PS uses a piecewise smoothing (PS) function to approximate the denoised image. To simulate an image with non-uniform intensity, the original single-target image F'(x,y) is decomposed into:

[0082] F'(x,y)=B(x,y)+J i (x,y)+n(x,y)

[0083] Where B(x,y) is the image with uneven gray levels; J i (x,y) is the real image, i = 1, 2, representing the foreground and background regions respectively; n(x,y) is the noisy image.

[0084] Using L 1 The data fidelity term measures the difference (L1PS) between the original single-object image F'(x,y) and the denoised image B(x,y)+J(x,y), i.e., L 1 Measure the noisy image n(x,y) to obtain a model for the worn edge image S(x,y):

[0085]

[0086] Where j1 and j2 are the average intensity values, i.e., gray values, of regions Ω1 and Ω2, respectively; λ, β, and v are positive parameters. For L 2 The norm measures the degree of gray-level unevenness. Let L be the length of the edge image P(x,y). The length of the edge image P(x,y) is expressed using L. 1 Using a norm as a metric preserves the smoothness of the edges. The optimized P(x,y) is used as the worn edge image S(x,y).

[0087] To solve the above model, we define a convex set C1∈[0,1], and use convex relaxation to transform the above equation into:

[0088]

[0089] The minimization problem is not convex with respect to (j1,j2,B(x,y),P(x,y)), but it is convex with respect to each variable. Each minimization problem has an exact solution, which can be computed using convex optimization algorithms. The problem is solved by alternately solving each minimization problem with respect to j, B(x,y), and P(x,y) until convergence. While this minimization process does not guarantee convergence to the global minimum (or even a local minimum), in practice it often yields good approximations. Therefore, to determine the numerical solution (j1,j2,B(x,y),P(x,y)), an alternating minimization method is used: with P(x,y) fixed, the minimums of j1,j2, and B(x,y) are computed, and the minimum of P(x,y) is computed based on the minimums of j1,j2, and B(x,y).

[0090] Step 6: Obtain the wear area.

[0091] The number of pixels within the wear area of ​​tool 6 is determined based on the wear edge image S(x,y) and the edge image P(x,y). The tool wear area S is then calculated based on the number of pixels within the wear area. If the calculated tool wear area S is less than a preset value, tool 6 continues to operate; otherwise, tool 6 is replaced in the control system module.

[0092] L1PS offers better segmentation for blurred edges, but its computational complexity is high and it cannot handle multi-target detection. To address this issue, this invention segments the image in step two, extracts single-target images, and then uses L1PS to process these single-target images to obtain accurate wear edges. Furthermore, image segmentation allows for convenient approximation of the edge of the worn cutting edge; thus, the wear area can be obtained by performing a simple XOR operation between the wear edge image and the edge image of the worn cutting edge.

[0093] Comparative Example 1

[0094] A method for detecting blade tip wear based on an improved L1PS model. The difference between this comparative example and Example 1 lies in the different edge extraction methods for the image S(x,y).

[0095] Comparative Example 1 uses the OTSU method to extract edges from the initial bottom edge image F(x,y). The extracted single-target binary image G'(x,y) in this comparative example is as follows: Figure 5 As shown, comparison Figure 4 and Figure 5 As can be seen, the segmentation method used in Example 1 can obtain an image that is closer to the worn blade.

[0096] In this comparative example, the method for obtaining the wear edge image S(x,y) is as follows:

[0097]

[0098] Where, λ∫ Ω |F(x)-s(x)| 2 dx is the data fitting term; F(x,y) is the initial bottom edge image; s(x) is the object to be optimized, whose initial value is a smooth approximation of the initial bottom edge image F(x,y), and the optimization result is used as the wear edge image S(x,y); Γ represents the regularization term; Γ represents the edges of the foreground and background regions; Ω\Γ represents the remaining edge region. For gradient operators; The L2 norm is used as a metric; Length(·) is the edge length; λ and β are positive parameters. X1 and X2 represent the horizontal and vertical directions of the image matrix.

[0099] Although MS has good mathematical properties, numerical solutions are very difficult due to the non-regularity and non-convexity of its edges.

