Tool wear detection method and device based on visual recognition

Abstract

The application discloses a tool wear detection method and device based on visual recognition and belongs to the technical field of tool wear detection, which comprises the following steps: S1: collecting a tool edge image.In the application, a deep learning model is used to perform pixel-level segmentation on the tool edge area, which can simultaneously identify multiple wear types such as damage, cracks and wear strips, thereby breaking through the limitation of the prior art that can only identify severe damage.Meanwhile, by calculating the ratio of the wear area to the total area of the tool edge, the percentage of wear degree is accurately quantified and output, replacing the traditional manual qualitative judgment method, and the detection accuracy and consistency are significantly improved.In addition, the optimized image segmentation algorithm such as Otsu method and Canny edge detection is used in combination with the image preprocessing step, which effectively enhances the contrast of the tool edge area and ensures that the tool contour can still be stably extracted under complex light and noise environment, thereby providing high-quality input images for subsequent wear identification.

Description

Technical Field

[0001] This invention belongs to the field of tool wear detection technology, specifically a tool wear detection method and device based on vision recognition. Background Technology

[0002] In the field of CNC machining, cutting tools, as core components directly involved in cutting, directly affect the machining accuracy, surface quality, and production efficiency of workpieces due to their wear condition. Over long-term use, cutting tools gradually develop various wear forms such as breakage, cracks, and wear bands. If these are not detected and replaced in time, they can easily lead to workpiece scrap or even machine tool damage. With the development of machine vision and image processing technologies, visual recognition-based tool wear detection methods, due to their advantages of being non-contact, fast, and highly automated, are gradually becoming an important means to replace traditional manual visual inspection, enabling real-time monitoring and evaluation of the cutting edge condition.

[0003] However, existing tool wear detection technologies still have many shortcomings. On the one hand, traditional manual visual inspection or microscopic measurement methods rely on the operator's experience and judgment, resulting in low detection efficiency, high subjectivity, poor consistency, and difficulty in integrating into automated production lines for real-time monitoring. On the other hand, most existing visual inspection methods can only identify severe tool damage and cannot simultaneously classify and identify multiple wear types such as cracks and wear bands at the pixel level. Furthermore, they lack precise quantitative assessment methods for the degree of wear and cannot output quantitative indicators such as wear percentage. In addition, existing technologies have not yet formed an end-to-end automated inspection process from image acquisition and wear recognition to remaining service life prediction, which cannot effectively avoid processing interruptions and workpiece scrapping caused by sudden tool failure, and cannot meet the needs of modern intelligent manufacturing for intelligent and efficient tool management. Summary of the Invention

[0004] The purpose of this invention is to provide a visual recognition-based method and apparatus for detecting tool wear, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution, comprising the following steps:

[0006] S1: Acquire the image of the cutting edge of the tool and preprocess the acquired image to enhance the contrast of the cutting edge area;

[0007] S2: Perform tool region segmentation on the preprocessed image, extract the tool contour and locate the cutting edge region;

[0008] S3: Use a deep learning model to perform pixel-level classification of the blade area, identify wear types including breakage, cracks, and wear bands, and calculate the geometric parameters of the wear area to obtain the wear percentage;

[0009] S4: Automatically identifies tool information and predicts the remaining tool life based on the current wear percentage, historical wear data, tool material, and machining parameters;

[0010] S5: Outputs wear percentage, wear type, tool information and remaining service life, and issues an alarm signal when the wear exceeds the set threshold;

[0011] The tool region segmentation employs either a threshold segmentation algorithm or an edge detection algorithm.

[0012] The specific steps are as follows:

[0013] The tool contour is extracted from the preprocessed image using a threshold segmentation algorithm or an edge detection algorithm, and the cutting edge region is located.

[0014] The threshold segmentation algorithm includes: calculating the image grayscale histogram, determining the optimal segmentation threshold using Otsu's method or an adaptive threshold segmentation method, binarizing the image, and separating the tool area from the background area.

[0015] The edge detection algorithm includes: using the Canny edge detection operator or the Sobel operator to calculate the image gradient magnitude, detecting tool edge points, and connecting the broken edges through morphological closing operations to form a continuous tool profile;

[0016] After extracting the tool profile, the cutting edge region is located based on the prior information of the tool geometry, and the region of interest containing only the cutting edge is cropped for subsequent wear feature extraction.

