Image recognition-based online monitoring method for warping deformation of fused deposition modeling product

Abstract

The application provides an image recognition-based fused deposition modeling product warping deformation online monitoring method, carries out initialization processing on a visual sensor collected signal, carries out convolution on an image color representation matrix, sets an excitation function, carries out post-processing on an excitation result, prepares for extracting a boundary contour, sets a target function, obtains boundary contour information based on the target function, calculates product warping deformation, obtains warping deformation results online until processing is completed, so that the fused deposition modeling product warping deformation is monitored online. The application carries out convolution on the image color representation matrix, improves recognition accuracy, provides a solid foundation for judging whether each pixel block is a boundary and automatically calculating warping deformation, realizes fused deposition modeling product warping deformation online monitoring, and effectively overcomes the problem of low monitoring system precision.

Description

Technical Field

[0001] This invention relates to the field of image recognition, and in particular to a method for monitoring fused deposition modeled products. Background Technology

[0002] In the field of fused deposition modeling (FDM), the dimensions and geometric accuracy of a product have a significant impact on its assembly and performance. Warpage significantly reduces the dimensions and geometric accuracy of the product. Traditionally, the method for measuring product molding is offline measurement after processing, which cannot control the product molding accuracy online. Once excessive warpage occurs during processing, the product molding accuracy is likely to fail to meet the design requirements and be scrapped. However, relevant research on product warpage detection at home and abroad has not yet considered online monitoring of product warpage during the fused deposition modeling process. [Pan Shenghu, Xie Lincheng, Liu Yunqiang, et al. Research on workpiece contour recognition method based on surface evaluation [J]. Journal of Engineering Design, 2023, 30(1):13-19.]. This literature proposes a workpiece contour recognition method based on surface evaluation, but it does not propose how to identify and quantify the warpage deformation of the workpiece. Traditional warpage measurement methods cannot determine whether the warpage deformation meets the design requirements during the molding process, and products with warpage deformation exceeding the design threshold during the molding process are obviously unqualified. Online monitoring of warpage deformation during processing is crucial for controlling the geometric and dimensional compliance of manufactured products. Therefore, developing an image recognition-based online monitoring method for warpage deformation in fused deposition modeling (FDM) products is of significant importance. Summary of the Invention

[0003] To overcome the shortcomings of existing technologies, this invention provides an online monitoring method for warpage deformation of fused deposition modeling (FDM) products based on image recognition. Warpage deformation of FDM products is typically monitored after processing. If the warpage deformation exceeds the design threshold during processing, the product is considered defective, and continuing processing wastes raw materials and time, reducing productivity. To monitor warpage deformation online during the FDM process and avoid the problem of ineffective processing due to unmonitored warpage deformation, this invention provides an online monitoring method for warpage deformation of FDM products based on image recognition. This allows warpage deformation information to be acquired during the processing of the FDM product, which is of significant importance for controlling the molding accuracy of the product.

[0004] The technical solution adopted by this invention to solve its technical problem includes the following steps:

[0005] Step (a): To increase contrast and improve monitoring effect, when acquiring visual information, the contrast ratio between the background color and the fused deposition modeling material should be no less than 80:1; set the horizontal reference to the upper surface of the fused deposition modeling equipment workbench, with a height of 0, and define the distance from the workbench along the direction perpendicular to the upper surface as height Z, in mm; when laying out the vision sensor, the distance Z from the workbench in the height direction of the vision sensor is 0, and the lens direction of the vision sensor is perpendicular to the longitudinal section of the workbench and 1000 mm away from the workbench. The vision sensor is then initialized by arranging the acquired signals in this manner.

[0006] Let the number of pixels acquired by the visual sensor at time t0 be a*b, and the color representation index of the pixel in the i-th row and j-th column be:

[0007] CI ij =RGB ij

[0008] CI ij RGB represents the color representation of the pixel in the i-th row and j-th column. ij Let be the coupling value of the red, green, and blue color components of the pixel in the i-th row and j-th column; t0 represents the initial time; a represents the number of pixels acquired by the horizontal vision sensor; b represents the number of pixels acquired by the vertical vision sensor; i represents the i-th row of pixels acquired by the vision sensor; j represents the j-th column of pixels acquired by the vision sensor.

