A tool wear monitoring system based on numerical control lathe tool use video

Abstract

The application relates to the technical field of tool monitoring, and relates to a tool wear monitoring system based on a numerical control lathe tool use video.The system comprises a video acquisition and motion analysis module, a texture evolution analysis module, a key frame determination and generation module, a visual feature extraction module, a time sequence change rule analysis module, a feature screening module and a wear determination module, can acquire a continuous video stream of a machining area, can identify a tool and a workpiece contact area, can track a pixel motion track, can record texture structure changes to generate a surface texture evolution sequence, can determine key video frames and generate a sequence according to the surface texture evolution sequence, can extract primary wear features, can analyze feature dimension time sequence change rules to obtain a time sequence change curve, can compare the curve with a reference curve form, can screen and refine wear features, and can generate a wear progress determination result according to a time sequence evolution trend.The application can improve the efficiency of tool wear monitoring.

Description

Technical Field

[0001] This invention relates to the field of tool monitoring technology, and in particular to a tool wear monitoring system based on video of CNC lathe tool usage. Background Technology

[0002] Existing tool wear monitoring methods mostly rely on manual inspection or data collection by a single sensor. Manual inspection is limited by operational experience, is highly subjective and inefficient, and is difficult to capture the dynamic wear details of the tool in real time during high-speed machining. Single sensors are easily affected by machining environment factors such as cutting vibration, oil, and temperature, resulting in insufficient accuracy of the collected data and an inability to comprehensively and accurately reflect the complete state of tool wear.

[0003] Traditional monitoring technologies lack in-depth mining of visual information in the processing area and fail to fully utilize key features such as motion trajectories and texture evolution contained in video streams. This results in incomplete and inaccurate extraction of wear features, leading to delayed judgment of wear progress and failure to provide timely warnings of tool failure risks. This affects processing accuracy and production continuity. At the same time, the data processing flow is cumbersome and the efficiency of converting effective monitoring information is low. Therefore, how to improve the efficiency of tool wear monitoring has become an urgent problem to be solved. Summary of the Invention

[0004] To achieve the above objectives, this invention provides a tool wear monitoring system based on video of CNC lathe tool usage. The system comprises a video acquisition and motion analysis module, a texture evolution analysis module, a keyframe determination and generation module, a visual feature extraction module, a temporal variation law analysis module, a feature filtering module, and a wear determination module, wherein:

[0005] The video acquisition and motion analysis module is used to acquire a continuous video stream of the CNC lathe machining area, identify the contact area between the target object and the workpiece in the continuous video stream, and track the displacement of pixels within the contact area to obtain the pixel motion trajectory of the target object.

[0006] The texture evolution analysis module is used to record the texture structure changes in the contact area to generate a surface texture evolution sequence of the target object.

[0007] The keyframe determination and generation module is used to determine key video frames in the continuous video stream based on the directional consistency of the pixel motion trajectory and the continuous smoothness of the surface texture evolution sequence, so as to generate a keyframe sequence of the target object.

[0008] The visual feature extraction module is used to extract the visual features of key regions of the target object in the keyframe sequence as the initial wear features of the target object.

[0009] The temporal variation pattern analysis module is used to analyze the temporal variation pattern of the feature dimension in the initial wear feature on the key frame sequence, and obtain the temporal variation curve of the feature dimension.

[0010] The feature filtering module is used to compare the time-series change curve with a preset reference curve in terms of shape, and based on the comparison result, filter out the feature dimensions that are consistent with the shape of the reference curve to obtain the refined wear characteristics of the target object.

[0011] The wear determination module is used to generate a wear process determination result of the target object based on the temporal evolution trend of the refined wear characteristics with the key frame sequence.

[0012] In a preferred embodiment, when the video acquisition and motion analysis module acquires a continuous video stream of the CNC lathe machining area, identifies the contact area between the target object and the workpiece in the continuous video stream, and tracks the displacement of pixels within the contact area to obtain the pixel motion trajectory of the target object, it is specifically used for:

[0013] Acquire a continuous video stream of the CNC lathe machining area, and extract the edge contours representing the workpiece position in the continuous video stream to determine the workpiece area of ​​the continuous video stream;

[0014] In adjacent video frames of the continuous video stream, the pixel brightness changes of the workpiece area are compared to obtain the contact area of ​​the target object;

[0015] The pixels with significant texture within the contact area are used as the feature points to be tracked for the target object.

[0016] The local image patch of the feature point to be tracked is matched with the corresponding image region of the subsequent video frame in the continuous video stream to obtain the new location information of the feature point to be tracked.

[0017] Starting from the original position of the feature point to be tracked and ending at the new position information, a movement path of the feature point to be tracked is constructed, and the movement path is integrated into the pixel motion trajectory of the target object.

[0018] In a preferred embodiment, when the texture evolution analysis module records the texture structure changes in the contact area to generate a surface texture evolution sequence of the target object, it is specifically used for:

[0019] Extract an image block of the contact area from the continuous video stream;

[0020] The image block is converted to grayscale to obtain a single-channel brightness image of the image block;

[0021] Gradient quantization is performed on the pixels in the single-channel brightness image to obtain the gradient magnitude and gradient direction angle of the pixels.

[0022] According to the gradient direction angle, the pixel is assigned to the corresponding preset direction interval, and the gradient magnitude of the pixel is accumulated into the accumulated value of the assigned direction interval to construct the gradient direction histogram of the single-channel brightness image.

[0023] Based on the gradient direction histogram, statistical features of texture distribution characteristics in the contact area are derived, and the statistical features are arranged into the texture feature vector of the target object.

[0024] Based on the temporal order of video frames in the continuous video stream, the texture feature vectors are temporally concatenated to obtain the surface texture evolution sequence of the target object.

[0025] In a preferred embodiment, the gradient magnitude is calculated using the following formula:

[0026] ;

[0027] In the formula, The gradient magnitude, The gradient component of the pixel in the horizontal direction. The gradient component of the pixel in the vertical direction. The norm order is used to control the fusion method of horizontal and vertical gradient components. This is the preset contrast enhancement factor. The local contrast of the pixel. The preset anisotropic modulation coefficients, It is a very small positive number. It is a logarithmic function.

[0028] In a preferred embodiment, when the keyframe determination and generation module determines key video frames in the continuous video stream based on the directional consistency of the pixel motion trajectory and the smoothness of the surface texture evolution sequence to generate a keyframe sequence for the target object, it is specifically used for:

[0029] The degree of directional deviation of the pixel motion trajectory is quantified to obtain the first evaluation sequence of the pixel motion trajectory;

[0030] The numerical transition amplitude of the surface texture evolution sequence is detected to obtain a second evaluation sequence of the surface texture evolution sequence;

[0031] The evaluation features of the video frames in the first evaluation sequence and the second evaluation sequence are coupled to obtain the comprehensive judgment index of the video frames.

[0032] Based on the comparison result between the comprehensive judgment index and the preset threshold, candidate keyframes that meet the conditions are marked in the continuous video stream.

[0033] The candidate keyframes are optimized for temporal distribution to obtain the keyframe sequence of the target object.

[0034] In a preferred embodiment, when the visual feature extraction module extracts visual features of key regions of the target object in the keyframe sequence as the initial wear features of the target object, it is specifically used for:

[0035] Based on the correspondence between the keyframe sequence and the continuous video stream, the contact area is mapped onto the keyframe image in the continuous video stream, and the image block corresponding to the contact area is extracted as a sub-image of the key area.

[0036] The sub-image is decomposed into multiple scales to obtain the scale space representation of the sub-image;

[0037] In the scale-space representation, texture features and contour features are extracted from the sub-image;

[0038] The texture features and the contour features are tensor synthesized to obtain the visual feature vector of the key region.

