A control system and method for a fully automatic dispensing machine

Abstract

The present application relates to the technical field of dispensing machine control, and discloses a control system and method for a fully automatic dispensing machine. After the dispensing task is started, the industrial camera collects images containing specific reference feature areas of the workpiece in real time; the industrial computer then detects and locates key reference feature points from the image and obtains their pixel coordinates. Based on the pre-calibrated coordinate system conversion relationship between the camera and the machine motion system, the feature points are converted into actual positions in world coordinates. By accurately comparing this actual position with the preset ideal position, the comprehensive deviation transformation parameters of the workpiece are calculated. Finally, based on this deviation parameter, the preset dispensing path is dynamically adjusted, and a dispensing drive instruction is generated to guide the dispensing head to the target position accurately, thereby effectively overcoming the influence of factors such as workpiece incoming material differences and positioning errors, and significantly improving the accuracy, stability and automation level of the dispensing operation.

Description

Technical Field

[0001] The present application relates to the field of glue dispensing machine control technology, and more specifically, to a control system and method for a fully automatic glue dispensing machine. Background Art

[0002] With the rapid development of modern manufacturing toward high precision, high efficiency, and high automation, automated dispensing technology has become an indispensable key process in numerous fields, including electronic packaging, automotive manufacturing, medical devices, and new energy. Dispensing quality, including accuracy of dispensing location, uniformity of dispensing volume, and consistency of dispensing line shape, directly impacts product performance, reliability, and service life. Traditional dispensing operations, whether manual or relying on automated equipment relying solely on mechanical positioning, face significant challenges. Manual dispensing is inefficient, inconsistent, and highly dependent on operator skill and experience. While early automated dispensing equipment improved production efficiency, its control methods were often based on preset fixed paths and assumptions about ideal workpiece positioning. These control methods lacked the ability to effectively adapt to and correct for factors such as material variability, positioning errors, fixture deviations, and minor deformations that may occur during the workpiece processing process.

[0003] Specifically, in the prior art, the motion trajectory of many fully automatic dispensing machines is pre-programmed or set through teaching when performing dispensing tasks. The premise of this method is that each workpiece to be processed can be placed accurately and consistently in the predetermined position. However, in the actual production environment, the shape tolerance of the workpiece, slight variations in surface features, and the inevitable positioning errors during the transmission and clamping process will cause the actual position and posture of the workpiece to deviate from the ideal state. If the dispensing system cannot perceive and compensate for these deviations, and the dispensing head still moves according to a fixed program path, it is very easy to cause the dispensing position to deviate from the target area, resulting in quality defects such as glue overflow, missing glue or glue line deviation, and in severe cases even leading to product scrapping, increasing production costs and reducing the pass rate. Summary of the Invention

[0004] In order to solve the problem that the existing automatic dispensing method is not adaptable enough when dealing with the actual position and posture changes of the workpiece, resulting in difficulty in ensuring dispensing accuracy, the present application is proposed. The embodiments of the present application propose a control system and method for a fully automatic dispensing machine.

[0005] According to one aspect of the present application, a control method for a fully automatic dispensing machine is provided, including: after detecting a dispensing task start instruction, issuing a camera trigger signal; an industrial camera collects original image data containing a workpiece reference feature area according to the camera trigger signal, and transmits the original image data containing the workpiece reference feature area to an industrial computer; the industrial computer detects and locates key reference feature points in the original image data containing the workpiece reference feature area to obtain a position set of the key reference feature points in pixel coordinates; based on the coordinate system calibration parameters of the camera and machine motion system, the key reference feature points are converted from the pixel coordinate system to the world coordinate system to obtain a position set of the key reference feature points in the world coordinate system; the position set of the key reference feature points in the world coordinate system is compared with the ideal position set of the key reference feature points in the world coordinate system to obtain a deviation transformation parameter; and based on the deviation transformation parameter, a dispensing drive instruction is generated.

[0006] In a possible implementation, the original image data containing the workpiece reference feature area is transmitted to the industrial computer via a high-speed interface.

[0007] In one possible implementation, the industrial computer detects and locates the key reference feature points in the original image data containing the workpiece reference feature area to obtain a position set of the key reference feature points in pixel coordinates, including: performing distortion correction, image noise reduction and normalization processing on the original image data containing the workpiece reference feature area to obtain a preprocessed image; and inputting the preprocessed image into a trained deep learning model to obtain a position set of the key reference feature points in pixel coordinates.

[0008] In one possible implementation, the trained deep learning model is a key point detection model based on a convolutional neural network.