[0009] Assigning values ​​to the elements of the image color representation matrix, the image color representation matrix becomes:

[0010]

[0011] Matc is the image color representation matrix;

[0012] Step (b): In order to effectively extract the features of image color and reduce redundant information, the image color representation matrix is ​​convolved;

[0013] Step (c): To perform high-dimensional mapping on linearly inseparable information, an activation function is set to activate the convolution result matrix RC(p, q). The saturation phenomenon of the activation function is that as the independent variable increases, the derivative of the dependent variable decreases, causing the gradient far from the origin to decay. When the independent variable approaches infinity, the derivative of the dependent variable approaches 0, and the gradient disappears. Let the activation function be f. ex (). When designing and selecting activation functions, saturation within the domain should be avoided as much as possible to reduce gradient decay and vanishing.

[0014] Step (d): Post-process the excitation results by extracting the excitation values ​​of the elements in the convolution result matrix of the (p+1)th row and qth column, the pth row and qth column, and the pth row and (q+1)th column, and then calculating the excitation values ​​using the following formula:

[0015]

[0016] In the formula, x p Let y be the horizontal coordinate of the p-th pixel; q Let q be the vertical coordinate of the q-th pixel.

[0017] Fr p，q The difference in color information between adjacent pixels is reflected and defined as the boundary feature recognition factor. By comparing it with the objective function in step (e), it is determined whether a pixel is a boundary point.

[0018] Step (e): To prepare for extracting the boundary contour, set the objective function as ta p,q ;

[0019] Step (f): Obtain boundary contour information based on the objective function;

[0020] The function to determine whether the pixel in the i-th row and j-th column is a boundary is:

[0021]

[0022] In the formula, fbo(i,j)=1 indicates that the pixel in the i-th row and j-th column is a boundary contour pixel; fbo(i,j)=0 indicates that the pixel in the i-th row and j-th column is a non-boundary contour pixel.

[0023] Step (g): Calculate the product warpage.

[0024] Step (h): Set the time interval from step (a) to step (g) to be Δt. Within the time range of t0 to t0+Δt, repeat the steps between (a) and (g) to obtain t0+nΔtn∈N online. + The warpage deformation fwd(p,q) at any given moment is recorded until the processing is complete, thus enabling online monitoring of the warpage deformation of the fused deposition modeled product.

[0025] The step of convolving the image color representation matrix in step (b) is as follows:

[0026] Let the convolution kernel be an m×n matrix, then the convolution kernel is:

[0027]

[0028] The color representation matrix is ​​convolved, and the result is:

[0029]

[0030] In the formula, RC(p, q) is the convolution result matrix, where p and q are the p-th row and q-th column of the pixel, respectively; p∈[1, a+1-m] and p∈N + q∈[1,b+1-n] and q∈N + .

[0031] In step (c), the specific implementation steps are as follows:

[0032] Let the saturation test function be f. fch ():

[0033]

[0034] In the formula, f′ ex (x) is the first derivative of the excitation function;

[0035] The excitation function f ex When substituting () into the saturation test function, make f fch (x) is greater than 0, thus avoiding saturation in the domain of the activation function and reducing gradient decay and vanishing;

[0036] Therefore, the excitation function f ex () is set as:

[0037]

[0038] In the formula, the parameter λ is a value between 0 and 1;

[0039] When designing or selecting an activation function, saturation within the domain should be avoided as much as possible to reduce gradient decay and vanishing. Let the activation function be f. ex ();

[0040] In step (e), the objective function is calculated using the following formula:

[0041]

[0042] In the formula, η is the threshold of the objective function; ta p,q Calculate the objective function value for the pixel in row p and column q.

