Online tool wear detection method based on machine vision

Abstract

The invention relates to the technical field of tool detection, in particular to a tool wear online detection method based on machine vision, which comprises the following steps of: 1, respectively acquiring an unworn image data set and a worn image data set; step 2, carrying out single-view depth-of-field fusion; step 3, splicing to form a multi-view panoramic depth cutter image; 4, extracting feature points on the surface of the object; space coordinate calculation is carried out, and sparse point cloud models of the unworn cutter and the worn cutter are obtained respectively; 5, dense reconstruction is carried out, and a three-dimensional model of the unworn cutter and a three-dimensional model of the worn cutter are constructed respectively; 6, performing visual subtraction on the point cloud data to obtain a point cloud model of the tool wear area; and 7, performing qualitative and quantitative analysis according to the point cloud model of the tool wear area to obtain the actual wear loss and the state of the tool. According to the invention, qualitative and quantitative analysis can be carried out on tool wear, the quality or machining precision of a machined product is prevented from being influenced, and the production cost is reduced.

Description

Technical Field

[0001] The invention relates to the technical field of tool detection, and in particular to an online tool wear detection method based on machine vision. Background Art

[0002] In modern metal cutting, the tool will inevitably wear and change its appearance, shape and cutting performance. For example, the cutting edge and blade will become blunt due to wear, and the cutting contact between the tool and the cut metal and the harsh processing environment can easily cause wear and even breakage of the tool. Therefore, tool wear is an unavoidable problem, so the measurement and monitoring of tool wear is of great significance to modern metal processing. The measurement and monitoring of tool wear can monitor the tool status in real time to judge the tool status to ensure the quality of the processed product or avoid problems such as reduced processing accuracy of the product caused by tool wear. With the continuous changes in modern manufacturing and the continuous improvement of requirements for mechanical processing automation, more efficient and automatic tool grinding measurement methods and devices are particularly important.

[0003] The main problem with the existing technology is that it is impossible to conduct qualitative and quantitative analysis on the wear of the tool. The wear measurement of the tool in the existing technology can only measure the circumference or area of ​​the wear area, but cannot detect the volume of the tool wear, which affects the subsequent use judgment of the tool, whether to continue to use or replace it. If it continues to be used, it may affect the quality or processing accuracy of the processed product. If it is replaced frequently, it will increase the production cost. Summary of the invention

[0004] The purpose of the present invention is to provide an online tool wear detection method based on machine vision, which can perform qualitative and quantitative analysis on tool wear, avoid affecting the quality or processing accuracy of processed products, and reduce production costs.

[0005] In order to achieve the above object, a method for online detection of tool wear based on machine vision is provided, comprising the following steps:

[0006] Step 1: Collect image information of unworn tools and worn tools of the same type respectively, and collect several tool images with single viewing angle and different depths of field at each set angle of tool rotation to obtain unworn image datasets and worn image datasets;

[0007] Step 2: Perform single-perspective depth-of-field fusion on the tool images in the unworn image dataset and the worn image dataset, and fuse the tool images with different depths of field at the same angle to obtain single-perspective full-depth tool images of the unworn tool and the worn tool respectively;

[0008] Step 3: stitching the single-view full-depth tool images at different viewing angles in the unworn image dataset and the worn image dataset into multi-view full-depth tool images by image stitching;

[0009] Step 4: Extract the surface feature points of the object from the multi-view full-depth tool image; calculate the spatial coordinates based on the extracted feature points, and calculate based on the spatial coordinates of the feature points to obtain the sparse point cloud models of the unworn tool and the worn tool respectively;

[0010] Step 5: Perform dense reconstruction on the sparse point cloud model to construct three-dimensional models of the unworn tool and the worn tool respectively;

[0011] Step 6: Obtain a point cloud model of the tool wear area by visually subtracting the point cloud data of the two three-dimensional models;

[0012] Step 7: Perform qualitative and quantitative analysis based on the point cloud model of the tool wear area to obtain the actual tool wear amount and tool status.

[0013] Further, the steps of qualitative and quantitative analysis in step 7 are as follows:

[0014] Step 701, measure the distance d between the tool and the lens, and obtain the conversion ratio between the actual tool size H and the imaged tool size h1 through the camera focal length f and the distance d, the formula is as follows:

[0015]

[0016] Step 702, obtaining the spatial size of the point cloud model of the tool wear area through step S6, denoted as ∑v;

[0017] Step 703, calculate the actual wear volume of the target tool under test, and the calculation formula is as follows:

[0018] V=k*∑v

[0019] Among them, V is the actual wear volume of the target tool being measured, and ∑v is the set of spatial point clouds in the tool wear area.