[0009] In one possible implementation, the preprocessed image is input into a trained deep learning model to obtain a position set of the key reference feature points in pixel coordinates, including: using a convolutional neural network model to extract image features of the preprocessed image to obtain an image feature matrix; performing convergence optimization on the image feature matrix based on tensor feature association gradient and dynamic tensor field to obtain an optimized image feature matrix; inputting the optimized image feature matrix into a softmax function to determine the probability value of each pixel as a key point; and determining the position set of the key reference feature points in pixel coordinates based on the probability value of each pixel as a key point.

[0010] In one possible implementation, the image feature matrix is ​​subjected to convergence optimization based on tensor feature associated gradients and dynamic tensor fields to obtain an optimized image feature matrix, including: performing a deconvolution operation on the image feature matrix to obtain a priori feature matrix in the gradient direction; deriving a differential field quantity distribution based on gradient response based on the image feature matrix and the priori feature matrix to obtain a gradient inverse response matrix; calculating a hyperspace superposition field of the gradient based on the image feature matrix and the priori feature matrix to obtain a hyperspace anchor matrix; using the hyperspace anchor matrix as a hyperspace anchor to normalize the convergence of the constraint field intensity center to obtain a canonical field intensity matrix; and using hyperparameters as weighting coefficients to fuse the gradient inverse response matrix and the canonical field intensity matrix to obtain the optimized image feature matrix.

[0011] In one possible implementation, the position set of the key reference feature points in the world coordinate system is compared with the ideal position set of the key reference feature points in the world coordinate system to obtain deviation transformation parameters, including: calculating the optimal 2D rigid body transformation matrix between the position set of the key reference feature points in the world coordinate system and the ideal position set of the key reference feature points in the world coordinate system, the optimal 2D rigid body transformation matrix including X-direction translation parameters, Y-direction translation parameters and rotation parameters around the Z axis; extracting the X-direction translation parameters, Y-direction translation parameters and rotation parameters around the Z axis from the optimal 2D rigid body transformation matrix as the deviation transformation parameters.

[0012] In one possible implementation, a dispensing drive instruction is generated based on the deviation transformation parameter, including: obtaining preset dispensing path data; inputting the deviation transformation parameter and the preset dispensing path data into a motion controller, and the motion controller fine-tuning the preset dispensing path data based on the deviation transformation parameter to obtain dispensing path adjustment data; and generating the dispensing drive instruction based on the dispensing path adjustment data.

[0013] According to another aspect of the present application, a control system for a fully automatic dispensing machine is provided, which is used to implement the above-mentioned control method for the fully automatic dispensing machine, including: a dispensing start trigger control module, which is used to send a camera trigger signal after detecting a dispensing task start instruction; an image acquisition and transmission module, which is used for the industrial camera to acquire original image data containing a workpiece reference feature area according to the camera trigger signal, and transmit the original image data containing the workpiece reference feature area to an industrial computer; an image feature point detection and positioning module, which is used for the industrial computer to detect and locate key reference feature points in the original image data containing the workpiece reference feature area to obtain a position set of the key reference feature points in pixel coordinates; a coordinate system conversion processing module, which is used to convert the key reference feature points from the pixel coordinate system to the world coordinate system based on the coordinate system calibration parameters of the camera and the machine motion system to obtain a position set of the key reference feature points in the world coordinate system; a feature point position comparison module, which is used to compare the position set of the key reference feature points in the world coordinate system with the ideal position set of the key reference feature points in the world coordinate system to obtain deviation transformation parameters; and a dispensing drive instruction generation module, which is used to generate a dispensing drive instruction based on the deviation transformation parameters.

[0014] Compared with the prior art, the control system and method of the fully automatic dispensing machine provided by the present application, after the dispensing task is started, the industrial camera collects images containing specific reference feature areas of the workpiece in real time; the industrial computer then detects and locates key reference feature points from the image and obtains their pixel coordinates. Based on the pre-calibrated coordinate system conversion relationship between the camera and the machine motion system, the feature points are converted into actual positions in world coordinates. By accurately comparing this actual position with the preset ideal position, the comprehensive deviation transformation parameters of the workpiece are calculated. Finally, based on this deviation parameter, the preset dispensing path is dynamically adjusted, and a dispensing drive instruction is generated to guide the dispensing head to the target position accurately, thereby effectively overcoming the influence of factors such as workpiece incoming material differences and positioning errors, and significantly improving the accuracy, stability and automation level of the dispensing operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0016] Figure 1 The figure illustrates a schematic flow chart of a control method for a fully automatic dispensing machine according to an embodiment of the present application.