[0039] Temporal feature aggregation is performed on the visual feature vector to obtain the initial wear features of the target object.

[0040] In a preferred embodiment, when the temporal variation pattern analysis module analyzes the temporal variation pattern of the feature dimension in the initial wear feature on the keyframe sequence to obtain the temporal variation curve of the feature dimension, it is specifically used for:

[0041] Using the time axis of the keyframe sequence as the temporal x-axis and the values ​​of the same feature dimension in the initial wear features as the y-axis, the original temporal sequence of the feature dimension is constructed.

[0042] The original time series sequence is subjected to trend decomposition to obtain the low-frequency components of the original time series sequence;

[0043] On the time-series horizontal axis, a linear trend of the low-frequency component is fitted, and the direction of the linear trend is taken as the overall change direction of the low-frequency component;

[0044] By projecting the data points in the low-frequency components onto the overall direction of change, a standard time-series signal of the low-frequency components is obtained.

[0045] The standard time-series signal is smoothed and interpolated to obtain the time-series variation curve of the feature dimension.

[0046] In a preferred embodiment, when the time-series variation pattern analysis module projects the data points in the low-frequency components onto the overall variation direction to obtain the standard time-series signal of the low-frequency components, it is specifically used for:

[0047] Based on the overall change direction, construct the projection direction vector of the overall change direction;

[0048] Identify outliers in the low-frequency components that deviate from the overall direction of change;

[0049] Based on the degree of deviation of the outliers, projected confidence weights are assigned to the outliers.

[0050] The projection confidence weights are applied to perform weighted correction on the projection values ​​of the outliers, and the original values ​​of the outliers are replaced according to the correction results to obtain the corrected low-frequency component sequence of the low-frequency components.

[0051] The data points in the corrected low-frequency component sequence are projected along the projection direction vector to obtain the standard time-series signal of the low-frequency component.

[0052] In a preferred embodiment, when the feature filtering module performs a morphological comparison between the time-series variation curve and a preset reference curve, and filters out feature dimensions consistent with the shape of the reference curve based on the comparison results to obtain the refined wear characteristics of the target object, it is specifically used for:

[0053] The timing variation curve is time-registered with a preset reference curve to obtain the aligned timing variation curve.

[0054] A local window is selected along the time axis on the aligned timing change curve;

[0055] Within the local window, the local morphological similarity between the aligned temporal change curve and the reference curve is calculated to obtain the local similarity value of the target object. The formula for calculating the local similarity value is as follows:

[0056] ;

[0057] In the formula, The local similarity value is... The preset amplitude error weighting coefficient, The number of sampling points within the local window. The aligned timing variation curve within the local window is the [missing information]. The value of each sampling point The reference curve within the local window is the first... The value of each sampling point These are the preset gradient error weighting coefficients. It is an exponential function;

[0058] When the local similarity value is higher than the preset consistency threshold, the feature dimension corresponding to the alignment time-series change curve is identified, and the candidate feature dimension of the target object is obtained.

[0059] The candidate feature dimensions are aggregated into the refined wear features of the target object.

[0060] In a preferred embodiment, when the wear determination module generates a wear process determination result for the target object based on the temporal evolution trend of the refined wear characteristics with the keyframe sequence, it is specifically used for:

[0061] Based on the temporal order of the keyframe sequence, the refining wear features are organized into a wear feature temporal evolution matrix;

[0062] The wear feature time-series evolution matrix is ​​matched piecewise with the historical wear stage reference pattern of the target object to obtain the shape matching mapping relationship of the target object.

[0063] Based on the morphological matching mapping relationship, identify the key stage boundary points in the wear feature time-series evolution matrix that correspond to the historical wear stage reference pattern;

[0064] Based on the key stage boundary points, the key frame sequence is divided into continuous wear stages;

[0065] The degree of wear is calibrated to obtain the stage identifier of the wear stage;

[0066] By combining the evolution order of the stage identifiers with the results of the wear stage division, a wear process determination result for the target object is generated.

[0067] Compared with the prior art, the present invention has the following beneficial effects:

[0068] 1. This invention accurately acquires continuous video streams of the CNC lathe machining area, efficiently identifies the contact area between the tool and the workpiece, tracks the displacement of pixels within the contact area to form a complete motion trajectory, and simultaneously records changes in texture structure to generate a surface texture evolution sequence. This comprehensively captures key visual information related to tool wear, making initial wear feature extraction more comprehensive and targeted. The keyframe determination mechanism filters valid frames based on the consistency of motion trajectory direction and the continuity of texture evolution, significantly reducing redundant data interference and improving the efficiency and accuracy of subsequent feature processing.

[0069] 2. This invention deeply analyzes the temporal variation patterns of initial wear characteristics to generate precise temporal variation curves. By comparing these curves with baseline curves, refined wear characteristics are selected. Combined with the temporal evolution trends of these characteristics, the wear process is scientifically determined. This technology can monitor tool wear status in real-time and dynamically, accurately classify wear stages and calibrate their degree, providing reliable data support for tool maintenance. It effectively ensures machining accuracy and production continuity, significantly improves the efficiency and accuracy of tool wear monitoring, and reduces production risks caused by tool failure. Attached Figure Description

[0070] Figure 1 This is a system architecture diagram of a tool wear monitoring system based on video of CNC lathe tool usage, provided in an embodiment of the present invention.

[0071] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0072] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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 belong to some, but not all, embodiments of the present invention. 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.

[0073] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “said” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0074] Depending on the context, the word "if" or "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0075] Furthermore, the timing of the steps in the following method embodiments is merely an example and not a strict limitation.

[0076] In practice, the server-side equipment deployed in a tool wear monitoring system based on CNC lathe tool usage video may consist of one or more devices. This tool wear monitoring system based on CNC lathe tool usage video can be implemented as: a business instance, a virtual machine, or hardware devices. For example, this tool wear monitoring system based on CNC lathe tool usage video can be implemented as a business instance deployed on one or more devices in a cloud node. Simply put, this tool wear monitoring system based on CNC lathe tool usage video can be understood as software deployed on a cloud node, used to provide tool wear monitoring based on CNC lathe tool usage video to various user terminals. Alternatively, this tool wear monitoring system based on CNC lathe tool usage video can also be implemented as a virtual machine deployed on one or more devices in a cloud node. This virtual machine contains application software for managing various user terminals. Alternatively, this tool wear monitoring system based on CNC lathe tool usage video can also be implemented as a server composed of numerous identical or different types of hardware devices, with one or more hardware devices configured to provide tool wear monitoring based on CNC lathe tool usage video to various user terminals.

[0077] In terms of implementation, a tool wear monitoring system based on CNC lathe tool usage video and the user terminal are mutually compatible. That is, if the tool wear monitoring system based on CNC lathe tool usage video is implemented as an application installed on a cloud service platform, then the user terminal is a client that establishes a communication connection with the application; or if the tool wear monitoring system based on CNC lathe tool usage video is implemented as a website, then the user terminal is implemented as a webpage; or if the tool wear monitoring system based on CNC lathe tool usage video is implemented as a cloud service platform, then the user terminal is implemented as a mini-program in an instant messaging application.

[0078] like Figure 1 The figure shown is a system architecture diagram of a tool wear monitoring system based on video of CNC lathe tool usage provided in an embodiment of the present invention.

