Adaptive adjustment system and method based on automatic burr removal equipment

CN118559940BActive Publication Date: 2024-11-05JIANGSU ZHONGKE CLOUD CONTROL INTELLIGENT IND EQUIP CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410609229.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-16
Publication Date
2024-11-05
Estimated Expiration
2044-05-16

AI Technical Summary

Technical Problem

Traditional burr removal methods rely on manual intervention, are inefficient and unstable, have safety hazards, and are difficult to ensure product quality and consistency.

Method used

Design an adaptive adjustment system based on automatic burr removal equipment, including visual recognition module, deep learning module and motor control module. It automatically detects and locates burrs using industrial cameras and deep learning models, and realizes automatic removal through motor control modules, combined with safety protection The module monitors current and temperature to ensure safe operation of the equipment.

Benefits of technology

The burr removal process is fully automated, which improves production efficiency and removal efficiency, enhances detection accuracy and safety, and reduces the need for manual intervention and resource waste.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118559940B_ABST
    Figure CN118559940B_ABST
Patent Text Reader

Abstract

The present invention discloses an adaptive adjustment system and method based on an automatic burr removal device, and belongs to the field of artificial intelligence technology. The present invention includes: a visual recognition module: using an industrial camera to aim at the burrs on the surface of a plastic product, continuously capturing images, performing preprocessing, and analyzing and extracting features from the preprocessed images; a deep learning module: collecting a data set containing samples with and without burrs, training a deep learning model, and the model will learn to extract features from the input image; a motor control module: controlling the speed and direction of the motor through a motor driver, the motor driver receives the specific information of the burrs and converts it into a signal that the motor can recognize; designing and implementing a motor control algorithm; a safety protection module: when an emergency occurs, the administrator can stop the motor by one-click stop; monitoring the current of the motor, monitoring the temperature of the motor, electronic components and the device itself, and executing corresponding measures when an abnormality occurs.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a self-adaptive adjustment system and method based on automatic burr removal equipment. Background Art

[0002] In the production process of plastic products, burrs often occur, which will reduce the quality of the product and affect the appearance and performance of the product. Traditionally, removing these burrs usually requires manual intervention, which is time-consuming, labor-intensive and inefficient. In order to improve efficiency and reduce costs, automated burr removal equipment has come into being. The Sobel operator is a commonly used edge detection operator that uses convolution operations to filter images to detect edges in images. In image processing, convolution operations are used to filter and extract features from images. The convolution operation multiplies a convolution kernel with each pixel of the image one by one and sums the results to obtain a filtered image. The Gaussian pyramid is an image pyramid generated by multiple downsampling and Gaussian filtering operations on the original image. The downsampling operation can reduce the resolution of the image, while the Gaussian filter can smooth the image and reduce the high-frequency details of the image.

[0003] Due to the intervention of human factors, traditional burr removal methods are easily affected by environmental factors, operator skill level and other factors, resulting in unstable removal effects and difficulty in ensuring product quality and consistency. In addition, traditional burr removal methods also have safety hazards. Operators need to approach running mechanical equipment to operate, which poses a risk of accidental injury. Summary of the invention

[0004] The object of the present invention is to provide an adaptive adjustment system and method based on an automatic burr removal device to solve the problems raised in the above background technology.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] The adaptive adjustment system based on the automatic burr removal equipment includes:

[0007] Visual recognition module: Use an industrial camera to aim at the burrs on the surface of the plastic product and continuously capture images within the working distance of the device; pre-process the captured images, including denoising, smoothing and contrast enhancement; use image processing technology to analyze and extract features from the pre-processed images to identify and locate the burrs;

[0008] Deep learning module: collect a data set containing samples with and without burrs, the data set contains images of plastic products of different types and sizes, and corresponding burr labels, the burr labels are used to mark the attributes of burrs; use the data set to train the deep learning model, during the training process the model will learn to extract features from the input image, and output when the parameters of the equipment are in a certain situation, burrs will be generated, when there are burrs, predict the location of the burrs and output the specific information of the burrs;

[0009] Motor control module: The motor is responsible for controlling the movement of the burr removal tool to remove burrs from the surface of plastic products; the motor speed and direction are controlled by the motor driver, which receives the specific information of the burrs sent from the deep learning module and converts it into a signal that the motor can recognize; the motor control algorithm is designed and implemented, and the control algorithm includes position control and speed control;

[0010] Safety protection module: When an emergency occurs, the administrator can stop the motor with one click; monitor the motor current, stop the motor when an overload is detected, and sound an alarm; monitor the temperature of the motor, electronic components, and the device itself, and stop the motor when the temperature exceeds the safe range to prevent overheating damage.

[0011] The visual recognition module uses the edge detection algorithm Sobel algorithm to detect the edge of the burr in the image and extract the shape and contour information of the burr; uses morphological processing technology, including corrosion, expansion, opening operation and closing operation, to perform morphological operations on the image to emphasize the structural characteristics of the burr, which is used to remove noise in the image, fill holes, and connect broken burrs; uses regional segmentation technology to segment the image into regions with similar features for locating the burr;

[0012] The gray-level co-occurrence matrix is ​​used to extract burr-related features from the preprocessed image, including gray-level features, texture features, directional features, edge features, geometric features and local features, to further identify and locate the burrs in the preprocessed image; the extracted features are matched with the predefined burr features using a feature matching algorithm to obtain the identified and located burrs; the identified and located burrs are post-processed to filter out areas of inappropriate size and remove misdetected burrs.