[0017] Figure 2The figure illustrates a schematic flow chart of step S3 in the control method of the fully automatic dispensing machine according to an embodiment of the present application.

[0018] Figure 3 The figure illustrates a schematic flow chart of step S5 in the control method of the fully automatic dispensing machine according to an embodiment of the present application.

[0019] Figure 4 The figure shows a schematic flow chart of step S6 in the control method of the fully automatic dispensing machine according to an embodiment of the present application.

[0020] Figure 5 The figure shows a schematic block diagram of a control system of a fully automatic dispensing machine according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0022] Figure 1 FIG2 is a schematic flow chart of a control method of a fully automatic dispensing machine according to an embodiment of the present application. Figure 1 As shown, the present application provides a control method for a fully automatic dispensing machine, including: S1: after detecting a dispensing task start instruction, sending a camera trigger signal; S2: the industrial camera collects original image data containing a workpiece reference feature area according to the camera trigger signal, and transmits the original image data containing the workpiece reference feature area to an industrial computer; S3: the industrial computer detects and locates the key reference feature points in the original image data containing the workpiece reference feature area to obtain a position set of the key reference feature points in pixel coordinates; S4: based on the coordinate system calibration parameters of the camera and the machine motion system, converting the key reference feature points from the pixel coordinate system to the world coordinate system to obtain a position set of the key reference feature points in the world coordinate system; S5: comparing the position set of the key reference feature points in the world coordinate system with the ideal position set of the key reference feature points in the world coordinate system to obtain a deviation transformation parameter; S6: generating a dispensing drive instruction based on the deviation transformation parameter.

[0023] Specifically, in step S1, after the start instruction of the dispensing task is detected, a camera trigger signal is issued. It should be understood that in the automated production process, the start instruction of the dispensing task indicates that a specific workpiece is ready and is about to enter the dispensing process. Triggering the camera at this time can ensure that the image captured by the camera is the latest state of the workpiece to be processed, including its actual position and posture information on the workbench. If the camera is triggered too early, the workpiece may not be in place or stable, and the captured image will have no reference value; if it is triggered too late, the entire dispensing cycle may be delayed, affecting production efficiency, and the dispensing action may even be mistakenly started before the camera imaging is completed. Therefore, the camera trigger is tightly coupled with the task start instruction, that is, the camera trigger signal is issued after the start instruction of the dispensing task is detected.

[0024] In one embodiment, the dispensing task start command can be issued by a higher-level control system (such as a PLC or MES system) after confirming that the workpiece has passed the previous process and has been accurately delivered to the dispensing station. In other embodiments of the present application, the dispensing task start command can also be generated by an operator manually pressing a start button through a human-machine interface (HMI); or it can be automatically triggered by a sensor (such as a photoelectric sensor or limit switch) detecting the workpiece's position. Upon detecting the dispensing task start command, the industrial computer software immediately executes a preset action, namely, issuing a camera trigger signal. Camera triggering methods are primarily hardware and software. In hardware triggering mode, the core controller sends a level signal, such as a rising edge, falling edge, or a pulse signal of a specific pulse width, to a dedicated trigger input pin of the industrial camera via its digital output port. This physical connection ensures low latency and high reliability of the trigger signal. In software triggering mode, the industrial computer and the industrial camera are connected via a data interface. When the control software on the industrial computer detects the dispensing task start command, it sends a specific software command to the camera through the camera's software development kit or application programming interface, instructing the camera to acquire images.

[0025] Specifically, in step S2, the industrial camera collects raw image data containing the reference feature area of ​​the workpiece according to the camera trigger signal, and transmits the raw image data containing the reference feature area of ​​the workpiece to the industrial computer. It should be understood that the trigger instruction is issued at the beginning of the dispensing task, that is, when the workpiece is basically in place. The camera only performs image acquisition after receiving this specific trigger, which means that the acquired image data can truly reflect the current state of the workpiece to be processed on the workbench. This event-driven acquisition method avoids the massive amount of invalid data and unnecessary computing resource consumption caused by continuous acquisition, ensuring that each frame of processed image is targeted, thereby maximizing the effectiveness and real-time performance of visual guidance.

[0026] Specifically, when the industrial camera receives a camera trigger signal from the industrial computer, its internal image acquisition process is activated. The industrial camera immediately captures an image frame according to its preset exposure parameters and other settings to collect raw image data containing the workpiece reference feature area. Here, it should be understood that the actual workpiece may have a complex structure, but it is usually only necessary to locate a few key, easily identifiable, and stable reference feature points on it, such as specific holes, edges, corners, or preset marking points, to accurately calculate the position and posture of the entire workpiece. By configuring the camera's field of view and resolution to just be able to clearly resolve these specific reference feature areas, the complexity of the subsequent image processing algorithm can be significantly simplified, processing time can be shortened, and the accuracy and anti-interference ability of feature extraction can be improved. Subsequently, the raw image data containing the workpiece reference feature area is transmitted to the industrial computer via a high-speed interface.