[0079] The tool wear monitoring system 100 based on CNC lathe tool usage video described in this invention can be set up in a cloud server. In terms of implementation, it can be used as one or more service devices, or as an application installed in the cloud (e.g., a mobile service operator's server, server cluster, etc.), or it can be developed into a website. Depending on the functions implemented, the tool wear monitoring system 100 based on CNC lathe tool usage video may include a video acquisition and motion analysis module 101, a texture evolution analysis module 102, a keyframe determination and generation module 103, a visual feature extraction module 104, a temporal change law analysis module 105, a feature filtering module 106, and a wear determination module 107. The modules described in this invention can also be called units, referring to a series of computer program segments that can be executed by an electronic device's processor and perform a fixed function, stored in the electronic device's memory.

[0080] In this embodiment of the invention, in a tool wear monitoring system based on CNC lathe tool usage video, each of the above modules can be implemented independently and can be called upon with other modules. This "calling" can be understood as one module connecting to multiple modules of another type and providing corresponding services to those connected modules. The tool wear monitoring system based on CNC lathe tool usage video provided by this embodiment of the invention allows for adjustment of the system's applicability by adding modules and directly calling them, without modifying the program code. This enables cluster-based horizontal expansion, achieving the goal of quickly and flexibly expanding the tool wear monitoring system based on CNC lathe tool usage video. In practical applications, the above modules can be set in the same device or different devices, or in virtual devices, such as service instances in a cloud server.

[0081] The following describes, with reference to specific embodiments, the various components and specific workflow of a tool wear monitoring system based on CNC lathe tool usage video:

[0082] The video acquisition and motion analysis module 101 is used to acquire a continuous video stream of the CNC lathe machining area, identify the contact area between the target object and the workpiece in the continuous video stream, and track the displacement of the pixels in the contact area to obtain the pixel motion trajectory of the target object.

[0083] In this embodiment of the invention, when the video acquisition and motion analysis module acquires a continuous video stream of the CNC lathe machining area, identifies the contact area between the target object and the workpiece in the continuous video stream, and tracks the displacement of pixels within the contact area to obtain the pixel motion trajectory of the target object, it is specifically used for:

[0084] Acquire a continuous video stream of the CNC lathe machining area, and extract the edge contours representing the workpiece position in the continuous video stream to determine the workpiece area of ​​the continuous video stream;

[0085] In adjacent video frames of the continuous video stream, the pixel brightness changes of the workpiece area are compared to obtain the contact area of ​​the target object;

[0086] The pixels with significant texture within the contact area are used as the feature points to be tracked for the target object.

[0087] The local image patch of the feature point to be tracked is matched with the corresponding image region of the subsequent video frame in the continuous video stream to obtain the new location information of the feature point to be tracked.

[0088] Starting from the original position of the feature point to be tracked and ending at the new position information, a movement path of the feature point to be tracked is constructed, and the movement path is integrated into the pixel motion trajectory of the target object.

[0089] A high-definition industrial camera is used to continuously capture images of the CNC lathe machining area. These continuously captured images are arranged in chronological order to form a continuous video stream of the CNC lathe machining area. Each video frame in the continuous video stream is converted to grayscale, retaining only brightness information. Then, the grayscale image is scanned row by row and column by column, comparing the brightness difference between each pixel and its adjacent pixels. When the brightness difference between adjacent pixels exceeds a fixed standard, the pixel is marked as an edge point. All consecutively connected edge points are connected sequentially to form a closed contour line. The image area enclosed by this closed contour line is the workpiece area representing the workpiece position.

[0090] Following the temporal order of the continuous video stream, two adjacent video frames are selected as a set of comparison images. The workpiece area in these two frames is extracted, and then the pixel brightness values ​​at the same coordinate position in the two sets of workpiece areas are compared pixel by pixel. The coordinates of all pixels whose brightness values ​​change are recorded. The pixels corresponding to these coordinates are integrated. When the integrated pixels form a continuous and complete area, this area is the contact area between the tool and the workpiece, where the tool is the target object.

[0091] Observe the brightness distribution of all pixels in the contact area. When the neighboring pixels around a certain pixel show a clear alternation of light and dark, or when the pixel and its surrounding pixels maintain a stable brightness difference, and this brightness distribution feature is unique in the contact area, the pixel is a pixel with significant texture. All pixels that meet the above characteristics are selected as the feature points to be tracked by the tool.

[0092] For each feature point to be tracked, a fixed number of pixels are extended upwards, downwards, leftwards, and rightwards from the feature point to form a rectangular local image patch. This local image patch contains the texture and brightness information of the feature point to be tracked and its surroundings. Each video frame following the current processing frame in the continuous video stream is selected sequentially. Within the workpiece region of each subsequent video frame, a search is performed region by region from left to right and from top to bottom. The brightness distribution pattern and texture structure of each searched region are compared with those of the local image patch. When a region is found to perfectly match the brightness distribution and texture structure of the local image patch, the coordinates of the center pixel of that region are the new position information of the tool's feature point to be tracked.

[0093] Record the pixel coordinates of each feature point to be tracked in the initial video frame, and use these coordinates as the starting point of the movement path. Then, use the pixel coordinates corresponding to the new position information obtained through matching in subsequent video frames, in chronological order, as the intermediate nodes or ending points of the movement path. Connect the starting point, intermediate nodes, and ending point with continuous lines to form the movement path of a single feature point to be tracked. Collect all the movement paths of the feature points to be tracked, and remove abnormal path segments that are discontinuous or deviate from the overall movement direction due to image interference. Summarize and integrate the remaining valid movement paths to finally form a pixel motion trajectory that can completely reflect the tool's movement state.

[0094] The beneficial effects are that by accurately acquiring video streams step by step, determining the workpiece area, locking the contact area, filtering feature points, matching position information, and constructing motion trajectories, the interaction process and motion state between the tool and the workpiece can be captured comprehensively and accurately. This provides real and reliable basic data for subsequent texture evolution analysis, key frame determination, and other steps, ensuring that the initial data acquisition stage of the entire monitoring system is accurate and efficient. At the same time, the clear and complete pixel motion trajectory also provides a key basis for subsequent wear feature extraction and wear process determination, improving the stability and accuracy of the entire monitoring system.

[0095] The texture evolution analysis module 102 is used to record the texture structure changes of the contact area to generate a surface texture evolution sequence of the target object.

[0096] In this embodiment of the invention, when the texture evolution analysis module records the texture structure changes of the contact area to generate a surface texture evolution sequence of the target object, it is specifically used for:

[0097] Extract an image block of the contact area from the continuous video stream;

[0098] The image block is converted to grayscale to obtain a single-channel brightness image of the image block;

[0099] Gradient quantization is performed on the pixels in the single-channel brightness image to obtain the gradient magnitude and gradient direction angle of the pixels.

[0100] According to the gradient direction angle, the pixel is assigned to the corresponding preset direction interval, and the gradient magnitude of the pixel is accumulated into the accumulated value of the assigned direction interval to construct the gradient direction histogram of the single-channel brightness image.

[0101] Based on the gradient direction histogram, statistical features of texture distribution characteristics in the contact area are derived, and the statistical features are arranged into the texture feature vector of the target object.

[0102] Based on the temporal order of video frames in the continuous video stream, the texture feature vectors are temporally concatenated to obtain the surface texture evolution sequence of the target object.

[0103] The formula for calculating the gradient magnitude is as follows:

[0104] ;

[0105] In the formula, The gradient magnitude, The gradient component of the pixel in the horizontal direction. The gradient component of the pixel in the vertical direction. The norm order is used to control the fusion method of horizontal and vertical gradient components. This is the preset contrast enhancement factor. The local contrast of the pixel. The preset anisotropic modulation coefficients, It is a very small positive number. It is a logarithmic function.