[0017] As a further preferred embodiment of this technical solution: the image preprocessing includes grayscale conversion, filtering and noise reduction, histogram equalization, and image enhancement processing;

[0018] As a further preferred embodiment of this technical solution: the deep learning model is a U-Net or ResNet network structure, which performs pixel-level classification on the blade area, identifies areas such as damage, cracks, and wear bands, and calculates geometric parameters such as the area, length, and width of the wear area to obtain the wear percentage.

[0019] As a further preferred embodiment of this technical solution: the geometric parameters of the wear region include the area, length, and width of the wear region; the wear percentage is calculated based on the ratio of the wear region area to the total area of ​​the tool cutting edge;

[0020] As a further preferred embodiment of this technical solution, the automatic identification method of the tool information includes: identifying the model, material, and batch information on the tool shank engraving or label through OCR, or identifying through QR code, or reading the tool information through an RFID reader;

[0021] As a further preferred embodiment of this technical solution: the prediction of the remaining service life adopts a regression model or an empirical model, wherein the empirical model is a linear wear model or an exponential degradation model;

[0022] As a further preferred embodiment of this technical solution: the alarm signal is an audible and visual alarm, and the location and degree of wear of the wear area are visually displayed on the screen;

[0023] As a further preferred embodiment of this technical solution: when acquiring the image of the cutting edge of the tool, an industrial camera is used in conjunction with a ring light source, a coaxial light source, or a backlight for illumination;

[0024] As a further preferred embodiment of this technical solution: the identification of wear type can also use traditional machine learning methods instead of deep learning models. The traditional machine learning methods include support vector machines or random forests, combined with hand-designed features for identification. The hand-designed features include HOG or LBP features.

[0025] As a further preferred embodiment of this technical solution, it includes: an image acquisition module, an image processing module, a data storage module, and an information output module;

[0026] The image acquisition module includes an industrial camera, a lens, and a ring light source, used to acquire images of the cutting edge of the tool;

[0027] The image processing module includes an embedded processor or industrial computer for running image processing and recognition algorithms, connected to the image acquisition module, for preprocessing the acquired images, segmenting the tool region, extracting wear features, and identifying tool information.

[0028] The data storage module includes: storing tool templates, historical wear data, and model parameters, and is connected to the image processing module for storing tool templates, historical wear data, and model parameters;

[0029] The information output module includes a display screen and an audible and visual alarm. It is also connected to the image processing module and the remaining service life prediction module. It is used to output the wear percentage, wear type, tool information and remaining service life, and to issue an alarm signal when the wear exceeds a set threshold.

[0030] Compared with the prior art, the beneficial effects of the present invention are:

[0031] 1. In this invention, the cutting edge region is segmented at the pixel level using a deep learning model, enabling simultaneous identification of multiple wear types such as breakage, cracks, and wear bands. This overcomes the limitation of existing technologies that can only identify severe breakage. Furthermore, by calculating the ratio of the wear area to the total cutting edge area, the percentage of wear is accurately quantified, replacing the traditional manual qualitative judgment method and significantly improving detection accuracy and consistency. In addition, optimized image segmentation algorithms such as Otsu's method and Canny edge detection are used, combined with image preprocessing steps, to effectively enhance the contrast of the cutting edge region. This ensures stable extraction of the cutting edge contour even under complex lighting and noise environments, providing high-quality input images for subsequent wear identification.

[0032] 2. In this invention, by integrating automatic tool information identification (OCR / QR code / RFID) and remaining service life prediction functions, based on the current wear level and historical wear data, combined with a linear wear model or exponential degradation model, the usable time of the tool or the number of remaining workpieces can be predicted in real time. This realizes an end-to-end automated detection process from image acquisition and wear identification to life prediction, which greatly improves the level of intelligence in tool management.

[0033] 3. In this invention, through the coordinated work of modules such as industrial camera, ring light source, industrial control computer, and sound and light alarm, a complete inspection can be completed within 2-3 seconds, and an alarm will be automatically triggered when the wear exceeds the threshold. This effectively reduces manual intervention, avoids workpiece scrapping caused by sudden tool failure, and significantly improves the production efficiency and reliability of CNC machining. Attached Figure Description

[0034] Figure 1 This is a flowchart of the tool wear detection method and apparatus based on vision recognition according to the present invention. Detailed Implementation

[0035] 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.