[0020] Further, the method further comprises the following steps:

[0021] Step 8: Use the tool analysis neural network model to analyze the tool status based on the actual tool wear and tool status, and predict the tool replacement time.

[0022] Further, the step 8 specifically includes the following steps:

[0023] Step 801, collect tool data sets and construct a tool analysis neural network model, the tool data sets include tool processing data, wear analysis data and tool status data, the processing data include milling speed, feed per tooth, milling depth, milling width and use environment parameters, the wear analysis data is the actual tool wear calculated in step 7, the tool status data is whether it can be used, if it can be used, then record the tool usage time from the use of the tool corresponding to the actual tool wear to the scrapping; when constructing the tool analysis neural network model, the data of the data sets are normalized and used as the input layer of the tool analysis neural network model, and the remaining tool usage time is used as the output layer of the artificial neural network;

[0024] Step 802: Obtain processing data and wear analysis data of the tool to be tested, and input the processing data and wear analysis data into the trained and optimized tool analysis neural network model to analyze and predict the state of the tool to be tested and the remaining use time of the tool;

[0025] Step 803: Analyze whether to replace the tool according to the state of the tool to be tested and the remaining use time of the tool. If the state is unusable, the remaining use time is 0, and a prompt is given to replace the tool; if the state is usable and the remaining use time exceeds the warning value, there is no need to replace the tool, and a prompt is given to indicate the remaining use time of the tool;

[0026] Step 804: When the machining efficiency of the tool drops significantly and does not meet the required machining speed requirement, the actual wear of the tool is qualitatively and quantitatively analyzed again through steps 1 to 7, and the data are collected and organized into an optimized data set, and the parameters of the established tool analysis neural network model are optimized and adjusted according to the optimized data set;

[0027] Step 805: Load the tool analysis neural network model into the tool wear system to monitor the tool status.

[0028] Furthermore, the tool wear analysis system includes a CCD camera for collecting a tool image of a tool to be tested on a machine tool, a CCD camera support frame for fixing the CCD camera on the machine tool, a cross slide for driving the CCD camera support frame to move horizontally and / or vertically on the machine tool, and a controller, wherein the controller is electrically connected to the CCD camera, the cross slide and the machine tool frame, and the controller loads a tool analysis neural network model

[0029] Furthermore, in step 1, the set angle of the tool for each rotation is 10°, and at least three single-view images with different depths of field are collected, for a total of 36 groups of images, which are collected and organized into an unworn image data set and a worn image data set.

[0030] Furthermore, the number of single-view images with different depths of field collected in step 1 is five.

[0031] Furthermore, the tool analysis neural network model in step 801 adopts a convolutional neural network.

[0032] Furthermore, the step 1 also includes the following steps:

[0033] Step 101: Perform noise reduction processing on tool images in the unworn image dataset and the worn image dataset.

[0034] Furthermore, in step 2, the depth of field fusion of the collected tool images is performed using wavelet transform technology.

[0035] Principles and advantages:

[0036] 1. This solution installs a CCD camera on the machine tool, collects images of the tool before and after wear through the CCD camera, and sends the tool image to the computer image processing system of the controller through the CCD camera to calculate the wear information and status of the tool on the machine tool. The point cloud reconstruction technology is used to build a three-dimensional model of the collected tool image, map and calculate the wear of the tool, and perform wear detection and status monitoring on the tool, realizing non-contact measurement, solving the problem of inconvenient detection and safety of non-detachable machine tool tools, making tool wear detection more convenient and quick, and suitable for the detection of non-detachable tools on traditional machine tools. It also solves the problem of the inability to accurately and quantitatively measure tool wear in traditional plane tool detection, thereby reducing the workload of manual operation, further ensuring the safety of the mechanical processing process, and greatly improving the efficiency and accuracy of tool wear detection. In addition, the status of the tool can be obtained to better control the efficiency of the processing.