[0106] From each video frame of the continuous video stream, based on the pixel coordinate range of the identified contact area, determine the pixel coordinates of the upper left and lower right corners of the contact area, extract all pixels contained between these two coordinates, and form a rectangular image area that completely corresponds to the contact area. This area is the image block of the contact area.

[0107] For each captured image block, process each pixel individually to obtain the brightness values ​​of the red, green, and blue channels of each pixel. Add the brightness values ​​of these three channels together and divide by 3 to obtain the grayscale value of the pixel. Replace the original three channel values ​​of the pixel with this grayscale value. After all pixels have been processed, a single-channel brightness image containing only grayscale information is formed.

[0108] For each pixel in a single-channel luminance image, the luminance difference between the pixel and its right-side neighbor is compared to obtain the luminance change value in the horizontal direction, and the luminance difference between the pixel and its lower-side neighbor is compared to obtain the luminance change value in the vertical direction.

[0109] The change in brightness in the horizontal direction is the gradient component in the horizontal direction, and the change in brightness in the vertical direction is the gradient component in the vertical direction.

[0110] The norm order used to control the fusion method of horizontal and vertical gradient components is a fixed value set in advance by the system based on the actual application scenario and accuracy requirements of CNC lathe tool wear monitoring, which is used to standardize the fusion logic of gradient components in the two directions.

[0111] The preset contrast enhancement coefficient is a fixed coefficient set in advance by the system based on the general characteristics of the image texture in the contact area, in order to enhance the influence of local contrast of pixels on gradient magnitude.

[0112] The local contrast of a pixel is calculated by selecting neighboring pixels within a 3×3 area around the pixel, calculating the brightness difference between these neighboring pixels and the pixel, and then averaging all the brightness differences to obtain a value that reflects the degree of local brightness difference.

[0113] The preset anisotropic modulation coefficient is a fixed coefficient set in advance by the system to adjust the effect of the difference between the horizontal and vertical gradient components on the final gradient amplitude, based on the requirements of tool texture monitoring.

[0114] The extremely small positive number is a fixed, tiny value set in advance by the system to avoid the denominator being zero during the calculation process. It does not affect the accuracy of the overall calculation result.

[0115] By combining the gradient components in both the horizontal and vertical directions and integrating them through a fixed fusion logic, a value reflecting the basic intensity of the overall brightness change of the pixel is obtained. Combined with the local contrast of the pixel, the influence of local contrast on brightness change is amplified through logarithmic operations, thereby strengthening the gradient signal corresponding to texture details. Then, by calculating the degree of difference between the horizontal and vertical gradient components and adjusting the contribution of this difference to the gradient amplitude with a preset coefficient, these three results are finally superimposed to obtain the gradient amplitude that accurately reflects the strength of the brightness change of the pixel.

[0116] Based on the proportional relationship between the gradient components in the horizontal and vertical directions, the specific direction of the brightness change of the pixel is determined, and the direction is expressed in the form of an angle, which is the gradient direction angle of the pixel.

[0117] Multiple evenly distributed directional intervals are pre-defined, dividing the range from 0 degrees to 360 degrees into several continuous intervals, with each interval corresponding to a fixed angular range.

[0118] For each pixel, examine the angular range of its gradient direction angle, assign the pixel to the corresponding preset direction interval, and add the gradient magnitude of the pixel to the cumulative value of the preset direction interval.

[0119] After all pixels have been allocated and their magnitudes accumulated, the cumulative gradient magnitudes of each interval are arranged in the order of the preset direction intervals to form a gradient direction histogram of the single-channel brightness image.

[0120] Analyze the cumulative gradient magnitude of each interval in the gradient direction histogram, calculate the proportion of the cumulative value of each interval to the sum of the cumulative values ​​of all intervals, find the value corresponding to the interval with the largest cumulative value, and calculate the average value of the cumulative values ​​of all intervals and the difference between the maximum and minimum values. These data are the statistical characteristics of the texture distribution in the contact area.

[0121] These statistical features are arranged in a predetermined fixed order to form an ordered set of values, which is the texture feature vector of the cutting tool.

[0122] According to the temporal order of video frames in a continuous video stream, the texture feature vectors corresponding to each video frame are sequentially connected. The end of the texture feature vector corresponding to the previous video frame is directly connected to the beginning of the texture feature vector corresponding to the next video frame, forming a continuous and ordered vector sequence. This sequence is the surface texture evolution sequence of the cutting tool.

[0123] The beneficial effects are that by performing precise processing of the contact area image step by step, from image block cropping, grayscale conversion, gradient component acquisition, parameter setting and gradient magnitude calculation, gradient direction angle determination, to direction interval allocation, histogram construction, statistical feature extraction and vector temporal stitching, the texture structure changes of the tool contact area are captured completely and accurately. The generated surface texture evolution sequence can truly reflect the texture state at different time points, providing comprehensive and reliable texture data support for subsequent keyframe determination, wear feature extraction and other links, effectively ensuring the accuracy and effectiveness of the entire wear monitoring system in analyzing texture information.

[0124] The keyframe determination and generation module 103 is used to determine key video frames in the continuous video stream based on the directional consistency of the pixel motion trajectory and the continuous smoothness of the surface texture evolution sequence, so as to generate a keyframe sequence of the target object.

[0125] In this embodiment of the invention, when the keyframe determination and generation module determines key video frames in the continuous video stream based on the directional consistency of the pixel motion trajectory and the continuous smoothness of the surface texture evolution sequence to generate a keyframe sequence for the target object, it is specifically used for:

[0126] The degree of directional deviation of the pixel motion trajectory is quantified to obtain the first evaluation sequence of the pixel motion trajectory;

[0127] The numerical transition amplitude of the surface texture evolution sequence is detected to obtain a second evaluation sequence of the surface texture evolution sequence;

[0128] The evaluation features of the video frames in the first evaluation sequence and the second evaluation sequence are coupled to obtain the comprehensive judgment index of the video frames.

[0129] Based on the comparison result between the comprehensive judgment index and the preset threshold, candidate keyframes that meet the conditions are marked in the continuous video stream.

[0130] The candidate keyframes are optimized for temporal distribution to obtain the keyframe sequence of the target object.

[0131] Using the overall motion direction of each pixel's motion trajectory as the reference direction, the actual direction of each motion segment in the trajectory is examined frame by frame. The deviation in direction is measured by measuring the angle between the actual direction and the reference direction. The angle is converted into a value within a fixed range, which directly corresponds to the degree of deviation in direction. The degree of deviation in direction corresponding to each video frame is arranged sequentially according to the time order of the continuous video stream to form the first evaluation sequence of the pixel motion trajectory.

[0132] For each texture feature vector in the surface texture evolution sequence, the texture feature vectors corresponding to two adjacent video frames are compared in chronological order. The difference between the feature vector value of the next frame and the corresponding dimension value of the previous frame is calculated in each dimension. The absolute value of each difference is taken and the absolute values ​​of all dimensions are accumulated to obtain the total value change between adjacent frames. This total value is the numerical transition amplitude of the surface texture evolution sequence. The numerical transition amplitudes corresponding to each video frame are arranged in order of video frame to form the second evaluation sequence of the surface texture evolution sequence.

[0133] Based on the time sequence of the continuous video stream, the evaluation value corresponding to each video frame in the first evaluation sequence is extracted, and the evaluation value corresponding to the same video frame in the second evaluation sequence is extracted. The two values ​​are added together to obtain the unique comprehensive value of the video frame. This comprehensive value is the comprehensive judgment index of the video frame. The value addition operation of all video frames is completed in time sequence to obtain the comprehensive judgment index of all video frames.