[0027] Specifically, in step S3, the industrial computer detects and locates the key reference feature points in the original image data containing the workpiece reference feature area to obtain the position set of the key reference feature points in pixel coordinates. It should be understood that in an actual production environment, the incoming position and posture of the workpiece may deviate due to slight errors in the transmission, clamping and other links, and the workpiece itself may also have certain deformations. If the dispensing machine relies entirely on a preset fixed path to operate, these deviations will lead to inaccurate dispensing positions, seriously affecting product quality and even causing product scrapping. Therefore, it is necessary to have the ability to perceive the actual position and posture of the workpiece and adjust the dispensing path accordingly. Based on this, the industrial computer detects and locates the key reference feature points in the original image data containing the workpiece reference feature area. By accurately identifying and locating representative feature points on the workpiece, the deviation of the workpiece relative to the ideal position can be calculated.

[0028] In one embodiment, Figure 2 As shown, the industrial computer detects and locates key reference feature points in the original image data containing the workpiece reference feature region to obtain a set of positions of the key reference feature points in pixel coordinates, including: S31: performing distortion correction, image noise reduction, and normalization on the original image data containing the workpiece reference feature region to obtain a preprocessed image; S32: inputting the preprocessed image into a trained deep learning model to obtain a set of positions of the key reference feature points in pixel coordinates. The trained deep learning model is a key point detection model based on a convolutional neural network.

[0029] First, a series of preprocessing operations are performed on the raw image data containing the workpiece reference feature region to obtain a preprocessed image. This eliminates or reduces noise and distortion introduced during image acquisition and converts the image data into a format more suitable for deep learning model processing. Specifically, the preprocessing operations include distortion correction, image noise reduction, and normalization. Distortion correction is performed to account for the fact that industrial camera lenses, particularly wide-angle or low-cost lenses, often introduce geometric distortion, primarily radial and tangential distortion. Radial distortion causes straight lines in the image to bend, with the curvature becoming more pronounced the further away from the image center. Tangential distortion is caused by the lens itself not being completely parallel to the image sensor plane. These distortions can cause deviations between the true position of object points in the image and their imaged positions. If not corrected, this directly affects the accuracy of keypoint positioning. In one specific embodiment, the intrinsic and extrinsic parameters and distortion coefficients can be obtained through camera calibration. For example, the Zhang Zhengyou calibration method can be used to calculate the camera's intrinsic parameter matrix K and distortion coefficient vector D by capturing images of a checkerboard calibration plate in different poses. After obtaining these parameters, any point in the original image can be mapped back to its ideal undistorted position through a mathematical model. By applying this inverse distortion model to each pixel of the original image, a distortion-corrected image can be generated.

[0030] Next, considering that the image will inevitably be interfered with by various noise sources during the acquisition and transmission process, such as sensor thermal noise, photon shot noise, readout circuit noise, etc. These noises will reduce the signal-to-noise ratio of the image, blur image details, and interfere with the precise identification of subsequent feature points. Therefore, image noise reduction is performed on the distortion-corrected image. In a specific embodiment, Gaussian filtering can be used. The Gaussian filter is a linear smoothing filter that uses a two-dimensional Gaussian function as the convolution kernel. The weight distribution of the Gaussian kernel is bell-shaped, with the center point having the largest weight and the pixel weight farther from the center having a smaller weight. Its function is to perform a weighted average of the pixel values ​​in the neighborhood, thereby smoothing the image and suppressing high-frequency noise.

[0031] The denoised image is then normalized. Normalization adjusts the pixel values ​​of an image to a predefined range or distribution, which is crucial for stable training and efficient convergence of deep learning models. Different images may have significant differences in overall brightness and contrast due to factors such as lighting conditions and exposure settings. Normalization can eliminate the effects of these differences. In a specific embodiment, minimum-maximum normalization can be used to linearly scale pixel values ​​to the range [0, 1] or [-1, 1].