[0013] The visual recognition module detects burr edges in the image through the Sobel algorithm, and calculates the gradient value of each pixel point by applying the Sobel operator to the image for convolution operation to find the burr edges in the image; the convolution kernel size of the Sobel operator is automatically selected according to the size and density of the burr, and the calculation is performed using a Gaussian pyramid;

[0014] The Sobel algorithm is a commonly used edge detection algorithm that uses convolution operations to calculate the gradient values ​​of pixels in an image to find the edges in the image. Its advantages include: Simple and easy to implement: The implementation of the Sobel algorithm is relatively simple, with a small amount of calculation, and is suitable for scenarios with high real-time requirements. Adjustable parameters: The convolution kernel size of the Sobel operator can be adjusted according to specific needs to adapt to glitches of different sizes and densities. The Gaussian pyramid is an image pyramid that is constructed through multiple downsampling and Gaussian filtering and is used for image scale space analysis and feature extraction. Its advantages include: Scale invariance: The Gaussian pyramid can analyze images at different scales to adapt to glitches of different sizes. Smoothness: The Gaussian pyramid can smooth images and reduce high-frequency details of images, which is beneficial for edge detection and feature extraction.

[0015] The Gaussian pyramid obtains a series of images by performing several downsampling and Gaussian filtering on the original image. Each level of image is more blurred and smaller than the previous level. Downsampling is achieved by downsampling the image to reduce the number of pixels in the image by a certain ratio. Gaussian filtering smoothes the image and reduces the high-frequency details of the image by performing convolution operations on the image. In the process of constructing the Gaussian pyramid, as the image is downsampled, the resolution of the image is reduced, and the size of the burr is also reduced accordingly.

[0016] In the process of constructing the Gaussian pyramid, the original image is downsampled several times. The downsampling operation is achieved by sampling the image in proportion. Let the original image be I, the downsampled image be I', and the sampling ratio be r. The downsampling operation is expressed as:

[0017] I'(x,y)=I(rx,ry)

[0018] Where (x, y) is the pixel coordinate of the downsampled image I', and (rx, ry) is the corresponding pixel coordinate of the original image I;

[0019] The Gaussian filter operation is implemented by convolving the image with a Gaussian kernel. The image after Gaussian filtering is I", the coordinates of the Gaussian kernel are (i, j), and the Gaussian kernel is G(i, j). The Gaussian filter operation is expressed as:

[0020]

[0021] Where (x', y') is the pixel coordinate of the image I' after Gaussian filtering, and k is the size of the Gaussian kernel;

[0022] At different levels of the Gaussian pyramid, the size and density of the burrs in the image are analyzed, the area and perimeter characteristics of the burrs are calculated, and the number of burrs is counted to evaluate the size and density of the burrs; according to the estimated size and density of the burrs at different levels in the Gaussian pyramid, the appropriate convolution kernel size of the Sobel operator is determined; when the burrs are larger or the density is higher, a larger convolution kernel size is selected; when the burrs are smaller or the density is lower, a smaller convolution kernel size is selected.

[0023] After the horizontal and vertical convolution operations of the Sobel operator, the horizontal gradient G x and the vertical gradient G y , the convolution kernel of the Sobel operator is used to calculate the gradient value of each pixel. The mathematical formula is as follows: G x =I*K x , G y =I*K y , * indicates convolution operation, K x is the horizontal convolution kernel of the Sobel operator, K y It is the vertical convolution kernel of the Sobel operator;

[0024] The convolution kernel size k*k dynamically selected by the Gaussian pyramid can ensure that the Sobel operator detects the burr edges in the image according to the estimated size and density of the burrs.

[0025] In the deep learning module, an image data set containing burrs and non-burrs is prepared, the image data is annotated to indicate which areas are burrs and which areas are not burrs, the image is input into the visual recognition module for acquisition, and a convolutional neural network is selected for training based on the image data set;

[0026] Get an image dataset D = {(U1, V1), (U2, V2), ..., (U N ,V N )}, where U N represents the Nth image, V N Yes with U N The corresponding annotation information, V N with U N The binary images of are the same size; in D, pixels with a V value of 1 represent burrs, and pixels with a V value of 0 represent non-burrs; the selected convolutional neural network model is f θ (S), where θ represents the parameters of the convolutional neural network model. The training goal is to optimize θ so that f θ Able to correctly predict the location of burrs from the input image S;

[0027] During the training process, the cross entropy loss function is used to measure the difference between the model's prediction results and the labeled information. The formula of the cross entropy loss function is expressed as:

[0028]

[0029] in, Represents the loss of the convolutional neural network model, which is used to measure the difference between the model prediction result and the true label; H×W represents the number of pixels in each image, which is used to calculate the loss of each sample in the loss function; V qe Represents image U q The label value of the e-th pixel in is a binary label. A value of 1 indicates that the pixel belongs to the burr area, and a value of 0 indicates that the pixel does not belong to the burr area. θ (U q ) e represents f θ The predicted probability that each pixel in the image belongs to the burr area.

[0030] The deep learning module extracts specific information of the burr, including burr position, burr shape, burr size and burr density, according to the result of the convolutional neural network analysis, and the specific information of the burr will be sent to the motor driver; the motor control module enables the motor driver to convert the specific information of the burr into a control signal that can be recognized by the motor after receiving the specific information of the burr from the deep learning module;

[0031] According to the specific information of the burr, the motor driver adjusts the speed of the motor to control the burr removal tool to move at a specific speed during operation; if the burrs are densely distributed or the burrs are large, the motor speed needs to be increased to increase the removal efficiency; conversely, if the burrs are sparsely distributed or the burrs are small, the motor speed needs to be reduced for removal operations; the motor driver is responsible for controlling the movement direction of the motor. According to the location and layout of the burrs, the motor driver will control the motor to rotate forward, reverse or stop to ensure that the burr removal tool can move in the predetermined path and direction and remove the burrs.