[0134] The system pre-sets a fixed comprehensive judgment threshold and compares the comprehensive judgment index of each video frame with the preset threshold one by one in chronological order. When the comprehensive judgment index of a video frame is greater than or equal to the preset threshold, the video frame is directly marked in the continuous video stream. All marked video frames together constitute candidate keyframes that meet the conditions.

[0135] A fixed minimum time interval is set as the time sequence distribution standard. All candidate keyframes are traversed in chronological order. Starting from the first candidate keyframe, the frame is retained and its time point is recorded. If the time interval between the subsequent candidate keyframe and the previous retained frame is less than the set minimum time interval, the candidate keyframe is directly removed. If the time interval is greater than or equal to the minimum time interval, the frame is retained. The selection of all candidate keyframes is completed according to this rule. The finally retained candidate keyframes are arranged in chronological order to obtain the tool keyframe sequence.

[0136] The beneficial effects are that by accurately quantifying the directional deviation of pixel motion trajectories and the numerical transition amplitude of surface texture evolution sequences, the comprehensive judgment index obtained by combining the two can fully reflect the criticality of video frames. After temporal distribution optimization and elimination of redundant candidate frames, the final key frame sequence not only retains the core information related to tool wear, but also effectively reduces the amount of data, providing an efficient and accurate data foundation for subsequent visual feature extraction and wear feature analysis, and improving the processing efficiency and judgment accuracy of the entire monitoring system.

[0137] The visual feature extraction module 104 is used to extract the visual features of the key regions of the target object in the key frame sequence as the initial wear features of the target object.

[0138] In this embodiment of the invention, when the visual feature extraction module extracts the visual features of key regions of the target object in the keyframe sequence as the initial wear features of the target object, it is specifically used for:

[0139] Based on the correspondence between the keyframe sequence and the continuous video stream, the contact area is mapped onto the keyframe image in the continuous video stream, and the image block corresponding to the contact area is extracted as a sub-image of the key area.

[0140] The sub-image is decomposed into multiple scales to obtain the scale space representation of the sub-image;

[0141] In the scale-space representation, texture features and contour features are extracted from the sub-image;

[0142] The texture features and the contour features are tensor synthesized to obtain the visual feature vector of the key region.

[0143] Temporal feature aggregation is performed on the visual feature vector to obtain the initial wear features of the target object.

[0144] Based on the timestamp of each keyframe in the keyframe sequence, a unique association is established with the corresponding video frame in the continuous video stream, clarifying the pixel coordinate correspondence between the two. The pixel coordinate range of the identified contact area is directly mapped to the same coordinate position in the keyframe image, and all pixels contained within this coordinate range are extracted to form an image block that perfectly matches the contact area. This image block is the sub-image of the key area.

[0145] A fixed scaling ratio is selected, and starting from the original size of the sub-image, the width and height of the sub-image are successively reduced according to the ratio to generate multiple sub-image copies of different sizes. Each copy retains the texture and contour information of the original image. At the same time, edge completion processing is performed on each reduced sub-image to ensure image integrity. All sub-images of different sizes and after completion are arranged from large to small according to the scaling ratio, which together constitute the scale space representation of the sub-image.

[0146] For each size of the sub-image in the scale space representation, the brightness distribution is analyzed pixel by pixel. The repetition pattern of brightness changes between adjacent pixels, the frequency of alternation between light and dark, and the distribution range of areas with the same brightness are statistically analyzed. These statistical results are the texture features of the sub-image. At the same time, the sub-image is scanned row by row and column by column, and the brightness difference between each pixel and its adjacent pixels is compared. When the brightness difference reaches a fixed standard, the pixel is marked as a contour point. All consecutively connected contour points are connected in sequence to form closed or continuous lines. The feature information corresponding to these lines is the contour feature of the sub-image.

[0147] The extracted texture features are transformed into an ordered set of values ​​in a preset order, and the contour features are also transformed into another ordered set of values ​​in the same preset order. Based on the dimensions of the two sets of values, a multidimensional data structure is constructed. The sets of values ​​corresponding to the texture features and the sets of values ​​corresponding to the contour features are embedded into the data structure in sequence according to the dimensions, forming a unified ordered sequence of values ​​containing texture and contour information. This sequence is the visual feature vector of the key region.

[0148] According to the chronological order of the keyframe sequence, the visual feature vectors corresponding to each keyframe are collected. Starting from the visual feature vector of the first keyframe, the visual feature vector of the next keyframe is directly connected to the end of the visual feature vector of the previous keyframe. The connection operation of the visual feature vectors of all keyframes is completed in sequence to form a continuous and ordered comprehensive feature set, which is the initial wear feature of the tool.

[0149] The beneficial effects are that by accurately mapping the contact area, decomposing sub-images at multiple scales, extracting dual features, synthesizing vectors from tensors, and aggregating features over time, the texture and contour information of key areas of the tool at different scales is fully captured. The generated primary wear features contain comprehensive and rich visual information, providing high-quality basic data for subsequent analysis of time-series changes, and effectively ensuring the accuracy and comprehensiveness of tool wear monitoring.

[0150] The temporal variation law analysis module 105 is used to analyze the temporal variation law of the feature dimension in the initial wear feature on the key frame sequence, and obtain the temporal variation curve of the feature dimension.

[0151] In this embodiment of the invention, when the temporal variation law analysis module analyzes the temporal variation law of the feature dimension in the initial wear feature on the keyframe sequence and obtains the temporal variation curve of the feature dimension, it is specifically used for:

[0152] Using the time axis of the keyframe sequence as the temporal x-axis and the values ​​of the same feature dimension in the initial wear features as the y-axis, the original temporal sequence of the feature dimension is constructed.

[0153] The original time series sequence is subjected to trend decomposition to obtain the low-frequency components of the original time series sequence;

[0154] On the time-series horizontal axis, a linear trend of the low-frequency component is fitted, and the direction of the linear trend is taken as the overall change direction of the low-frequency component;

[0155] By projecting the data points in the low-frequency components onto the overall direction of change, a standard time-series signal of the low-frequency components is obtained.

[0156] The standard time-series signal is smoothed and interpolated to obtain the time-series variation curve of the feature dimension.

[0157] When the time-series variation pattern analysis module projects the data points in the low-frequency components onto the overall variation direction to obtain the standard time-series signal of the low-frequency components, it is specifically used for:

[0158] Based on the overall change direction, construct the projection direction vector of the overall change direction;

[0159] Identify outliers in the low-frequency components that deviate from the overall direction of change;

[0160] Based on the degree of deviation of the outliers, projected confidence weights are assigned to the outliers.

[0161] The projection confidence weights are applied to perform weighted correction on the projection values ​​of the outliers, and the original values ​​of the outliers are replaced according to the correction results to obtain the corrected low-frequency component sequence of the low-frequency components.

[0162] The data points in the corrected low-frequency component sequence are projected along the projection direction vector to obtain the standard time-series signal of the low-frequency component.

[0163] Using the chronological order of the keyframe sequence as the temporal axis, each scale on the axis corresponds to a specific time point of a keyframe. The same feature dimension is selected from the initial wear features, and the specific value of each feature dimension corresponding to each keyframe is extracted. Each time point is matched one-to-one with the corresponding feature dimension value, and these correspondences are arranged in chronological order to form the original temporal sequence of the feature dimension.

[0164] A fixed-width sliding window is used to traverse the original time series. The window starts from the beginning of the series and moves one data point at a time. The average value of all data points in each window is calculated. These average values ​​are arranged in the order in which the window moves to form a new sequence. This sequence is the low-frequency component of the original time series. In this way, the fast fluctuation part in the original sequence is removed, while the long-term trend is preserved.