[0032] After the aforementioned distortion correction, image denoising, and normalization, the resulting preprocessed image is fed as input into a trained convolutional neural network-based keypoint detection model. Convolutional neural networks, due to their powerful feature extraction capabilities and effective modeling of spatial hierarchical information, are well-suited for object localization tasks in images. In one specific embodiment, the convolutional neural network-based keypoint detection model comprises an input layer, a convolutional neural network model, and a softmax function. Specifically, the input layer is used to receive the preprocessed image. The convolutional neural network model comprises multiple convolutional layers, multiple pooling layers, and a fully connected layer. In one specific embodiment, the convolutional neural network model comprises: a first convolutional layer using 32 3×3 convolutional kernels with a stride of 1, padding of 1, and a ReLU activation function; a first pooling layer using max pooling with a pooling window size of 2×2 and a stride of 2; and a second convolutional layer using 64 3×3 convolutional kernels with a stride of 1, padding of 1, and a ReLU activation function. The second pooling layer performs max pooling with a pooling window size of 2×2 and a stride of 2. The third convolutional layer uses 128 convolution kernels of size 3×3, a stride of 1, padding of 1, and a Reinforced Lu (ReLU) activation function. The third pooling layer also performs max pooling with a pooling window size of 2×2 and a stride of 2. After the convolution and pooling operations, the 3D feature map is flattened into a 1D vector, then resized through a fully connected layer, ultimately outputting an image feature matrix. This image feature matrix is ​​input into a Softmax function to generate a probability value for each pixel being a keypoint, forming a probability heatmap where each pixel value represents the probability of that pixel being a keypoint. In one specific embodiment, a threshold is applied to this probability heatmap to filter out points deemed noisy or unimportant. For example, setting a threshold of 0.7 (this is just an example and can be adjusted based on actual conditions) means that a pixel is considered a potential keypoint only if the probability value at that location is greater than or equal to 0.7. Next, the entire probability heatmap is traversed to find all pixel locations that meet the above conditions. For each such pixel location, its surrounding area is further checked to ensure that it is indeed a local maximum. This is achieved by comparing the probability values ​​of the pixel with its neighboring pixels (for example, the 8-neighborhood, i.e., neighbors in the upper, lower, left, right, and four diagonal directions). If the probability value of the current pixel is higher than that of all its neighboring pixels, it is confirmed to be a key reference feature point. The coordinates (x, y) of these key points are recorded to obtain the set of key reference feature points in pixel coordinates.

[0033] In particular, since the key point detection model based on the convolutional neural network extracts the image features of the preprocessed image through the convolution kernel to determine the probability of the pixel feature as the key point to perform key point detection, considering the width and height dimensions of the image processed by the convolutional neural network and the channel dimension of the convolutional neural network itself, the convolution kernel of the convolutional neural network is essentially a dynamic modeling of the dynamic tensor field of the tensor feature. Moreover, since the key reference feature points subsequently need to be converted from the pixel coordinate system to the world coordinate system and the coordinate deviation transformation parameters need to be calculated, the convolutional neural network needs to have strong constraints on the associated gradient direction in the nonlinear tensor space and the field intensity convergence center of the dynamic tensor field, thereby improving the linear-nonlinear characteristic expression effect of the key point space in the image semantic feature scenario.

[0034] Based on this, in another embodiment of the present application, the preprocessed image is input into a trained deep learning model to obtain a position set of the key reference feature points in pixel coordinates, including: using a convolutional neural network model to extract image features of the preprocessed image to obtain an image feature matrix; performing convergence optimization based on tensor feature association gradient and dynamic tensor field on the image feature matrix to obtain an optimized image feature matrix; inputting the optimized image feature matrix into a softmax function to determine the probability value of each pixel as a key point; and determining the position set of the key reference feature points in pixel coordinates based on the probability value of each pixel as a key point.

[0035] Specifically, the image feature matrix is ​​subjected to convergence optimization based on tensor feature correlation gradient and dynamic tensor field to obtain an optimized image feature matrix, including: the image feature matrix output by the convolutional neural network model Perform deconvolution operation to obtain the prior feature matrix in the gradient direction, that is, .

[0036] Next, the gradient response-based differential field quantity distribution is derived based on the image feature matrix and the prior feature matrix to obtain the gradient inverse response matrix, namely: ;in, represents the image feature matrix, represents the prior feature matrix, represents element-wise subtraction, Represents the image feature matrix The inverse matrix of represents matrix multiplication, represents the gradient inverse response matrix.

[0037] That is, the differential field in the nonlinear tensor space is approximated by the gradient inverse response of the image feature matrix.

[0038] Then, based on the image feature matrix and the prior feature matrix, the hyperspace superposition field of the gradient is calculated to obtain the hyperspace anchor point matrix, that is; ;in, represents the transpose symbol, represents the hyperspace anchor matrix.