[0036] Example:

[0037] Please see Figure 1 As shown, the present invention provides a technical solution: a tool wear detection method and apparatus based on visual recognition, specifically including the following steps S1 to S5:

[0038] S1: Image Acquisition and Preprocessing

[0039] Image acquisition: A high-resolution industrial camera is used in conjunction with a ring LED light source. The camera is fixedly installed inside the CNC machine tool or the tool inspection station. After the tool is processed or during periodic shutdown, the control system triggers the camera to take pictures and acquire the original image of the tool cutting edge area. The light source illuminates the tool surface evenly at a 45° angle to avoid shadows and reflection interference.

[0040] Image preprocessing:

[0041] The acquired raw images are processed sequentially as follows:

[0042] Grayscale conversion: Converting a color image to a grayscale image reduces the amount of data.

[0043] Filtering and noise reduction: Median filtering or Gaussian filtering is used to remove noise generated during image acquisition;

[0044] Histogram equalization: Adjusts the grayscale distribution of an image to enhance overall contrast;

[0045] Image enhancement: Gamma correction or sharpening filtering is used to further highlight the details of the blade edge and worn areas;

[0046] After the above preprocessing, the contrast between the blade area and the background is significantly improved, which facilitates subsequent segmentation.

[0047] S2: Tool area segmentation and tool edge positioning

[0048] This step uses a threshold segmentation algorithm or an edge detection algorithm to extract the tool contour and locate the cutting edge region from the preprocessed image, as detailed below:

[0049] Threshold segmentation algorithm: Calculate the gray-level histogram of the preprocessed image, use the Otsu method to automatically find the optimal segmentation threshold T that maximizes the inter-class variance, set pixels with gray values ​​greater than T as foreground and pixels with gray values ​​less than T as background to obtain a binarized image. If the lighting is uneven, an adaptive threshold segmentation method is used to divide the image into multiple sub-blocks, and the threshold is calculated independently for each block.

[0050] Edge detection algorithm: First, smooth the image with a Gaussian filter, then calculate the gradient magnitude and direction of each pixel, retain the points with the largest local gradient through non-maximum suppression, and then use double threshold detection and connect the edges. Since the tool surface may produce broken edges, morphological closing operation is subsequently used to connect the broken parts into a continuous tool contour.

[0051] Blade region localization: After extracting the complete tool outline, based on the prior information of the tool's geometry, such as the blade being located at the lower edge of the tool outline and the blade tip angle typically being 60°, 90°, or 120°, the blade region is automatically located. A rectangular region containing only the blade is cropped from the original image, i.e., the region of interest, for subsequent wear feature extraction.

[0052] S3: Wear Zone Identification and Wear Degree Calculation Based on Deep Learning

[0053] Deep learning model selection: In this embodiment, the U-Net network structure is used to perform pixel-level classification of the blade region. U-Net consists of two parts: an encoder and a decoder. The encoder extracts multi-scale features through continuous convolution and downsampling, and the decoder restores the resolution through upsampling. Skip connections are used to fuse the shallow spatial features of the encoder with the deep semantic features of the decoder, thereby accurately segmenting the wear region. To improve the deep feature extraction capability, Res-UNet can also be used to alleviate the gradient vanishing problem.

[0054] Model training: 5,000 images of tool cutting edges with different wear levels were collected in advance. Professionals labeled three types of regions pixel by pixel: normal region, damaged region (chipping), crack region (micro-crack), and wear zone region (band wear). The model was trained using the cross-entropy loss function and Adam optimizer until the accuracy reached over 95%.

[0055] Wear Recognition and Quantization: The region of interest of the blade clipped by S2 is input into the trained U-Net model. The model outputs a segmentation map of the same size as the input. Each pixel is classified as normal, damaged, cracked, or wear band. The number of pixels in each type of wear region is counted. Combined with the pixel-to-actual size conversion coefficient calibrated by the camera, the area (square millimeters), length (length of the maximum wear band along the blade direction), and width (average wear width perpendicular to the blade direction) of the wear region are calculated.