Claims

1. A method for detecting tool tip wear based on an improved L1PS model, characterized in that: Includes the following steps: Step 1: Acquire the bottom edge image of the tool as the initial bottom edge image F(x,y). Segment the initial bottom edge image F(x,y) to obtain the binarized image G(x,y) as follows: Where L is the maximum level in the grayscale image; The optimal threshold; the optimal threshold The segmentation threshold that maximizes the inter-class variance E'(t); Step 2: Divide the initial bottom edge image F(x,y) and the binarized image G(x,y) into blocks to obtain the single-target original image F'(x,y) and single-target binary image G'(x,y) corresponding to each tooth of the bottom edge; calculate the edge image P(x,y) of each single-target image G'(x,y); Step 3: Using each edge image P(x,y) as a mask, perform edge extraction on the original single-target image F'(x,y) to obtain the wear edge image S(x,y); the process of obtaining the wear edge image S(x,y) is as follows: Where j1 and j2 are the average gray values ​​of the foreground and background regions of the original single-target image F'(x,y), respectively; B(x,y) is the foreground region image; C1 is the convex set; λ, β, v are positive parameters; Ω is the bounded image domain; and the optimized result is... As the image of the worn edge, S(x,y); Step 4: Perform an XOR operation on the wear edge image S(x,y) and the edge image P(x,y) to obtain the wear area S of the cutting edge.

2. The tool tip wear detection method based on the improved L1PS model according to claim 1, characterized in that: In step one, for any segmentation threshold t, the expression for its corresponding inter-class variance E'(t) is: in, The weights are: ω0 and ω1 are the cumulative probabilities of classes D0 and D1, respectively; μ0 and μ1 are the average gray values ​​of classes D0 and D1, respectively; μ is the weight of the initial bottom edge image F(x,y). The threshold t divides all pixels of the initial bottom edge image F(x,y) into class D0 (gray values ​​less than or equal to threshold t) and class D1 (gray values ​​greater than threshold t); μ0 and μ1 are the average gray values ​​of classes D0 and D1, respectively; μ1 is the weight of the bottom edge image F(x,y). t Let F(x,y) be the global average gray value of the initial bottom edge image.

3. The tool tip wear detection method based on the improved L1PS model according to claim 2, characterized in that: In step one, weights The expression is: Where p(t+i) is the probability of grayscale value t+i appearing.

4. The tool tip wear detection method based on the improved L1PS model according to claim 2, characterized in that: In step one, the expressions for the cumulative probabilities ω0 and ω1 are: Where p(k) is the probability of gray value k appearing.

5. The tool tip wear detection method based on the improved L1PS model according to claim 4, characterized in that: The expression for the probability p(k) of the grayscale value k appearing is: Where M is the number of rows in the image matrix; N is the number of columns in the image matrix.

6. The tool tip wear detection method based on the improved L1PS model according to claim 1, characterized in that: In step two, the method for obtaining the single-target image G'(x,y) is as follows: The Hough transform is used to detect the longest line in a binarized image G(x,y). The centroid of the smallest bounding rectangle of the foreground region of the binarized image G(x,y) is taken as the rotation point (a,b). The longest line is rotated n-1 times around the rotation point (a,b). Each rotation... Degree; where n is the number of teeth on the bottom edge; Zero matrix Zero(max(a1,a2),max(b1,b2)) is established using the maximum value of the coordinates of the two vertices (a1,b1) and (a2,b2) of the longest straight line, and the image of the largest connected component is retained to obtain the single target image G'(x,y).

7. The tool tip wear detection method based on the improved L1PS model according to claim 1, characterized in that: In step four, after obtaining the wear area S, the wear area S is compared with the preset wear limit threshold. If the wear area S is greater than the wear limit threshold, the tool is replaced.

8. A tool tip wear detection device based on an improved L1PS model, characterized in that: The tool tip wear detection device is used to perform a tool tip wear detection method based on an improved L1PS model as described in claim 1. The tool tip wear detection device includes a control terminal and a data acquisition module. The data acquisition module is installed on the control system module and is used to acquire the bottom edge image of the tool (6) and transmit it to the control terminal. The control terminal is used to process the bottom edge image of the tool (6) to obtain the wear area of ​​the tool (6).

9. The tool tip wear detection device based on the improved L1PS model according to claim 8, characterized in that: The data acquisition module includes a fixed stage (3) and a camera and microscope (5) mounted on the fixed stage (3); the microscope (5) is mounted on the lens of the camera.

10. A tool tip wear detection device based on an improved L1PS model according to claim 9, characterized in that: The microscope (5) is equipped with a ring light source (7) and a CNC line cold light source (8) on the outside of the tube to improve the problem of uneven light intensity in the bottom edge image of the cutting tool (6).

Images