[0165] On the time-series horizontal axis, mark the time coordinate and corresponding numerical coordinate of each data point in the low-frequency component, and draw the distribution graph of all data points. Then draw a straight line, making the line as close as possible to all data points, ensuring that the sum of the vertical distances from each data point to the line is minimized. This straight line is the linear trend of the low-frequency component. The overall direction of change is determined by the direction of the line from left to right. If the line is upward, the overall direction of change is rising; if the line is downward, it is falling; and if the line is horizontal, it is stable.

[0166] Based on the determined overall direction of change, the magnitude of the numerical change is determined by taking the unit length of the time series horizontal axis as a reference and combining the slope of the linear trend. For example, when the overall direction of change is upward, a vector from the lower left to the upper right is constructed based on the fixed magnitude of the numerical increase per unit time. The direction of this vector is completely consistent with the overall direction of change, which is the projection direction vector of the overall direction of change.

[0167] Calculate the vertical distance from each data point in the low-frequency component to the linear trend line, set a fixed distance threshold, and compare the vertical distance of each data point with the threshold. When the vertical distance of a data point is greater than the threshold, the data point is an outlier that deviates from the overall direction of change. Record the position and value of all outliers in sequence.

[0168] For each outlier, calculate the ratio of its vertical distance to the threshold. The larger the ratio, the greater the deviation. Assign a corresponding projection confidence weight based on this ratio. The greater the deviation, the smaller the weight. The smaller the deviation, the closer the weight is to 1. For example, the weight is set to 0.9 when the ratio is 1 and 0.7 when the ratio is 2, to ensure that the weight and the degree of deviation have an inverse relationship.

[0169] First, calculate the projection value of each outlier on the projection direction vector, that is, find the value corresponding to the outlier along the projection direction vector. Then, multiply the projection value by the projection confidence weight assigned to the outlier to obtain the corrected projection value. Replace the original value of the outlier with this corrected projection value. After all outliers are replaced, the corrected low-frequency component sequence of the low-frequency components is formed.

[0170] For each data point in the corrected low-frequency component sequence, extend the projection direction vector from the location of the data point to the projection direction vector to find the corresponding position of each data point on the projection direction vector, record the value corresponding to that position, and arrange the corresponding values ​​of all data points in time order to obtain the standard time sequence signal of the low-frequency component.

[0171] Traverse adjacent data points in the standard time-series signal, calculate the numerical difference between two adjacent data points, and when the difference exceeds a fixed standard, insert a new value between the two data points. This value is the average of the values ​​of the two adjacent data points. Repeat this process to check the difference and interpolate the values ​​of all adjacent data points, so that the numerical change of the entire signal is smooth and the time-series change curve of the feature dimension is obtained.

[0172] The beneficial effects are that by constructing the original time series step by step, extracting low-frequency components, determining the overall direction of change, correcting outliers, and performing projection and smooth interpolation, the core change law of the same dimension in the initial wear characteristics is fully preserved, the influence of fluctuation interference and abnormal data is effectively eliminated, and the generated time series change curve can accurately reflect the time series evolution of the feature dimension. This provides accurate and reliable data support for subsequent morphological comparison with the baseline curve and feature selection, and improves the accuracy of the entire monitoring system in analyzing wear characteristics.

[0173] The feature filtering module 106 is used to compare the time-series change curve with a preset reference curve in terms of shape, and based on the comparison result, filter out the feature dimensions that are consistent with the shape of the reference curve to obtain the refined wear characteristics of the target object.

[0174] In this embodiment of the invention, when the feature filtering module performs a morphological comparison between the time-series change curve and a preset reference curve, and filters out feature dimensions consistent with the shape of the reference curve based on the comparison results to obtain the refined wear characteristics of the target object, it is specifically used for:

[0175] The timing variation curve is time-registered with a preset reference curve to obtain the aligned timing variation curve.

[0176] A local window is selected along the time axis on the aligned timing change curve;

[0177] Within the local window, the local morphological similarity between the aligned temporal change curve and the reference curve is calculated to obtain the local similarity value of the target object. The formula for calculating the local similarity value is as follows:

[0178] ;

[0179] In the formula, The local similarity value is... The preset amplitude error weighting coefficient, The number of sampling points within the local window. The aligned timing variation curve within the local window is the [missing information]. The value of each sampling point The reference curve within the local window is the first... The value of each sampling point These are the preset gradient error weighting coefficients. It is an exponential function;

[0180] When the local similarity value is higher than the preset consistency threshold, the feature dimension corresponding to the alignment time-series change curve is identified, and the candidate feature dimension of the target object is obtained.

[0181] The candidate feature dimensions are aggregated into the refined wear features of the target object.

[0182] Using the time axis of the preset baseline curve as a standard, the start time point, end time point, and key time nodes of the baseline curve are extracted. The time points in the time series curve that match the characteristics of these key time nodes are then found. By adjusting the time scale of the time series curve, the key time points of the time series curve are made to completely coincide with the key time nodes of the baseline curve, ensuring that the two curves correspond one-to-one in the time dimension. The adjusted time series curve is the aligned time series curve.

[0183] A fixed time length is set as the standard for the local window. Starting from the starting time point of the aligned time series curve, a curve segment within the fixed time length is extracted as the first local window. Then, the window moves along the time axis to the ending time point at a fixed time step. After each move, a curve segment of the same time length is extracted as a new local window, until the extracted local window covers the entire time axis of the aligned time series curve.

[0184] The preset amplitude error weighting coefficient is a fixed coefficient set in advance by the system based on the importance attached to the difference in the amplitude of the curve values ​​in tool wear monitoring, combined with historical monitoring data and accuracy requirements. It is used to adjust the influence ratio of amplitude error in the overall calculation. The preset gradient error weighting coefficient is a fixed coefficient set in advance by the system based on the importance attached to the difference in the curve change trend in tool wear monitoring, and with reference to a large amount of measured data and judgment criteria. It is used to adjust the contribution ratio of gradient error in the overall calculation.

[0185] The number of sampling points within a local window is a value obtained by counting each point within the current local window. Specifically, it is the total number of sampling points shared by the aligned time-series change curve and the baseline curve within the local window, which is determined directly by counting the number of sampling points covered by the local window.

[0186] The time series variation curve after alignment is within the local window. The value of the sampling point is obtained by locating the time range corresponding to the current local window from the aligned time change curve after time registration, and extracting the value of the sampling point within that range. The specific value corresponding to the sampling position, the baseline curve within the local window, the _th _ The value of the sampling point is obtained by finding the interval corresponding to the current local window time range from the preset baseline curve, and extracting the value of the sampling point within that interval. The specific value corresponding to the sampling position, and the relationship between the sampling position and the first value of the time series change curve after alignment. The locations of each sampling point correspond perfectly in time.

[0187] Within each local window, the absolute difference between the aligned temporal variation curve values ​​and the reference curve values ​​for all corresponding sampling points is first calculated. These absolute differences are summed and divided by the number of sampling points to obtain the average amplitude error. This average amplitude error is then multiplied by a preset amplitude error weighting coefficient to obtain the amplitude error contribution value. Simultaneously, the numerical changes of adjacent sampling points on the aligned temporal variation curve and adjacent sampling points on the reference curve are calculated. The difference between the two sets of changes is calculated for each pair. All differences are summed and divided by the total number of changes to obtain the average gradient error. This average gradient error is then multiplied by a preset gradient error weighting coefficient to obtain the gradient error contribution value. The amplitude error contribution value and the gradient error contribution value are summed to obtain the comprehensive error value. This comprehensive error value is then negative and substituted into a fixed exponential function for calculation. The exponential function converts the comprehensive error value into a value between 0 and 1. The final result is the local similarity value of the tool. The closer the value is to 1, the more similar the shapes of the two curves are within the local window.