[0037] 2. This solution constructs a tool analysis neural network model, which can further analyze and predict the status of the tool to be tested and the remaining usage time of the tool according to the actual wear of the tool; when the status is unusable, the remaining usage time is 0, and the tool is prompted to be replaced; when the status is usable and the remaining usage time exceeds the warning value, there is no need to replace the tool, and the remaining usage time of the tool is prompted. In addition, the tool wear analysis system can be used to continuously collect tool data and collect and organize them into optimized data sets to optimize the tool analysis neural network model. The tool analysis neural network model that meets the optimization requirements is installed in the computer image processing system of the controller to monitor the tool status in real time, so as to fully utilize the tool and reduce the tool cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flowchart of a method for online tool wear detection based on machine vision according to an embodiment of the present invention;

[0039] Figure 2 It is the structural diagram of the tool wear analysis system;

[0040] Figure 3 Schematic diagram of quantitative analysis of tool wear. DETAILED DESCRIPTION

[0041] The following is further described in detail through specific implementation methods:

[0042] The reference numerals in the drawings of the specification include: machine tool 1, transverse sliding guide rail 2, CCD camera support frame 3, tool to be measured 4, CCD camera 5, longitudinal sliding guide rail 6, connection node 7, signal transmission cable 8, controller 9.

[0043] Example

[0044] A machine vision-based online tool wear detection method is basically as follows Figure 1 As shown, the following steps are included:

[0045] Step 1: The tool wear analysis system is used to collect image information of unworn tools and worn tools of the same type. A number of tool images with single viewing angles and different depths of field are collected at each set angle of tool rotation to obtain an unworn image dataset and a worn image dataset corresponding to the unworn tool.

[0046] like Figure 2 As shown, the tool wear analysis system includes a CCD camera for collecting tool images of the tool to be tested on the machine tool, a CCD camera support frame for fixing the CCD camera on the machine tool, a cross slide for driving the CCD camera support frame to move horizontally and / or vertically on the machine tool, and a controller, the controller is electrically connected to the CCD camera, the cross slide and the machine tool frame, and the controller is loaded with a tool analysis neural network model. The cross slide includes a horizontal sliding guide rail and a vertical sliding guide rail, both of which are a combination of a traditional lead screw and a stepper motor.

[0047] In step 1, the set angle of the tool for each rotation is 10°, and at least three single-view images with different depths of field are collected (in this embodiment, 5 tool images at different depths of field are collected), with a total of 36 groups of images, which are collected and organized into unworn image data sets and worn image data sets.

[0048] The step 1 also includes the following steps:

[0049] Step 101: Perform noise reduction processing on the tool images in the unworn image dataset and the worn image dataset to improve the quality of the tool images and the accuracy of the wear amount.

[0050] Step 2: Perform single-perspective depth-of-field fusion on the tool images in the unworn image dataset and the worn image dataset, and fuse the tool images with different depths of field at the same angle to obtain single-perspective full-depth tool images of the unworn tool and the worn tool respectively; In this embodiment, the depth-of-field fusion of the collected tool images is performed based on wavelet transform technology, and the tool images with different depths of field at the same angle are fused to obtain a single-perspective full-depth tool image;

[0051] Step 3: stitching the single-view full-depth tool images at different viewing angles in the unworn image dataset and the worn image dataset into multi-view full-depth tool images by image stitching; in this embodiment, the image stitching technology based on feature point extraction is used to stitch 36 groups of tool images at different viewing angles into a multi-view full-depth tool image;

[0052] Step 4: Extract surface feature points of the multi-view full-depth tool image; calculate the spatial coordinates based on the extracted feature points, and calculate based on the spatial coordinates of the feature points to obtain sparse point cloud models of the unworn tool and the worn tool respectively; in this embodiment, the SIFT algorithm is used to extract surface feature points of the multi-view full-depth tool image; then the SFM algorithm is used to calibrate the CCD camera parameters, and the spatial coordinates of the feature points are calculated through the CCD camera parameters.

[0053] Step 5: densely reconstruct the sparse point cloud model to construct three-dimensional models of the unworn tool and the worn tool respectively; in this embodiment, the sparse point cloud is densely reconstructed by using the CMVS and PMVS algorithms to construct a complete three-dimensional model of the tool.

[0054] Step 6: Obtain a point cloud model of the tool wear area by visually subtracting the point cloud data of the two three-dimensional models;

[0055] Step 7: Perform qualitative and quantitative analysis based on the point cloud model of the tool wear area to obtain the actual tool wear amount and tool status.

[0056] The steps of qualitative and quantitative analysis in step 7 are as follows:

[0057] Step 701: Figure 3 As shown, the distance d between the tool and the lens is measured, and the conversion ratio between the actual tool size H and the imaged tool size h1 is obtained through the camera focal length f and the distance d. The formula is as follows:

[0058]

[0059] Step 702: Obtain the spatial size of the point cloud model of the tool wear area through step S6, denoted as Σv.

[0060] Step 703, calculate the actual wear volume of the target tool under test, and the calculation formula is as follows:

[0061] V=k*Σv

[0062] Among them, V is the actual wear volume of the target tool under test, and Σv is the set of spatial point clouds in the tool wear area.