[0032] The core of the convolutional neural network (CNN) is the convolution layer, which contains a series of convolution kernels. Each convolution kernel is a small window that slides over the input image and performs convolution operations with each local area of ​​the image. The convolution operation can extract local features such as edges, textures, etc. The pooling layer usually follows the convolution layer to reduce the spatial size of the feature map and reduce the number of parameters. The pooling operation includes maximum pooling and average pooling, which take the maximum value or average value in the pooling window as the output.

[0033] After each convolutional layer or pooling layer, a nonlinear activation function ReLU is applied. The activation function introduces nonlinearity and increases the expressiveness of the model. The last few layers of CNN are usually fully connected layers, which flatten the outputs of the convolutional and pooling layers into one-dimensional vectors and connect them to one or more fully connected layers. The fully connected layer is responsible for associating high-level features with specific categories.

[0034] The motor control module first determines the target position of the burr removal tool in the position control algorithm, which is obtained by the deep learning module or determined by the operator's input, and the target position is expressed as an absolute coordinate or an offset relative to the current position; when the target position is determined, the position control algorithm calculates the error between the current position and the target position, which is obtained by subtracting the current position from the target position, and is expressed as a distance or angle difference;

[0035] Position control uses a PID controller to adjust the movement of the motor to minimize the position error. The PID controller generates a control signal based on the size and rate of change of the position error to adjust the speed and direction of the motor. In the PID controller, the proportional term P is used to generate a control signal based on the size of the position error, the integral term I is used to eliminate steady-state errors, and the differential term D is used to suppress changes in the error.

[0036] The generated PID control signal is sent to the motor driver to adjust the input voltage or current of the motor, thereby controlling the speed and direction of the motor; the motor driver controls the movement of the motor according to the received control signal, so that the motor moves toward the target position; during the operation of the motor, the position control algorithm continuously monitors the position error and adjusts the control signal in real time to maintain operation near the target position. It is necessary to adjust the PID parameters or limit the amplitude of the control signal according to the real-time feedback signal.

[0037] The motor control module first determines a target speed of the burr removal tool in a speed control algorithm, wherein the target speed is obtained by the deep learning module or determined by an operator's input; when the target speed is determined, the speed control algorithm calculates an error between the current speed and the target speed, expressed as a difference between the current speed and the target speed;

[0038] Speed ​​control uses a proportional controller to adjust the speed of the motor, and the control signal is proportional to the speed error; the generation of the control signal is determined by the size and rate of change of the speed error; the generated control signal is sent to the motor driver to adjust the input voltage or current to the motor, thereby controlling the speed of the motor; the motor driver adjusts the speed of the motor based on the received control signal; during the operation of the motor, the speed control algorithm continuously monitors the speed error and adjusts the control signal in real time.

[0039] In the safety protection module, the device is equipped with an emergency stop button, and the administrator can immediately stop the rotation of the motor by pressing the emergency stop button; the safety protection module monitors the current of the motor and immediately stops the operation of the motor when a current overload is detected; when an overload is detected, the system automatically stops the operation of the motor and issues an alarm to alert the administrator; the safety protection module monitors the temperature of the motor, electronic components and the device itself, and stops the operation of the motor when the temperature exceeds a safe range; when the temperature exceeds a set threshold, the system automatically stops the operation of the motor and issues an alarm to alert the administrator; the safety protection module is integrated with the human-machine interface of the device, and the alarm alerts the administrator through sound, light signals and display screens; after a security incident occurs, the device implements an automatic reset function, and can restart the device and resume normal operation after the problem is solved.

[0040] The steps of the adaptive adjustment method based on the burr automatic removal equipment are as follows:

[0041] S100, use an industrial camera to focus on the burrs on the surface of plastic products and continuously capture images within the working distance of the device;

[0042] S200, preprocessing the captured image, and using image processing technology to analyze and extract features from the preprocessed image to identify and locate burrs;

[0043] S300, collecting a data set including samples with burrs and samples without burrs, and using the data set to train a deep learning model;

[0044] S400, during the training process, the model will learn to extract features from the input image and output whether there is a burr or no burr, and when there is a burr, the specific information of the burr will be output;

[0045] S500, controlling the speed and direction of the motor through the motor driver, the motor driver receiving the specific information of the burr sent from the deep learning module and converting it into a signal that the motor can recognize;

[0046] S600, designing and implementing a motor control algorithm, wherein the control algorithm includes position control and speed control;

[0047] S700, when an emergency occurs, the administrator can stop the motor with one click;

[0048] S800, monitors the current of the motor, stops the operation of the motor when an overload is detected, and issues an alarm; monitors the temperature of the motor, electronic components, and the device itself, and stops the operation of the motor when the temperature exceeds the safe range to prevent overheating damage.

[0049] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0050] The present invention utilizes a visual recognition module and a deep learning module to realize automatic detection and positioning of burrs, making the burr removal process fully automated and eliminating the need for manual intervention, thereby greatly improving production efficiency.

[0051] With the help of a deep learning module, the present invention can accurately identify information such as the location, shape, size, and density of burrs, with higher accuracy than traditional manual inspection or simple image processing methods.

[0052] The present invention is combined with a motor control module, and can adjust the speed and direction of the motor according to the specific information of the burr, thereby optimizing the burr removal process, improving the removal efficiency, and reducing the waste of resources and time.