[0039] Then, the hyperspace anchor matrix is ​​used as the hyperspace anchor to normalize the constraint field intensity center convergence to obtain the normalized field intensity matrix, that is, ;in, represents the gauge field strength matrix.

[0040] Then, the gradient field and the gauge field are superimposed by using the hyperparameter as the weighting coefficient, that is, the gradient inverse response matrix and the gauge field intensity matrix are fused by using the hyperparameter as the weighting coefficient to obtain the optimized image feature matrix, that is, ,in, and represents a learnable hyperparameter, represents element-wise multiplication, represents element-wise addition, By representing the optimized image feature matrix, the accuracy of derivation of spatial linear-nonlinear characteristics of nonlinear differential manifolds based on tensor feature correlation gradients and dynamic composite manifold field intensity convergence mapping can be improved by superimposing geometric transformation representations, that is, the coordinate system space field quantity transfer learning effect of the convolutional neural network is enhanced, thereby improving the detection accuracy of key reference feature points in coordinate system transformation and coordinate deviation transformation scenarios.

[0041] Specifically, in step S4, based on the coordinate system calibration parameters of the camera and machine motion system, the key reference feature points are converted from the pixel coordinate system to the world coordinate system to obtain a set of key reference feature point positions in the world coordinate system. It should be understood that the image captured by an industrial camera is essentially a two-dimensional pixel array, where the positions of key reference feature points are described in units of pixels. These pixel coordinates are relative to the camera image sensor and do not directly carry any absolute information about the workpiece's size or position in real physical space. Whether a feature occupying 10 pixels in the image corresponds to 1 mm or 1 cm in the physical world depends entirely on the camera's optical settings, distance from the workpiece, and lens parameters. On the other hand, the machine motion system operates in a well-defined physical space, namely the world coordinate system (typically a three-dimensional or two-dimensional Cartesian coordinate system fixed to the worktable). Dispensing drive instructions, such as moving the dispensing head to a certain position, must be specified in physical units in the world coordinate system.

[0042] Therefore, if we only stay at the pixel coordinate level, we will not be able to understand the actual position of the workpiece in the physical world, let alone calculate the actual deviation between it and the ideal physical position. Without this deviation information expressed in physical units, it is impossible to generate correct and effective compensation instructions to adjust the motion path of the dispensing head. Based on this, in this application, the key reference feature points are further converted from the pixel coordinate system to the world coordinate system based on the coordinate system calibration parameters of the camera and the machine motion system to obtain the position set of the key reference feature points in the world coordinate system. Ordinary technicians in this field should know that the calibration between the coordinate systems of the camera and the machine motion system is a routine process, wherein the purpose of the calibration between the camera and the machine motion system is to determine the coordinate system calibration parameters of the camera and the machine motion system.

[0043] In one embodiment, the key reference feature points can be transformed from the pixel coordinate system to the world coordinate system through a plane homography transformation. Specifically, there is a direct mapping relationship between the pixel coordinates (u, v) and the world coordinates (Xw, Yw) on the plane via a 3x3 homography matrix H: ; where s is a scale factor, represents transpose, Represents the homography matrix, which can be obtained by collecting several known world coordinates on the calibration plate The points and their corresponding image pixel coordinates , and then solve it by an optimization algorithm such as the least square method. Of course, this is just an example, and the key reference feature points can also be converted from the pixel coordinate system to the world coordinate system by other geometric transformation methods.

[0044] Specifically, in step S5, the set of positions of the key reference feature points in the world coordinate system is compared with the set of ideal positions of the key reference feature points in the world coordinate system to obtain deviation transformation parameters. It should be understood that by comparing the actual set of positions of the key reference feature points measured in the world coordinate system with the set of world coordinate positions where the key reference feature points should be when the workpiece is in an absolutely ideal state, the actual deviation difference can be accurately perceived.

[0045] In one embodiment, Figure 3As shown, the position set of the key reference feature points in the world coordinate system is compared with the ideal position set of the key reference feature points in the world coordinate system to obtain deviation transformation parameters, including: S51: calculating the optimal 2D rigid body transformation matrix between the position set of the key reference feature points in the world coordinate system and the ideal position set of the key reference feature points in the world coordinate system, the optimal 2D rigid body transformation matrix including X-direction translation parameters, Y-direction translation parameters and rotation parameters around the Z axis; S52: extracting the X-direction translation parameters, Y-direction translation parameters and rotation parameters around the Z axis from the optimal 2D rigid body transformation matrix as the deviation transformation parameters.