[0063] Step 8: Use the tool analysis neural network model to analyze the tool status based on the actual tool wear and tool status, and predict the tool replacement time.

[0064] The step 8 specifically includes the following steps:

[0065] Step 801, collect tool data sets and construct a tool analysis neural network model, the tool data sets include tool processing data, wear analysis data and tool status data, the processing data include milling speed, feed per tooth, milling depth, milling width and use environment parameters, the wear analysis data is the actual tool wear calculated in step 7, the tool status data is whether it can be used, if it can be used, then the tool usage time from the tool corresponding to the actual tool wear to the time it is scrapped is recorded; if it cannot be used, then the tool usage time from the tool corresponding to the actual tool wear to the time it is scrapped is 0, and no recording is required; when constructing the tool analysis neural network model, the data of the data set is normalized and used as the input layer of the tool analysis neural network model, and the remaining tool usage time is used as the output layer of the artificial neural network;

[0066] Step 802: Obtain processing data and wear analysis data of the tool to be tested, and input the processing data and wear analysis data into the trained and optimized tool analysis neural network model to analyze and predict the state of the tool to be tested and the remaining use time of the tool;

[0067] Step 803: Analyze whether to replace the tool based on the status of the tool to be tested and the remaining usage time of the tool. If the status is unusable, the remaining usage time is 0, and a prompt is given to replace the tool. If the status is usable and the remaining usage time exceeds the warning value, there is no need to replace the tool, and a prompt is given for the remaining usage time of the tool. The warning value refers to the threshold that satisfies the minimum processing time, because changing the tool midway may affect the processing accuracy and quality. Therefore, tools with a remaining usage time less than the warning value need to be replaced and reserved for processing scenarios with corresponding warning values ​​for full use.

[0068] Step 804: When the machining efficiency of the tool drops significantly and does not meet the required machining speed requirement, the actual wear of the tool is qualitatively and quantitatively analyzed again through steps 1 to 7, and the evaluation of the remaining use time of the tool by technical experts is collected and organized into an optimized data set, and the parameters of the established tool analysis neural network model are optimized and adjusted according to the optimized data set; whether the machining efficiency of the tool drops significantly or not, and whether the required machining speed is met or not, data collection is performed through random sampling, and specifically the machining data of the tool is collected.

[0069] Step 805: Load the tool analysis neural network model into the tool wear system to monitor the tool status. The status monitoring includes timed status monitoring and real-time status monitoring.

[0070] 1. This solution realizes non-contact measurement, solves the problem of inconvenient detection and safety of non-detachable machine tool tools, makes tool wear detection more convenient and quick, and is suitable for the detection of non-detachable tools on traditional machine tools. It also solves the problem that the tool wear cannot be accurately and quantitatively measured in traditional plane tool detection, thereby reducing the workload of manual operation, further ensuring the safety of the mechanical processing process, and greatly improving the efficiency and accuracy of tool wear detection. In addition, the status of the tool can be obtained to better control the efficiency of the processing.

[0071] 2. This solution constructs a tool analysis neural network model, which can further analyze and predict the status of the tool to be tested and the remaining service life of the tool based on the actual wear of the tool; and can continuously collect data to optimize the tool analysis neural network model during use, and then install the tool analysis neural network model that meets the optimization requirements into the computer image processing system of the controller to monitor the tool status in real time, so as to achieve full utilization of the tool and greatly reduce the tool cost.

[0072] The above is only an embodiment of the present invention. The common sense such as the known specific structure and characteristics in the scheme is not described in detail here. The ordinary technicians in the relevant field know all the common technical knowledge in the technical field of the invention before the application date or priority date, can obtain all the existing technologies in the field, and have the ability to apply the conventional experimental means before that date. The ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several deformations and improvements can be made, which should also be regarded as the protection scope of the present invention, which will not affect the effect of the implementation of the present invention and the practicality of the patent. The protection scope required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A machine vision-based online tool wear detection method, characterized in that: The following steps are involved: Step 1: Collect image information of unworn tools and worn tools of the same type respectively, and collect several tool images with single viewing angle and different depths of field at each set angle of tool rotation to obtain unworn image datasets and worn image datasets; Step 2: Perform single-perspective depth-of-field fusion on the tool images in the unworn image dataset and the worn image dataset, and fuse the tool images with different depths of field at the same angle to obtain single-perspective full-depth tool images of the unworn tool and the worn tool respectively; Step 3: stitching the single-view full-depth tool images at different viewing angles in the unworn image dataset and the worn image dataset into multi-view full-depth tool images by image stitching; Step 4: Extract the surface feature points of the object from the multi-view full-depth tool image; calculate the spatial coordinates based on the extracted feature points, and calculate based on the spatial coordinates of the feature points to obtain the sparse point cloud models of the unworn tool and the worn tool respectively; Step 5: Perform dense reconstruction on the sparse point cloud model to construct three-dimensional models of the unworn tool and the worn tool respectively; Step 6: Obtain a point cloud model of the tool wear area by visually subtracting the point cloud data of the two three-dimensional models; Step 7: Perform qualitative and quantitative analysis based on the point cloud model of the tool wear area to obtain the actual tool wear amount and tool status.