[0053] The application of the safety protection module can monitor the current and temperature of the motor. Once an abnormal situation is found, the operation of the motor will be stopped in time to ensure the safety of the equipment and operators. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0055] Figure 1 It is a system structure diagram of the adaptive adjustment system based on the automatic burr removal device of the present invention;

[0056] Figure 2 It is a schematic diagram of the steps of the adaptive adjustment method based on the automatic burr removal equipment of the present invention. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] See also Figure 1 and Figure 2 , the present invention provides a technical solution:

[0059] According to one embodiment of the present invention, Figure 1 As shown in the system structure diagram of the adaptive adjustment system based on the automatic burr removal device, the adaptive adjustment system based on the automatic burr removal device includes:

[0060] Visual recognition module: Use an industrial camera to aim at the burrs on the surface of the plastic product and continuously capture images within the working distance of the device; pre-process the captured images, including denoising, smoothing and contrast enhancement; use image processing technology to analyze and extract features from the pre-processed images to identify and locate the burrs;

[0061] Deep learning module: collect a data set containing samples with and without burrs, the data set contains images of plastic products of different types and sizes, and corresponding burr labels, the burr labels are used to mark the attributes of burrs; use the data set to train the deep learning model, during the training process the model will learn to extract features from the input image, and output when the parameters of the equipment are in a certain situation, burrs will be generated, when there are burrs, predict the location of the burrs and output the specific information of the burrs;

[0062] Motor control module: The motor is responsible for controlling the movement of the burr removal tool to remove burrs from the surface of plastic products; the motor speed and direction are controlled by the motor driver, which receives the specific information of the burrs sent from the deep learning module and converts it into a signal that the motor can recognize; the motor control algorithm is designed and implemented, and the control algorithm includes position control and speed control;

[0063] Safety protection module: When an emergency occurs, the administrator can stop the motor with one click; monitor the motor current, stop the motor when an overload is detected, and sound an alarm; monitor the temperature of the motor, electronic components, and the device itself, and stop the motor when the temperature exceeds the safe range to prevent overheating damage.

[0064] The visual recognition module uses the edge detection algorithm Sobel algorithm to detect the edge of the burr in the image and extract the shape and contour information of the burr; uses morphological processing technology, including corrosion, expansion, opening operation and closing operation, to perform morphological operations on the image to emphasize the structural characteristics of the burr, which is used to remove noise in the image, fill holes, and connect broken burrs; uses regional segmentation technology to segment the image into regions with similar features for locating the burr;

[0065] The gray-level co-occurrence matrix is ​​used to extract burr-related features from the preprocessed image, including gray-level features, texture features, directional features, edge features, geometric features and local features, to further identify and locate the burrs in the preprocessed image; the extracted features are matched with the predefined burr features using a feature matching algorithm to obtain the identified and located burrs; the identified and located burrs are post-processed to filter out areas of inappropriate size and remove misdetected burrs.

[0066] The visual recognition module detects burr edges in the image through the Sobel algorithm, and calculates the gradient value of each pixel point by applying the Sobel operator to the image for convolution operation to find the burr edges in the image; the convolution kernel size of the Sobel operator is automatically selected according to the size and density of the burr, and the calculation is performed using a Gaussian pyramid;

[0067] The Gaussian pyramid obtains a series of images by performing several downsampling and Gaussian filtering on the original image. Each level of image is more blurred and smaller than the previous level. Downsampling is achieved by downsampling the image to reduce the number of pixels in the image by a certain ratio. Gaussian filtering smoothes the image and reduces the high-frequency details of the image by performing convolution operations on the image. In the process of constructing the Gaussian pyramid, as the image is downsampled, the resolution of the image is reduced, and the size of the burr is also reduced accordingly.

[0068] In the process of constructing the Gaussian pyramid, the original image is downsampled several times. The downsampling operation is achieved by sampling the image in proportion. Let the original image be I, the downsampled image be I', and the sampling ratio be r. The downsampling operation is expressed as:

[0069] I'(x,y)=I(rx,ry)

[0070] Where (x, y) is the pixel coordinate of the downsampled image I', and (rx, ry) is the corresponding pixel coordinate of the original image I;

[0071] The Gaussian filter operation is implemented by convolving the image with a Gaussian kernel. The image after Gaussian filtering is I", the coordinates of the Gaussian kernel are (i, j), and the Gaussian kernel is G(i, j). The Gaussian filter operation is expressed as:

[0072]

[0073] Where (x', y') is the pixel coordinate of the image I' after Gaussian filtering, and k is the size of the Gaussian kernel;

[0074] At different levels of the Gaussian pyramid, the size and density of the burrs in the image are analyzed, the area and perimeter characteristics of the burrs are calculated, and the number of burrs is counted to evaluate the size and density of the burrs; according to the estimated size and density of the burrs at different levels in the Gaussian pyramid, the appropriate convolution kernel size of the Sobel operator is determined; when the burrs are larger or the density is higher, a larger convolution kernel size is selected; when the burrs are smaller or the density is lower, a smaller convolution kernel size is selected.

[0075] After the horizontal and vertical convolution operations of the Sobel operator, the horizontal gradient G x and the vertical gradient G y, the convolution kernel of the Sobel operator is used to calculate the gradient value of each pixel. The mathematical formula is as follows: G x =I*K x , G y =I*K y , * indicates convolution operation, K x is the horizontal convolution kernel of the Sobel operator, K y It is the vertical convolution kernel of the Sobel operator;

[0076] The convolution kernel size k*k dynamically selected by the Gaussian pyramid can ensure that the Sobel operator detects the burr edges in the image according to the estimated size and density of the burrs.