[0056] The formula for calculating the percentage of wear is:

[0057] Wear percentage = (Total area of ​​worn area / Total area of ​​cutting edge) × 100%;

[0058] The total area of ​​the cutting edge of the tool is obtained in advance through calibration;

[0059] S4: Automatic Tool Information Recognition and Remaining Service Prediction

[0060] Tool information recognition

[0061] This embodiment employs two optional solutions:

[0062] OCR recognition: The engravings installed on the tool shank include: model, material, batch number, and the characters are read using the Tesseract OCR engine;

[0063] RFID-assisted identification: A passive RFID tag is embedded in the tool shank, and an RFID reader is installed at the inspection station to read the tool's unique ID and associated information in a non-contact manner;

[0064] The identified tool information can then be correlated with current wear data and stored in a database;

[0065] Remaining service life prediction: Retrieve historical wear data for the tool from the database, including the percentage of wear and the corresponding number of parts processed or the time for each inspection. Combine this data with the current wear level and use a linear wear model for prediction.

[0066] Assume that the wear degree W and the number of processed parts N approximately satisfy a linear relationship: By fitting historical data using the least squares method to obtain coefficients a and b, the wear level is determined when it reaches the scrap threshold. (When typically set at 80%), the remaining number of workable parts ;

[0067] S5: Results Output and Alarms

[0068] The industrial touchscreen displays the following information in both digital and graphical formats: percentage of wear, type of wear, tool model, and number of remaining workable parts. Simultaneously, different colors are overlaid on the segmented image to show the wear area: red indicates breakage, yellow indicates cracks, and blue indicates wear bands. When the percentage of wear exceeds a set threshold, the control system issues an audible and visual alarm and automatically notifies the machine operator to replace the tool.

[0069] Example 2: Vision-based tool wear detection device

[0070] This embodiment provides an apparatus for implementing the above method. Referring to the figure, the apparatus includes: an image acquisition module, an image processing module, a data storage module, and an information output module.

[0071] The device includes, in sequence, an industrial camera, a lens, a ring light source, an embedded processor or industrial computer, a solid-state drive, a display screen, and an audible and visual alarm.

[0072] The device's workflow is as follows:

[0073] After the operator places the tool to be inspected at the inspection station, presses the start button, the industrial control computer triggers the camera to take pictures and turns on the ring light source. After acquiring the image, it automatically executes steps S1-S5 and finally outputs the results on the display screen. The whole process takes about 2-3 seconds, realizing rapid, automatic and quantitative detection of tool wear.

[0074] Alternative embodiments

[0075] In practical applications, some technical features of the above embodiments can be replaced with equivalent substitutions, including but not limited to:

[0076] Light source replacement: For highly reflective cutting tools, replace the ring light source with a coaxial light source or a backlight source to reduce specular reflection interference;

[0077] Wear identification model alternative: If on-site computing power is limited, deep learning can be avoided and support vector machine combined with HOG features can be used to classify wear areas. Specifically, extract the HOG feature vector of the blade area image, input it into the trained SVM classifier, and output the wear type.

[0078] Tool information identification replacement: For scenarios where tools are frequently changed, RFID readers are preferred over OCR to achieve non-contact, batch reading;

[0079] Lifetime prediction model alternatives: When the wear process is nonlinear, Gaussian process regression or shallow neural networks can be used to replace linear regression, resulting in higher prediction accuracy.

[0080] All of the above alternative solutions fall within the protection scope of this invention.

[0081] Technical effectiveness verification:

[0082] The method and apparatus of this embodiment were used to inspect 30 milling cutters in a factory, and the results were compared with those obtained by manual microscopy. The average absolute error of the wear percentage was 2.3%, the accuracy of wear type identification was 96.7%, and the average time for a single inspection was 2.8 seconds. Compared with manual visual inspection, which takes an average of 30 seconds per piece and has an error of ±15%, this invention significantly improves the inspection efficiency and accuracy, and can predict the remaining tool life in advance, effectively reducing the scrapping of workpieces due to sudden tool failure.