2. The online tool wear detection method based on machine vision according to claim 1, characterized in that: The steps of qualitative and quantitative analysis in step 7 are as follows: Step 701, measure the distance d between the tool and the lens, and obtain the conversion ratio between the actual tool size H and the imaged tool size h1 through the camera focal length f and the distance d, the formula is as follows: Step 702, obtaining the spatial size of the point cloud model of the tool wear area through step S6, denoted as Σv; Step 703, calculate the actual wear volume of the target tool under test, and the calculation formula is as follows: v=k*∑v Among them, V is the actual wear volume of the target tool being measured, and ∑v is the set of spatial point clouds in the tool wear area.

3. The online tool wear detection method based on machine vision according to claim 2 is characterized in that: The following steps are also included: Step 8: Use the tool analysis neural network model to analyze the tool status based on the actual tool wear and tool status, and predict the tool replacement time.

4. The method for online tool wear detection based on machine vision according to claim 3, characterized in that: The step 8 specifically comprises the following steps: Step 801, collect tool data sets and construct a tool analysis neural network model, the tool data sets include tool processing data, wear analysis data and tool status data, the processing data include milling speed, feed per tooth, milling depth, milling width and use environment parameters, the wear analysis data is the actual tool wear calculated in step 7, the tool status data is whether it can be used, if it can be used, then record the tool usage time from the use of the tool corresponding to the actual tool wear to the scrapping; when constructing the tool analysis neural network model, the data of the data sets are normalized and used as the input layer of the tool analysis neural network model, and the remaining tool usage time is used as the output layer of the artificial neural network; Step 802: Obtain processing data and wear analysis data of the tool to be tested, and input the processing data and wear analysis data into the trained and optimized tool analysis neural network model to analyze and predict the state of the tool to be tested and the remaining use time of the tool; Step 803: Analyze whether to replace the tool according to the state of the tool to be tested and the remaining use time of the tool. If the state is unusable, the remaining use time is 0, and a prompt is given to replace the tool; if the state is usable and the remaining use time exceeds the warning value, there is no need to replace the tool, and a prompt is given to indicate the remaining use time of the tool; Step 804: When the machining efficiency of the tool drops significantly and does not meet the required machining speed requirement, the actual wear of the tool is qualitatively and quantitatively analyzed again through steps 1 to 7, and the data are collected and organized into an optimized data set, and the parameters of the established tool analysis neural network model are optimized and adjusted according to the optimized data set; Step 805: Load the tool analysis neural network model into the tool wear system to monitor the tool status.

5. The method for online tool wear detection based on machine vision according to claim 4, characterized in that: The tool wear analysis system includes a CCD camera for collecting tool images of a tool to be tested on a machine tool, a CCD camera support frame for fixing the CCD camera on the machine tool, a cross slide for driving the CCD camera support frame to move horizontally and / or vertically on the machine tool, and a controller, wherein the controller is electrically connected to the CCD camera, the cross slide and the machine tool frame, and the controller loads a tool analysis neural network model.

6. The online tool wear detection method based on machine vision according to claim 1, characterized in that: In the step 1, the set angle of the tool for each rotation is 10°, and at least three single-view images with different depths of field are collected, for a total of 36 groups of images, which are collected and organized into an unworn image data set and a worn image data set.

7. The method for online tool wear detection based on machine vision according to claim 6, characterized in that: In step 1, five single-view images with different depths of field are collected.

8. The method for online tool wear detection based on machine vision according to claim 1, characterized in that: In step 801, the tool analysis neural network model adopts a convolutional neural network.

9. The online tool wear detection method based on machine vision according to claim 1, characterized in that: The step 1 also includes the following steps: Step 101: Perform noise reduction processing on tool images in the unworn image dataset and the worn image dataset.

10. The online tool wear detection method based on machine vision according to claim 1, characterized in that: In the step 2, the depth of field fusion of the collected tool images is performed using wavelet transform technology.

Images