[0077] In the deep learning module, an image data set containing burrs and non-burrs is prepared, the image data is annotated to indicate which areas are burrs and which areas are not burrs, the image is input into the visual recognition module for acquisition, and a convolutional neural network is selected for training based on the image data set;

[0078] Get an image dataset D = {(U1, V1), (U2, V2), ..., (U N ,V N )}, where U N represents the Nth image, V N Yes with U N The corresponding annotation information, V N with U N The binary images of are the same size; in D, pixels with a V value of 1 represent burrs, and pixels with a V value of 0 represent non-burrs; the selected convolutional neural network model is f θ (S), where θ represents the parameters of the convolutional neural network model. The training goal is to optimize θ so that f θ Able to correctly predict the location of burrs from the input image S;

[0079] During the training process, the cross entropy loss function is used to measure the difference between the model's prediction results and the labeled information. The formula of the cross entropy loss function is expressed as:

[0080]

[0081] in, Represents the loss of the convolutional neural network model, which is used to measure the difference between the model prediction result and the true label; H×W represents the number of pixels in each image, which is used to calculate the loss of each sample in the loss function; V qe Represents image U q The label value of the e-th pixel in is a binary label. A value of 1 indicates that the pixel belongs to the burr area, and a value of 0 indicates that the pixel does not belong to the burr area.θ (U q ) e represents f θ The predicted probability that each pixel in the image belongs to the burr area.

[0082] The deep learning module extracts specific information of the burr, including burr position, burr shape, burr size and burr density, according to the result of the convolutional neural network analysis, and the specific information of the burr will be sent to the motor driver; the motor control module enables the motor driver to convert the specific information of the burr into a control signal that can be recognized by the motor after receiving the specific information of the burr from the deep learning module;

[0083] According to the specific information of the burr, the motor driver adjusts the speed of the motor to control the burr removal tool to move at a specific speed during operation; if the burrs are densely distributed or the burrs are large, the motor speed needs to be increased to increase the removal efficiency; conversely, if the burrs are sparsely distributed or the burrs are small, the motor speed needs to be reduced for removal operations; the motor driver is responsible for controlling the movement direction of the motor. According to the location and layout of the burrs, the motor driver will control the motor to rotate forward, reverse or stop to ensure that the burr removal tool can move in the predetermined path and direction and remove the burrs.

[0084] The core of the convolutional neural network (CNN) is the convolution layer, which contains a series of convolution kernels. Each convolution kernel is a small window that slides over the input image and performs convolution operations with each local area of ​​the image. The convolution operation can extract local features such as edges, textures, etc. The pooling layer usually follows the convolution layer to reduce the spatial size of the feature map and reduce the number of parameters. The pooling operation includes maximum pooling and average pooling, which take the maximum value or average value in the pooling window as the output.

[0085] After each convolutional layer or pooling layer, a nonlinear activation function ReLU is applied. The activation function introduces nonlinearity and increases the expressiveness of the model. The last few layers of CNN are usually fully connected layers, which flatten the outputs of the convolutional and pooling layers into one-dimensional vectors and connect them to one or more fully connected layers. The fully connected layer is responsible for associating high-level features with specific categories.

[0086] The motor control module first determines the target position of the burr removal tool in the position control algorithm, which is obtained by the deep learning module or determined by the operator's input, and the target position is expressed as an absolute coordinate or an offset relative to the current position; when the target position is determined, the position control algorithm calculates the error between the current position and the target position, which is obtained by subtracting the current position from the target position, and is expressed as a distance or angle difference;

[0087] Position control uses a PID controller to adjust the movement of the motor to minimize the position error. The PID controller generates a control signal based on the size and rate of change of the position error to adjust the speed and direction of the motor. In the PID controller, the proportional term P is used to generate a control signal based on the size of the position error, the integral term I is used to eliminate steady-state errors, and the differential term D is used to suppress changes in the error.

[0088] The generated PID control signal is sent to the motor driver to adjust the input voltage or current of the motor, thereby controlling the speed and direction of the motor; the motor driver controls the movement of the motor according to the received control signal, so that the motor moves toward the target position; during the operation of the motor, the position control algorithm continuously monitors the position error and adjusts the control signal in real time to maintain operation near the target position. It is necessary to adjust the PID parameters or limit the amplitude of the control signal according to the real-time feedback signal.

[0089] The motor control module first determines a target speed of the burr removal tool in a speed control algorithm, wherein the target speed is obtained by the deep learning module or determined by an operator's input; when the target speed is determined, the speed control algorithm calculates an error between the current speed and the target speed, expressed as a difference between the current speed and the target speed;

[0090] Speed ​​control uses a proportional controller to adjust the speed of the motor, and the control signal is proportional to the speed error; the generation of the control signal is determined by the size and rate of change of the speed error; the generated control signal is sent to the motor driver to adjust the input voltage or current to the motor, thereby controlling the speed of the motor; the motor driver adjusts the speed of the motor based on the received control signal; during the operation of the motor, the speed control algorithm continuously monitors the speed error and adjusts the control signal in real time.

[0091] In the safety protection module, the device is equipped with an emergency stop button, and the administrator can immediately stop the rotation of the motor by pressing the emergency stop button; the safety protection module monitors the current of the motor and immediately stops the operation of the motor when a current overload is detected; when an overload is detected, the system automatically stops the operation of the motor and issues an alarm to alert the administrator; the safety protection module monitors the temperature of the motor, electronic components and the device itself, and stops the operation of the motor when the temperature exceeds a safe range; when the temperature exceeds a set threshold, the system automatically stops the operation of the motor and issues an alarm to alert the administrator; the safety protection module is integrated with the human-machine interface of the device, and the alarm alerts the administrator through sound, light signals and display screens; after a security incident occurs, the device implements an automatic reset function, and can restart the device and resume normal operation after the problem is solved.