[0188] The system pre-sets a fixed consistency threshold and compares the local similarity value of each local window with this threshold one by one. When the local similarity value of a local window is higher than the pre-set consistency threshold, the feature dimension associated with the aligned time-series change curve of the local window is determined and marked as a feature dimension that meets the conditions. All marked feature dimensions together constitute the candidate feature dimensions of the tool.

[0189] All identified candidate feature dimensions are collected and arranged according to their original order in the primary wear features. Duplicate feature dimensions are removed to form an ordered and unique set of feature dimensions, which is the refining wear feature of the tool.

[0190] The beneficial effects are that temporal registration ensures the consistency of two curves in the time dimension, the selection of local windows enables a fine comparison of curve shapes, the similarity value calculated based on the weighted average of amplitude error and gradient error can comprehensively and accurately quantify the local shape matching of the two curves, and the refined wear features obtained by threshold screening and aggregation eliminate redundant features unrelated to tool wear and retain the core effective features, providing high-quality feature data support for the accurate determination of subsequent wear processes, and improving the reliability and accuracy of the wear monitoring system.

[0191] The wear determination module 107 is used to generate a wear process determination result of the target object based on the temporal evolution trend of the refined wear characteristics with the key frame sequence.

[0192] In this embodiment of the invention, when the wear determination module generates a wear process determination result for the target object based on the temporal evolution trend of the refined wear characteristics with the keyframe sequence, it is specifically used for:

[0193] Based on the temporal order of the keyframe sequence, the refining wear features are organized into a wear feature temporal evolution matrix;

[0194] The wear feature time-series evolution matrix is ​​matched piecewise with the historical wear stage reference pattern of the target object to obtain the shape matching mapping relationship of the target object.

[0195] Based on the morphological matching mapping relationship, identify the key stage boundary points in the wear feature time-series evolution matrix that correspond to the historical wear stage reference pattern;

[0196] Based on the key stage boundary points, the key frame sequence is divided into continuous wear stages;

[0197] The degree of wear is calibrated to obtain the stage identifier of the wear stage;

[0198] By combining the evolution order of the stage identifiers with the results of the wear stage division, a wear process determination result for the target object is generated.

[0199] According to the chronological order of the keyframe sequence, the refined wear features corresponding to each keyframe are treated as a row of data, and each feature dimension contained in the refined wear features is treated as a column of data. The specific values ​​of each dimension in the refined wear features of each keyframe are filled into the intersection of the corresponding row and column. After all the keyframe data are arranged in order, a two-dimensional data structure is formed, which is the wear feature temporal evolution matrix.

[0200] The system pre-stores historical wear stage reference patterns corresponding to different tool wear stages. Each reference pattern is a two-dimensional data structure with the same dimension as the wear feature time-series evolution matrix, containing the feature value distribution and change trend of the corresponding wear stage. The wear feature time-series evolution matrix is ​​divided into multiple continuous sub-matrices of fixed length in chronological order. The number of rows in each sub-matrix is ​​consistent with the number of rows in the historical wear stage reference patterns. Each sub-matrix is ​​compared with all historical wear stage reference patterns, and the absolute difference between the corresponding position values ​​of the sub-matrix and each reference pattern is calculated. All absolute differences are summed to obtain the total difference value. The reference pattern with the smallest total difference value is the reference pattern matched by that sub-matrix. The correspondence between each sub-matrix and its matched reference pattern is recorded to form a tool morphology matching mapping relationship.

[0201] Based on the morphological matching mapping relationship, the starting and ending row numbers of each sub-matrix in the wear feature time-series evolution matrix are located. These row numbers correspond to specific key frames in the key frame sequence. The key frames corresponding to the connecting row numbers between sub-matrices are the key stage boundary points. Each boundary point marks the transition of the wear stage from one stage corresponding to a historical reference pattern to another stage. All key frames corresponding to the boundary points are arranged in chronological order to form a key stage boundary point set.

[0202] Starting with the first keyframe of the keyframe sequence and ending with the keyframe corresponding to the first key stage boundary point, a first continuous keyframe segment is formed. Starting with the keyframe corresponding to the boundary point, and ending with the keyframe corresponding to the next key stage boundary point, a second keyframe segment is formed. All key stage boundary points are processed in sequence according to this rule until the last keyframe of the keyframe sequence is covered. Each keyframe segment is a continuous wear stage.

[0203] Fixed stage identification rules are preset to clearly define the wear severity levels corresponding to different historical wear stage reference patterns. For example, a wear stage matching the initial wear reference pattern is labeled as 1, one matching the intermediate wear reference pattern as 2, and one matching the late wear reference pattern as 3. Based on the historical wear stage reference pattern matched for each wear stage, the corresponding wear severity level is found and used as the stage identifier for that wear stage, thus completing the severity labeling for all wear stages.

[0204] Organize all wear stages in chronological order, record the keyframe range and corresponding time interval of each wear stage, and integrate the evolution order of stage identifiers, keyframe range of each stage, time interval and corresponding wear degree description into a complete information set. This information set is the result of tool wear process determination.

[0205] The beneficial effects are that by organizing the refining wear characteristics into a time-series evolution matrix and combining it with historical wear stage reference patterns for precise segment-by-segment matching, the key dividing points of wear stages can be accurately identified, enabling the scientific division and degree calibration of wear stages. The final wear process judgment result comprehensively and clearly reflects the wear evolution of the tool from the start of use to the present moment, providing workers with an intuitive and reliable tool status reference, facilitating timely formulation of maintenance or replacement strategies, effectively reducing the risk of machining failures caused by excessive tool wear, and ensuring the stability and continuity of production.

[0206] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0207] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0208] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A tool wear monitoring system based on video of a CNC lathe tool use, characterized by, The system comprises a video acquisition and motion analysis module, a texture evolution analysis module, a key frame determination and generation module, a visual feature extraction module, a time sequence change rule analysis module, a feature screening module and a wear determination module, wherein: The video acquisition and motion analysis module is configured to acquire a continuous video stream of a machining area of a numerical control lathe, identify a contact area of a target object and a workpiece in the continuous video stream, and track displacement of a pixel point in the contact area to obtain a pixel motion trajectory of the target object. The texture evolution analysis module is configured to record texture structure changes of the contact area to generate a surface texture evolution sequence of the target object. The key frame determination and generation module is configured to determine key video frames in the continuous video stream according to direction consistency of the pixel motion trajectory and continuous flatness of the surface texture evolution sequence to generate a key frame sequence of the target object, including: quantifying a direction deviation degree of the pixel motion trajectory to obtain a first evaluation sequence of the pixel motion trajectory; detecting a numerical transition amplitude of the surface texture evolution sequence to obtain a second evaluation sequence of the surface texture evolution sequence; performing feature coupling of evaluation features of video frames in the first evaluation sequence and the second evaluation sequence to obtain a comprehensive determination index of the video frames; marking out candidate key frames meeting a condition in the continuous video stream according to a comparison result of the comprehensive determination index and a preset threshold; performing time sequence distribution optimization on the candidate key frames to obtain the key frame sequence of the target object; The visual feature extraction module is configured to extract visual features of a target object key region in the key frame sequence as primary wear features of the target object. The time sequence change rule analysis module is configured to analyze a time sequence change rule of a feature dimension in the primary wear features on the key frame sequence to obtain a time sequence change curve of the feature dimension. The feature screening module is configured to perform shape comparison of the time sequence change curve and a preset reference curve, and screen out a feature dimension consistent in shape with the reference curve according to a comparison result to obtain refined wear features of the target object. The wear determination module is configured to generate a wear progress determination result of the target object according to a time sequence evolution trend of the refined wear features along with the key frame sequence.