[0043] In step (g), the step of calculating the product warpage deformation is as follows:

[0044] Let the pixel in the i-th row be at a horizontal position of z. i The horizontal position of the pixel in row b is taken as the reference position, and the distance between pixels in rows is l. r The unit is mm; then the calculated result of the warpage deformation value of the fused deposition modeled product at time t0 is fwd(p,q), as shown below:

[0045] fwd(p,q)=max{(ib)*l r *fbo(i,j)}i∈[1,p]∩N + j∈[1,q]∩N +

[0046] After calculating the warpage deformation value of the fused deposition modeled product at time t0, the time interval required from step (a) to step (g) is obtained by using a timer.

[0047] The beneficial effects of this invention are that, apart from requiring a properly positioned visual sensor, the entire process—including the visual sensor signal acquisition initialization processing module, convolution calculation module, function activation module, feature recognition module, and product warpage deformation calculation module—can be automated on a computer using this method, thereby achieving online monitoring of warpage deformation in fused deposition modeling (FDM) products. The innovation of this method lies in addressing the issue that, during the actual FDM process, the product color may differ too little from the ambient base color, leading to low accuracy in contour recognition and inaccurate warpage deformation calculation. Therefore, convolution is applied to the image color representation matrix to improve recognition accuracy. Furthermore, a feature recognition factor is designed based on the steepest descent principle, providing a solid foundation for determining whether each pixel block is a boundary and for automatically calculating warpage deformation. Therefore, this method achieves online monitoring of warpage deformation in FDM products and effectively overcomes the problem of low accuracy in monitoring systems. Attached Figure Description

[0048] Figure 1 This is a flowchart of the steps of the method described in this invention.

[0049] Figure 2 This is a schematic diagram of the visual sensor layout. Detailed Implementation

[0050] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0051] To address the aforementioned problems, this invention proposes an online monitoring method for warpage deformation of fused deposition modeled (FDM) products based on image recognition. This method is used for online monitoring of warpage deformation of FDM products and is characterized by the following steps:

[0052] An image recognition-based online monitoring method for warpage deformation of fused deposition modeled products is used to monitor product warpage deformation during the fused deposition modeling process. The specific implementation method is as follows:

[0053] (a): Let the initial processing time for fused deposition modeling be t0. Based on the material color of the fused deposition modeling product, set the background color to enhance the contrast between the product color and the background color. For example, if the material color of the fused deposition modeling product is white, then set the background color to black. Select a vision sensor resolution of 1920*1080. Position the vision sensor so that the lens is parallel to surface A, such as... Figure 2 As shown, the signals acquired by the vision sensor are initialized.

[0054] (b): Select an appropriate convolution kernel. In order to effectively extract the color features of the image and sharpen the boundaries, the Laplacian operator can be set as the convolution kernel for convolution operations.

[0055] (c): Set the activation function to activate the convolution result. To reduce gradient vanishing and bias, the ReLU function, an unsaturated activation function, is selected.

[0056] (d): Post-process the excitation results, perform feature recognition, and calculate the feature recognition factor.

[0057] (e): Determine the threshold η of the objective function, with an initial value of 0. Based on the online monitoring effect of warping deformation, determine the optimal threshold.

[0058] (f): Obtain the boundary contour based on the step-type decision boundary function and the objective function.

[0059] (g): Based on the horizontal position reference, the distance between row pixels and the calculated boundary contour, calculate the product warping deformation under the constraint that the distance between the boundary contour pixels and the horizontal position reference is the maximum.

[0060] (h): Let the processing time of the system from step (a) to step (g) be Δt. Then, at t0+Δt, the process of steps (a) to (g) continues until the fused deposition modeling process ends, thereby realizing online monitoring of warpage deformation of fused deposition modeling products based on image recognition.