[0092] According to another embodiment of the present invention, Figure 2The steps of the adaptive adjustment method based on the burr automatic removal device are shown in the schematic diagram. The steps are as follows:

[0093] The steps of the adaptive adjustment method based on the burr automatic removal equipment are as follows:

[0094] S100, use an industrial camera to focus on the burrs on the surface of plastic products and continuously capture images within the working distance of the device;

[0095] S200, preprocessing the captured image, and using image processing technology to analyze and extract features from the preprocessed image to identify and locate burrs;

[0096] S300, collecting a data set including samples with burrs and samples without burrs, and using the data set to train a deep learning model;

[0097] S400, during the training process, the model will learn to extract features from the input image and output whether there is a burr or no burr, and when there is a burr, the specific information of the burr will be output;

[0098] S500, controlling the speed and direction of the motor through the motor driver, the motor driver receiving the specific information of the burr sent from the deep learning module and converting it into a signal that the motor can recognize;

[0099] S600, designing and implementing a motor control algorithm, wherein the control algorithm includes position control and speed control;

[0100] S700, when an emergency occurs, the administrator can stop the motor with one click;

[0101] S800, monitors the current of the motor, stops the operation of the motor when an overload is detected, and issues an alarm; monitors the temperature of the motor, electronic components, and the device itself, and stops the operation of the motor when the temperature exceeds the safe range to prevent overheating damage.

[0102] According to step S100, the industrial camera captures images at a fixed frame rate, capturing 10 images per second, and assuming that each image contains 1000x1000 pixels. Some of the data obtained are as follows:

[0103] Image number: 1, time: 0.1 seconds, glitch detection result: glitch exists;

[0104] Image number: 2, time: 0.2 seconds, glitch detection result: glitch exists;

[0105] Image number: 3, time: 0.3 seconds, burr detection result: no burrs;

[0106] Image number: 4, time: 0.4 seconds, burr detection result: no burrs;

[0107] Image number: 5, time: 0.5 seconds, glitch detection result: glitch exists;

[0108] Image number: 6, time: 0.6 seconds, glitch detection result: glitch exists;

[0109] Image number: 7, time: 0.7 seconds, glitch detection result: glitch exists;

[0110] Image number: 8, time: 0.8 seconds, burr detection result: no burrs;

[0111] Image number: 9, time: 0.9 seconds, glitch detection result: glitch exists;

[0112] Image number: 10, time: 1.0 second, glitch detection result: glitch exists.

[0113] According to step S200, the image processing technology is used to analyze and extract features from the pre-processed image, and the results obtained based on part of the data in S100 are as follows:

[0114] Image number: 1, burr preprocessing features: position (100, 200), shape (circle), size (20x20 pixels);

[0115] Image number: 2, burr preprocessing features: position (150,300), shape (ellipse), size (30x15 pixels);

[0116] Image number: 3, features after burr preprocessing: no burrs;

[0117] Image number: 4, features after burr preprocessing: no burrs;

[0118] Image number: 5, burr preprocessing features: position (200, 400), shape (circular), size (25x25 pixels);

[0119] Image number: 6, features after burr preprocessing: position (300, 100), shape (ellipse), size (40x20 pixels);

[0120] Image number: 7, burr preprocessing features: position (400, 300), shape (circular), size (15x15 pixels);

[0121] Image number: 8, features after burr preprocessing: no burrs;

[0122] Image number: 9, burr preprocessing features: position (500, 200), shape (circular), size (30x30 pixels);

[0123] Image number: 10, burr preprocessing features: position (600,500), shape (ellipse), size (35x20 pixels).

[0124] According to step S300, we collect a dataset containing samples with burrs and samples without burrs, each sample is an image, and contains labels about burrs and samples without burrs. The dataset contains 1000 samples, of which 500 samples have burrs and 500 samples do not have burrs. The sample numbers in the dataset are: 1-500, corresponding to images with burrs; the sample numbers in the dataset are: 501-1000, corresponding to images without burrs.

[0125] According to step S400, the model is trained using the prepared data set. During the training process, the model continuously adjusts parameters through the back propagation algorithm to minimize the loss function between the predicted value and the true value. After the training is completed, the model can predict new images. When the input image passes through the model, the model will output a predicted value indicating whether there are burrs in the image. If there are burrs, the model will also output specific information about the burrs, such as location, shape, size, etc.

[0126] According to step S500, based on the burr position information output by the deep learning module, the motor driver will calculate the distance and direction that the motor needs to move. For example, if the burr position is located in the upper left corner of the image, the motor will move to the upper left corner so as to move the removal tool to the burr position. The shape and size information of the burr can also affect the movement mode of the motor. If the burr is large and regular in shape, the motor can choose to use a larger force and speed for the removal operation. On the contrary, if the burr is small or irregular in shape, the motor will use a smaller force and speed to avoid damaging the surrounding plastic surface. The density information of the burr can affect the movement speed and frequency of the motor. If the burr density is high, the motor may move the removal tool at a faster speed to improve the removal efficiency. On the contrary, if the burr density is low, the motor may move at a slower speed to ensure that each burr is effectively removed.

[0127] According to step S800, the current of the motor is monitored in real time while the motor is running. If an abnormal increase in current is detected, it indicates that the motor may have encountered an overload condition, which may be caused by the burr removal tool being stuck or other abnormal conditions. At this time, the system will immediately stop the operation of the motor and issue an alarm to alert the operator. Similarly, temperature sensors are installed on key components of the motor and equipment to monitor their temperature regularly. If the monitored temperature exceeds the set safety threshold, it indicates that the equipment may be at risk of overheating. To prevent equipment damage or safety accidents such as fire, the system will immediately stop the operation of the motor and alert the operator to handle it through an alarm or other means.