2. A tool wear monitoring system based on video of a CNC lathe tool use according to claim 1, characterized in that, When the video acquisition and motion analysis module is executed to acquire a continuous video stream of a machining area of a numerical control lathe, identify a contact area of a target object and a workpiece in the continuous video stream, and track displacement of a pixel point in the contact area to obtain a pixel motion trajectory of the target object, it is specifically configured to: acquire a continuous video stream of a machining area of a numerical control lathe, extract an edge contour representing a workpiece position in the continuous video stream to determine a workpiece area of the continuous video stream; compare pixel brightness changes of the workpiece area in adjacent video frames of the continuous video stream to obtain a contact area of a target object; take a pixel point with significant texture in the contact area as a to-be-tracked feature point of the target object; Matching the local image block of the to-be-tracked feature point with a corresponding image region of a subsequent video frame in the continuous video stream to obtain new position information of the to-be-tracked feature point; Taking the original position of the to-be-tracked feature point as a starting point and the new position information as an ending point, a moving path of the to-be-tracked feature point is constructed, and the moving path is integrated into a pixel motion trajectory of the target object.

3. A tool wear monitoring system based on video of a CNC lathe tool use according to claim 1, wherein, When the texture evolution analysis module performs recording texture structure changes of the contact area to generate a surface texture evolution sequence of the target object, the texture evolution analysis module is specifically configured to: Capture an image block of the contact area in the continuous video stream; Perform grayscale processing on the image block to obtain a single-channel luminance image of the image block; Perform gradient quantization on the pixels in the single-channel luminance image to obtain gradient amplitudes and gradient direction angles of the pixels; According to the gradient direction angles, the pixels are assigned to corresponding preset direction intervals, and the gradient amplitudes of the pixels are accumulated into accumulated values of the assigned direction intervals to construct a gradient direction histogram of the single-channel luminance image; According to the gradient direction histogram, statistical features of texture distribution characteristics in the contact area are derived, and the statistical features are arranged into a texture feature vector of the target object; Based on a time sequence of video frames in the continuous video stream, the texture feature vector is time-series spliced to obtain a surface texture evolution sequence of the target object.

4. A tool wear monitoring system based on video of a CNC lathe tool use according to claim 3, characterized in that, The calculation formula of the gradient amplitude is as follows: ； In the formula, is the gradient amplitude of the pixel point, is the gradient component of the pixel point in the horizontal direction, is the gradient component of the pixel point in the vertical direction, is the norm order used for controlling the fusion mode of the horizontal and vertical gradient components, is a preset contrast enhancement coefficient, is the local region contrast of the pixel point, is a preset anisotropic modulation coefficient, is a very small positive number, is a logarithmic function.

5. A tool wear monitoring system based on video of a lathe tool for use with a numerically controlled lathe as claimed in claim 1, characterized in that, When the visual feature extraction module performs extraction of visual features of a key region of a target object in the key frame sequence as primary wear features of the target object, the visual feature extraction module is specifically configured to: According to a correspondence relationship between the key frame sequence and the continuous video stream, the contact area is mapped to a key frame image in the continuous video stream, and an image block corresponding to the contact area is captured as a sub-image of the key region; Perform multi-scale decomposition on the sub-image to obtain a scale space representation of the sub-image; In the scale space representation, texture features and contour features in the sub-image are extracted; The texture features and the contour features are tensor-synthesized to obtain a visual feature vector of the key region; The visual feature vector is time-series feature aggregated to obtain the primary wear features of the target object.

6. A tool wear monitoring system based on video of a lathe tool for use with a numerically controlled lathe as claimed in claim 1, characterized in that, When the time-series change rule analysis module performs analysis of a time-series change rule of a feature dimension in the primary wear features on the key frame sequence to obtain a time-series change curve of the feature dimension, the time-series change rule analysis module is specifically configured to: Taking a time axis of the key frame sequence as a time-series horizontal coordinate and a value of a same feature dimension in the primary wear features as a vertical coordinate, an original time-series sequence of the feature dimension is constructed; Performing trend decomposition on the original time-series sequence to obtain a low-frequency component of the original time-series sequence; Fitting a linear trend of the low-frequency component on the time-series horizontal coordinate, and taking a direction of the linear trend as an overall change direction of the low-frequency component; Projecting data points in the low-frequency component to the overall change direction to obtain a standard time-series signal of the low-frequency component; and Smoothly interpolate the standard time sequence signal to obtain the time sequence variation curve of the feature dimension.

7. A tool wear monitoring system based on video of a CNC lathe tool use according to claim 6, characterized in that, When performing the projecting of the data points in the low-frequency component to the overall variation direction to obtain the standard time sequence signal of the low-frequency component, the time sequence variation rule analysis module is specifically configured to: construct a projection direction vector of the overall variation direction according to the overall variation direction; identify an abnormal point deviating from the overall variation direction in the low-frequency component; assign a projection confidence weight to the abnormal point according to a deviation degree of the abnormal point; perform weighted correction on a projection value of the abnormal point by applying the projection confidence weight, and replace an original value of the abnormal point with a correction result to obtain a corrected low-frequency component sequence of the low-frequency component; project the data points in the corrected low-frequency component sequence along the projection direction vector to obtain the standard time sequence signal of the low-frequency component.

8. A tool wear monitoring system based on video of a lathe tool for use with a numerically controlled lathe as claimed in claim 1, characterized by, When performing the shape comparison of the time sequence variation curve with a preset reference curve, and filtering out a feature dimension with a shape consistent with the reference curve according to a comparison result to obtain the refined wear feature of the target object, the feature filtering module is specifically configured to: perform time sequence registration of the time sequence variation curve with the preset reference curve to obtain an aligned time sequence variation curve of the time sequence variation curve; select a local window along a time axis on the aligned time sequence variation curve; calculate a local shape similarity between the aligned time sequence variation curve and the reference curve in the local window to obtain a local similarity value of the target object, wherein a calculation formula of the local similarity value is as follows: ； In the formula, The local similarity value is... The preset amplitude error weighting coefficient, The number of sampling points within the local window. The aligned timing variation curve within the local window is the [missing information]. The value of each sampling point The reference curve within the local window is the first... The value of each sampling point These are the preset gradient error weighting coefficients. It is an exponential function; when the local similarity value is higher than a preset consistency threshold, identify a feature dimension corresponding to the aligned time sequence variation curve to obtain a candidate feature dimension of the target object; aggregate the candidate feature dimension into the refined wear feature of the target object.

9. A tool wear monitoring system based on video of a lathe tool for use with a numerically controlled lathe as claimed in claim 1, characterized by, When performing the generating of a wear process determination result of the target object according to a time sequence evolution trend of the refined wear feature along the key frame sequence, the wear determination module is specifically configured to: organize the refined wear feature into a wear feature time sequence evolution matrix according to a time sequence of the key frame sequence; perform segment-by-segment shape matching of the wear feature time sequence evolution matrix with a historical wear stage reference mode of the target object to obtain a shape matching mapping relationship of the target object; identify a key stage demarcation point in the wear feature time sequence evolution matrix corresponding to the historical wear stage reference mode based on the shape matching mapping relationship; divide the key frame sequence into continuous wear stages according to the key stage demarcation point; degree calibrate the wear stages to obtain stage identifiers of the wear stages; generate the wear process determination result of the target object by comprehensively considering an evolution order of the stage identifiers and a division result of the wear stages.

Images