Claims

1. A method for online monitoring of warpage deformation of fused deposition modeled products based on image recognition, characterized in that... Includes the following steps: Step (a): To increase contrast and improve monitoring effect, when acquiring visual information, the contrast ratio between the background color and the fused deposition modeling material should be no less than 80:1; set the horizontal reference to the upper surface of the fused deposition modeling equipment workbench, with a height of 0, and define the distance from the workbench along the direction perpendicular to the upper surface as height Z (in mm); when laying out the vision sensor, the distance Z from the workbench in the height direction of the vision sensor is 0, and the lens direction of the vision sensor is perpendicular to the longitudinal section of the workbench and 1000 mm away from the workbench. The vision sensor is then initialized by performing this initialization process on the acquired signals. Suppose At time t, the visual sensor collects a*b pixels, and the color representation index of the pixel in the i-th row and j-th column is: ； The color representation index of the pixel in the i-th row and j-th column; The coupling values ​​of the red, green, and blue color components of the pixel in the i-th row and j-th column; Indicates the initial time; a represents the number of pixels collected by the horizontal vision sensor; b represents the number of pixels collected by the vertical vision sensor. i represents the i-th row of pixels acquired by the vision sensor; j represents the j-th column of pixels acquired by the vision sensor. Assigning values ​​to the elements of the image color representation matrix, the image color representation matrix becomes: ； The image color representation matrix; Step (b): In order to effectively extract the features of image color and reduce redundant information, the image color representation matrix is ​​convolved; Step (c): In order to perform high-dimensional mapping on linearly inseparable information, an activation function is set, and the convolution result matrix is... The activation function saturates because as the independent variable increases, the derivative of the dependent variable decreases, causing the gradient to decay away from the origin. When the independent variable approaches infinity, the derivative of the dependent variable approaches zero, and the gradient disappears. Let the activation function be... Excitation functions should be designed and selected to avoid saturation within their domain. Step (d): Post-process the activation results by extracting the activation values ​​of the elements in the convolution result matrix of the (p+1)th row and qth column, the pth row and qth column, and the pth row and (q+1)th column, and then calculating the activation values ​​using the following formula: ； In the formula, Let p be the horizontal coordinate of the p-th pixel. Let be the vertical coordinate of the q-th pixel; The difference in color information between adjacent pixels is reflected and defined as the boundary feature recognition factor. By comparing it with the objective function in step (e), it is determined whether a pixel is a boundary point. Step (e): To prepare for extracting the boundary contour, the objective function is set as follows. ; In step (e), the objective function is calculated using the following formula: ； In the formula, The threshold of the objective function; Calculate the objective function value for the pixel in row p and column q. Step (f): Obtain boundary contour information based on the objective function; The function to determine whether the pixel in the i-th row and j-th column is a boundary is: ； In the formula, The pixel in the i-th row and j-th column is a boundary contour pixel; This indicates that the pixel in the i-th row and j-th column is a non-boundary contour pixel; Step (g): Calculate the product warpage. In step (g), the step of calculating the product warpage deformation is as follows: Let the pixel in the i-th row be at a horizontal position of . The horizontal position of the pixel in row b is taken as the reference position, and the distance between pixels in rows is... The unit is mm; then The calculated warpage value of the fused deposition modeled product at any given time is: As shown below: ； Calculate After determining the warpage deformation value of the fused deposition modeled product, a timer is used to obtain the time interval required to complete step (a) to step (g). Step (h): Set the time interval from step (a) to the completion of step (g) to be [value]. ,exist arrive Within the time frame, repeat steps (a) through (g) to obtain online results. Warping deformation of time As a result, the warpage deformation of the fused deposition modeled product can be monitored online until the processing is completed.

2. The online monitoring method for warpage deformation of fused deposition modeled products based on image recognition according to claim 1, characterized in that: The step of convolving the image color representation matrix in step (b) is as follows: Let the convolution kernel be If the matrix is ​​such that the convolution kernel is: ； The color representation matrix is ​​convolved, and the result is: ； In the formula, The result matrix of the convolution. , These are the p-th row and q-th column of the pixel, respectively; , .

3. The online monitoring method for warpage deformation of fused deposition modeled products based on image recognition according to claim 1, characterized in that: In step (c), the specific implementation steps are as follows: Let the saturation test function be... : ； In the formula, The first derivative of the activation function; Excitation function When substituting into the saturation test function, make A value greater than 0 is used to avoid saturation within the domain of the activation function. Therefore, the activation function Set as: ； In the formula, the parameter A value between 0 and 1; When designing or selecting an excitation function, saturation within its domain should be avoided; let the excitation function be... .

Images