[0128] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0129] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. Adaptive adjustment system based on automatic burr removal equipment, Features: include: Visual recognition module: Use an industrial camera to focus on the burrs on the surface of plastic products and continuously capture images within the working distance of the device; Preprocessing the captured image, wherein the preprocessing includes denoising, smoothing and contrast enhancement; analyzing and extracting features from the preprocessed image using image processing technology to identify and locate burrs; The visual recognition module uses the edge detection algorithm Sobel algorithm to detect the edge of the burr in the image and extract the shape and contour information of the burr; uses morphological processing technology, including corrosion, expansion, opening operation and closing operation, to perform morphological operations on the image to emphasize the structural characteristics of the burr, which is used to remove noise in the image, fill holes, and connect broken burrs; uses regional segmentation technology to segment the image into regions with similar features for locating the burr; Using the gray level co-occurrence matrix from the preprocessed image, the features related to the burrs are extracted, including gray level features, texture features, direction features, edge features, geometric features and local features, so as to further identify and locate the burrs in the preprocessed image; Use feature matching algorithm to match the extracted features with predefined burr features to obtain identified and located burrs; perform post-processing on the identified and located burrs to filter out areas of inappropriate size and remove misdetected burrs; The visual recognition module detects burr edges in the image through the Sobel algorithm, and calculates the gradient value of each pixel point by applying the Sobel operator to the image for convolution operation to find the burr edges in the image; the convolution kernel size of the Sobel operator is automatically selected according to the size and density of the burr, and the calculation is performed using a Gaussian pyramid; The Gaussian pyramid obtains a series of images by performing several downsampling and Gaussian filtering on the original image. Each level of image is more blurred and smaller than the previous level. Downsampling is achieved by downsampling the image to reduce the number of pixels in the image by a certain ratio. Gaussian filtering smoothes the image and reduces the high-frequency details of the image by performing convolution operations on the image. In the process of constructing the Gaussian pyramid, as the image is downsampled, the resolution of the image is reduced, and the size of the burr is also reduced accordingly. In the process of constructing the Gaussian pyramid, the original image is downsampled several times. The downsampling operation is achieved by sampling the image in proportion. Let the original image be I and the downsampled image be , the sampling ratio is r, then the downsampling operation is expressed as: ; in, is the downsampled image The pixel coordinates of is the corresponding pixel coordinate of the original image I; The Gaussian filtering operation is achieved by convolving the image with a Gaussian kernel. The image after Gaussian filtering is , the coordinates of the Gaussian kernel are (i, j), and the Gaussian kernel is G(i, j). Then the Gaussian filtering operation is expressed as: ; in, The image is after Gaussian filtering. The pixel coordinates of , k is the size of the Gaussian kernel; At different levels of the Gaussian pyramid, the size and density of the burrs in the image are analyzed, the area and perimeter characteristics of the burrs are calculated, and the number of burrs is counted to evaluate the size and density of the burrs; according to the estimated size and density of the burrs at different levels in the Gaussian pyramid, the appropriate convolution kernel size of the Sobel operator is determined; when the burrs are larger or the density is higher, a larger convolution kernel size is selected; when the burrs are smaller or the density is lower, a smaller convolution kernel size is selected; Deep learning module: collect a data set containing samples with and without burrs, the data set contains images of plastic products of different types and sizes, and corresponding burr labels, the burr labels are used to mark the attributes of burrs; use the data set to train the deep learning model, during the training process the model will learn to extract features from the input image, and output when the parameters of the equipment are in a certain situation, burrs will be generated, when there are burrs, predict the location of the burrs and output the specific information of the burrs; Motor control module: The motor is responsible for controlling the movement of the burr removal tool to remove burrs from the surface of plastic products; the motor speed and direction are controlled by the motor driver, which receives the specific information of the burrs sent from the deep learning module and converts it into a signal that the motor can recognize; the motor control algorithm is designed and implemented, and the control algorithm includes position control and speed control; Safety protection module: When an emergency occurs, the administrator can stop the motor with one click; monitor the motor current, stop the motor when an overload is detected, and sound an alarm; monitor the temperature of the motor, electronic components, and the device itself, and stop the motor when the temperature exceeds the safe range to prevent overheating damage.

2. The adaptive adjustment system based on the automatic burr removal device according to claim 1, Features: After the horizontal and vertical convolution operations of the Sobel operator, the horizontal gradients are obtained respectively. and vertical gradient , the convolution kernel of the Sobel operator is used to calculate the gradient value of each pixel. The mathematical formula is as follows: , , * indicates convolution operation, is the horizontal convolution kernel of the Sobel operator, It is the vertical convolution kernel of the Sobel operator; The convolution kernel size k*k dynamically selected by the Gaussian pyramid can ensure that the Sobel operator detects the burr edges in the image according to the estimated size and density of the burrs.

3. The adaptive adjustment system based on the automatic burr removal device according to claim 1, Features: In the deep learning module, an image data set containing burrs and non-burrs is prepared, the image data is annotated to indicate which areas are burrs and which areas are not burrs, the image is input into the visual recognition module for acquisition, and a convolutional neural network is selected for training based on the image data set; Get an image dataset ,in represents the Nth image, is with The corresponding annotation information, and The binary images of are the same size; in D, pixels with a V value of 1 represent burrs, and pixels with a V value of 0 represent non-burrs; the selected convolutional neural network model is ,in Represents the parameters of the convolutional neural network model. The training goal is to optimize Make Can be from the input image The location of the burr can be correctly predicted in the During the training process, the cross entropy loss function is used to measure the difference between the model's prediction results and the labeled information. The formula of the cross entropy loss function is expressed as: ; in, Represents the loss of the convolutional neural network model, which is used to measure the difference between the model prediction results and the true labels; Represents the number of pixels in each image, which is used to calculate the loss of each sample in the loss function; Representing images The label value of the e-th pixel in is a binary label. A value of 1 indicates that the pixel belongs to the burr area, and a value of 0 indicates that the pixel does not belong to the burr area. express The predicted probability that each pixel in the image belongs to the burr area.