[0083] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A tool wear detection method based on vision recognition, characterized in that, Includes the following steps: S1: Acquire the image of the cutting edge of the tool and preprocess the acquired image to enhance the contrast of the cutting edge area; S2: Perform tool region segmentation on the preprocessed image, extract the tool contour and locate the cutting edge region; S3: Use a deep learning model to perform pixel-level classification of the blade area, identify wear types including breakage, cracks, and wear bands, and calculate the geometric parameters of the wear area to obtain the wear percentage; S4: Automatically identifies tool information and predicts the remaining tool life based on the current wear percentage, historical wear data, tool material, and machining parameters; S5: Outputs wear percentage, wear type, tool information and remaining service life, and issues an alarm signal when the wear exceeds the set threshold; The tool region segmentation employs either a threshold segmentation algorithm or an edge detection algorithm. The specific steps are as follows: The tool contour is extracted from the preprocessed image using a threshold segmentation algorithm or an edge detection algorithm, and the cutting edge region is located. The threshold segmentation algorithm includes: calculating the image grayscale histogram, determining the optimal segmentation threshold using Otsu's method or an adaptive threshold segmentation method, binarizing the image, and separating the tool area from the background area. The edge detection algorithm includes: using the Canny edge detection operator or the Sobel operator to calculate the image gradient magnitude, detecting tool edge points, and connecting the broken edges through morphological closing operations to form a continuous tool profile; After extracting the tool profile, the cutting edge region is located based on the prior information of the tool geometry, and the region of interest containing only the cutting edge is cropped for subsequent wear feature extraction.

2. The tool wear detection method based on vision recognition according to claim 1, characterized in that: The image preprocessing includes grayscale conversion, filtering and noise reduction, histogram equalization, and image enhancement.

3. The tool wear detection method and apparatus based on vision recognition according to claim 1, characterized in that: The deep learning model is a U-Net or ResNet network structure, which performs pixel-level classification of the blade area, identifies areas such as damage, cracks, and wear bands, and calculates geometric parameters such as the area, length, and width of the wear area to obtain the wear percentage.

4. The tool wear detection method based on vision recognition according to claim 1, characterized in that: The geometric parameters of the wear region include the area, length, and width of the wear region; the wear percentage is calculated based on the ratio of the wear region area to the total area of ​​the tool cutting edge.

5. The tool wear detection method based on vision recognition according to claim 1, characterized in that: The automatic identification methods for the tool information include: identifying the model, material, and batch information on the tool shank engraving or label through OCR, or identifying it through a QR code, or reading the tool information through an RFID reader.

6. The tool wear detection method based on vision recognition according to claim 1, characterized in that: The prediction of the remaining service life is made using a regression model or an empirical model, wherein the empirical model is a linear wear model or an exponential degradation model.

7. The tool wear detection method based on vision recognition according to claim 1, characterized in that: The alarm signal is an audible and visual alarm, and the location and degree of wear of the worn area are visually displayed on the screen.

8. The tool wear detection method based on vision recognition according to claim 1, characterized in that: When acquiring images of the cutting edge of the tool, an industrial camera is used in conjunction with a ring light source, a coaxial light source, or a backlight for illumination.

9. The tool wear detection method based on vision recognition according to claim 1, characterized in that: The identification of wear types can also use traditional machine learning methods instead of deep learning models. These traditional machine learning methods include support vector machines or random forests, combined with hand-designed features for identification. These hand-designed features include HOG or LBP features.

10. The tool wear detection device based on vision recognition according to claims 1-9, characterized in that: include: Image acquisition module, image processing module, data storage module, information output module; The image acquisition module includes an industrial camera, a lens, and a ring light source, used to acquire images of the cutting edge of the tool; The image processing module includes an embedded processor or industrial computer for running image processing and recognition algorithms, connected to the image acquisition module, for preprocessing the acquired images, segmenting the tool region, extracting wear features, and identifying tool information. The data storage module includes: storing tool templates, historical wear data, and model parameters, and is connected to the image processing module for storing tool templates, historical wear data, and model parameters; The information output module includes a display screen and an audible and visual alarm. It is also connected to the image processing module and the remaining service life prediction module. It is used to output the wear percentage, wear type, tool information and remaining service life, and to issue an alarm signal when the wear exceeds a set threshold.

Images