[0046] Specifically, after the industrial computer obtains the actual positions of key reference feature points in the world coordinate system, it also requires a benchmark: the ideal positions of these key reference feature points in the world coordinate system when the workpiece is in its ideal position. This ideal position set is predefined and stored in the system, typically during workpiece design or initial process programming. For example, this can be directly extracted from the CAD model, or during teach-in programming, the workpiece is placed in an absolute standard position, and the world coordinates of these reference points are then recorded using a manually guided robot or vision system.

[0047] Next, the position set of the key reference feature points in the world coordinate system is compared with the ideal position set of the key reference feature points in the world coordinate system. The process of comparing the two position sets is of course not a simple point-by-point coordinate difference, because the deviation of the workpiece is usually a combination of translation and rotation, rather than independent random movement of each point. Therefore, in this application, by finding an optimal 2D rigid body transformation matrix, the transformation can transform the ideal position set as a whole to a state closest to the actual position set. The so-called 2D rigid body transformation means that the relative distance and angle between the points inside the object remain unchanged during the transformation process, and only involves translation and rotation within the plane. This is consistent with the actual situation of workpiece deviation in most planar dispensing applications. Once this transformation matrix is ​​obtained, it accurately describes the overall geometric transformation from the ideal state to the current actual state, and the translation parameters and rotation parameters it contains are the deviation transformation parameters required.

[0048] In one embodiment, a rigid body transformation solution method for a two-dimensional point set is used to calculate the optimal 2D rigid body transformation matrix between the position set of the key reference feature points in the world coordinate system and the ideal position set of the key reference feature points in the world coordinate system. Ordinary technicians in this field should know that the rigid body transformation solution method for a two-dimensional point set is a conventional process, so it will not be traced back.

[0049] Specifically, in step S6, a dispensing drive instruction is generated based on the deviation transformation parameters. It should be understood that the fully automatic dispensing machine performs tasks according to a series of pre-programmed or taught path points (including position and posture information, as well as dispensing action instructions). This preset path is designed based on the workpiece being in an ideal position. If this preset path is used directly and the actual position of the workpiece deviates, the dispensing will deviate from the target. Therefore, based on the deviation transformation parameters, this preset standard dispensing path is dynamically adjusted in real time to generate a new actual dispensing path that can accurately match the current actual position of the workpiece. Based on this corrected actual dispensing path, a truly effective dispensing drive instruction is generated, guiding the dispensing head to complete the dispensing action (such as dispensing, marking, filling, etc.) at the correct position on the workpiece, while precisely controlling the start and stop of the dispensing valve.

[0050] In one embodiment, Figure 4 As shown, based on the deviation transformation parameters, a dispensing drive instruction is generated, including: S61: obtaining preset dispensing path data; S62: inputting the deviation transformation parameters and the preset dispensing path data into a motion controller, and the motion controller fine-tunes the preset dispensing path data based on the deviation transformation parameters to obtain dispensing path adjustment data; S63: generating the dispensing drive instruction based on the dispensing path adjustment data.

[0051] Specifically, the pre-set dispensing path data is first acquired. This acquisition can be accomplished through offline programming, teach-through programming, and vision-assisted teaching. Offline programming utilizes CAD / CAM software to plan the dispensing path directly on a digital 3D model of the workpiece. The software generates a series of path points with world coordinates (X, Y, Z, and possible poses A, B, C), along with dispensing instructions (such as valve opening, valve closing, and glue volume control) to be executed at or between these points. This data is then downloaded to the controller of the fully automatic dispensing machine. With teach-through programming, the operator manually guides the robotic dispensing head to key dispensing locations on the workpiece using a handheld teach pendant, recording the world coordinates of these points and the corresponding dispensing movements. This method is intuitive, but accuracy and efficiency may be affected by human error. Vision-assisted teaching, in conjunction with a vision system, allows the operator to click on the target dispensing location on a camera image. The system then automatically converts the pixel coordinates to world coordinates and records them.

[0052] The standard dispensing path data is then transformed based on the deviation transformation parameters to obtain the actual dispensing path data. Specifically, the deviation transformation parameters calculated by the industrial computer through the previous visual processing process are applied to each path point in the standard dispensing path to generate a new, compensated actual dispensing path. This transformation process is based on the mathematical principles of rigid body transformation. For any ideal path point in the standard dispensing path, its corresponding point at the current actual workpiece position must be calculated. The specific transformation process is as follows: First, if the rotation is performed about the center of mass of the ideal point set, the ideal point must be translated to a coordinate system with the center of mass as the origin, rotated, translated back, and finally the overall translation deviation is applied. However, a more direct approach is to define the deviation parameters (ΔX, ΔY, Δθ) as the parameters that directly transform the ideal coordinate system to the actual coordinate system (that is, first rotate Δθ about the origin, then translate ΔX and ΔY to obtain the actual coordinates). The industrial computer needs to traverse all path points with defined spatial positions in the standard dispensing path and apply the above transformation formula to the coordinates of each point to obtain the new coordinates. If the path also contains posture information (for example, the orientation of the dispensing needle), and the deviation transformation also includes posture deviation (for example, if the workpiece not only translates and rotates, but also tilts), then the posture information must also be transformed accordingly. For 2D plane dispensing, only XY translation and rotation around the Z axis are usually considered, and the dispensing head posture (perpendicular to the plane) remains unchanged. By performing this transformation on all relevant points on the standard path, a set of actual dispensing path data is generated.