4. The adaptive adjustment system based on the automatic burr removal device according to claim 1, Features: The deep learning module extracts specific information of the burr, including burr position, burr shape, burr size and burr density, according to the result of the convolutional neural network analysis, and the specific information of the burr will be sent to the motor driver; the motor control module enables the motor driver to convert the specific information of the burr into a control signal that can be recognized by the motor after receiving the specific information of the burr from the deep learning module; According to the specific information of the burr, the motor driver adjusts the speed of the motor to control the burr removal tool to move at a specific speed during operation; if the burrs are densely distributed or the burrs are large, the motor speed needs to be increased to increase the removal efficiency; On the contrary, if the burrs are sparsely distributed or small, it is necessary to reduce the motor speed to remove them; The motor driver is responsible for controlling the direction of movement of the motor. According to the location and layout of the burrs, the motor driver will control the motor to rotate forward, reverse or stop to ensure that the burr removal tool can move in the predetermined path and direction and remove the burrs.

5. The adaptive adjustment system based on the automatic burr removal device according to claim 1, Features: The motor control module first determines the target position of the burr removal tool in the position control algorithm, which is obtained by the deep learning module or determined by the operator's input, and the target position is expressed as an absolute coordinate or an offset relative to the current position; when the target position is determined, the position control algorithm calculates the error between the current position and the target position, which is obtained by subtracting the current position from the target position, and is expressed as a distance or angle difference; Position control uses a PID controller to adjust the movement of the motor to minimize the position error. The PID controller generates a control signal based on the size and rate of change of the position error to adjust the speed and direction of the motor. In the PID controller, the proportional term P is used to generate a control signal based on the size of the position error, the integral term I is used to eliminate steady-state errors, and the differential term D is used to suppress changes in the error. The generated PID control signal is sent to the motor driver to adjust the input voltage or current to the motor, thereby controlling the speed and direction of the motor; The motor driver controls the movement of the motor according to the received control signal, so that the motor moves toward the target position; During the operation of the motor, the position control algorithm continuously monitors the position error and adjusts the control signal in real time to maintain operation near the target position. It is necessary to adjust the PID parameters or limit the amplitude of the control signal based on the real-time feedback signal.

6. The adaptive adjustment system based on the automatic burr removal device according to claim 1, Features: The motor control module first determines the target speed of the burr removal tool in the speed control algorithm, and the target speed is obtained by the deep learning module or determined by the operator's input; When the target speed is determined, the speed control algorithm calculates the error between the current speed and the target speed, expressed as the difference between the current speed and the target speed; Speed ​​control uses a proportional controller to adjust the speed of the motor, and the control signal is proportional to the speed error; the generation of the control signal is determined based on the size and rate of change of the speed error; the generated control signal is sent to the motor driver to adjust the input voltage or current to the motor, thereby controlling the speed of the motor; the motor driver adjusts the speed of the motor based on the received control signal; During motor operation, the speed control algorithm continuously monitors the speed error and adjusts the control signal in real time.

7. The adaptive adjustment system based on the automatic burr removal device according to claim 1, Features: In the safety protection module, the device is equipped with an emergency stop button, and the administrator can immediately stop the rotation of the motor by pressing the emergency stop button; the safety protection module monitors the current of the motor and immediately stops the operation of the motor when a current overload is detected; when an overload is detected, the system automatically stops the operation of the motor and issues an alarm to alert the administrator; the safety protection module monitors the temperature of the motor, electronic components and the device itself, and stops the operation of the motor when the temperature exceeds a safe range; when the temperature exceeds a set threshold, the system automatically stops the operation of the motor and issues an alarm to alert the administrator; the safety protection module is integrated with the human-machine interface of the device, and the alarm alerts the administrator through sound, light signals and display screens; after a security incident occurs, the device implements an automatic reset function, and can restart the device and resume normal operation after the problem is solved.

8. An adaptive adjustment method based on an automatic burr removal device, applied to an adaptive adjustment system based on an automatic burr removal device according to claims 1 to 7, Features: Here are the steps: S100, use an industrial camera to focus on the burrs on the surface of plastic products and continuously capture images within the working distance of the device; S200, preprocessing the captured image, and using image processing technology to analyze and extract features from the preprocessed image to identify and locate burrs; S300, collecting a data set including samples with burrs and samples without burrs, and using the data set to train a deep learning model; S400, during the training process, the model will learn to extract features from the input image and output whether there is a burr or no burr, and when there is a burr, the specific information of the burr will be output; S500, controlling the speed and direction of the motor through the motor driver, the motor driver receiving the specific information of the burr sent from the deep learning module and converting it into a signal that the motor can recognize; S600, designing and implementing a motor control algorithm, wherein the control algorithm includes position control and speed control; S700, when an emergency occurs, the administrator can stop the motor with one click; S800, monitors the current of the motor, stops the operation of the motor when an overload is detected, and issues an alarm; monitors the temperature of the motor, electronic components, and the device itself, and stops the operation of the motor when the temperature exceeds the safe range to prevent overheating damage.