[0053] Finally, based on the actual dispensing path data, dispensing drive instructions for controlling the movement of the dispensing head and the opening and closing of the dispensing valve are generated. That is, the industrial computer converts the actual dispensing path data (including world coordinates, motion type, dispensing action, etc.) into dispensing drive instructions that the controller can directly understand and execute. In a specific embodiment, the dispensing drive instructions are generated based on the dispensing path adjustment data, including: first, path interpolation. That is, the controller needs to use a specific interpolation algorithm (such as linear interpolation, circular interpolation, spline interpolation) to generate a series of dense intermediate target position points from the discrete path points on the actual dispensing path (such as the transformed start and end points) to ensure that the dispensing head can move smoothly and continuously along the corrected trajectory. Then, speed and acceleration planning. To ensure smooth and efficient movement, the controller will plan corresponding speed and acceleration curves (such as trapezoidal or S-shaped curves) for the interpolated path points based on the set maximum speed, acceleration, and other parameters. Next, I / O control instructions are generated. This involves generating corresponding digital output (DO) or analog output (AO) instructions based on the dispensing action information (such as when to open and close the valve, and glue volume control parameters) contained in the actual dispensing path data. For example, when the dispensing head moves to a certain point on the actual path, the controller sends a DO signal to the dispensing valve controller, causing it to open the valve and begin dispensing glue. When it reaches another point, it sends another signal to close the valve. If glue volume control is required, analog outputs may also be used to control dispensing pressure or time. Finally, instructions are serialized and sent. This means that all calculated joint target values, velocities, accelerations, and I / O control signals are organized in chronological order into a command sequence executable by the robot controller. These instructions ultimately drive the robot's servo motors and dispensing valve actuators.

[0054] In summary, the control method of the fully automatic dispensing machine provided by the present application is that after the dispensing task is started, the industrial camera collects an image containing a specific reference feature area of ​​the workpiece in real time; the industrial computer then detects and locates the key reference feature points from the image and obtains their pixel coordinates. Based on the pre-calibrated coordinate system conversion relationship between the camera and the machine motion system, the feature points are converted into actual positions in the world coordinates. By accurately comparing this actual position with the preset ideal position, the comprehensive deviation transformation parameters of the workpiece are calculated. Finally, based on this deviation parameter, the preset dispensing path is dynamically adjusted, and a dispensing drive instruction is generated to guide the dispensing head to the target position accurately, thereby effectively overcoming the influence of factors such as workpiece incoming material differences and positioning errors, and significantly improving the accuracy, stability and automation level of the dispensing operation.

[0055] The present application also provides a control system for a fully automatic glue dispensing machine, which is used to implement the control method of the fully automatic glue dispensing machine, such as Figure 5As shown, the control system 500 of the fully automatic dispensing machine includes: a dispensing start trigger control module 501, which is used to send a camera trigger signal after detecting the dispensing task start instruction; an image acquisition and transmission module 502, which is used for the industrial camera to acquire the original image data containing the workpiece reference feature area according to the camera trigger signal, and transmit the original image data containing the workpiece reference feature area to the industrial computer; an image feature point detection and positioning module 503, which is used for the industrial computer to detect and locate the key reference feature points in the original image data containing the workpiece reference feature area to obtain the key reference feature points. A position set in pixel coordinates; a coordinate system conversion processing module 504, used to convert the key reference feature points from the pixel coordinate system to the world coordinate system based on the coordinate system calibration parameters of the camera and the machine motion system to obtain the position set of the key reference feature points in the world coordinate system; a feature point position comparison module 505, used to compare the position set of the key reference feature points in the world coordinate system with the ideal position set of the key reference feature points in the world coordinate system to obtain deviation transformation parameters; a dispensing drive instruction generation module 506, used to generate a dispensing drive instruction based on the deviation transformation parameters.

[0056] The basic principles of the present application have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, and effects mentioned in this application are merely illustrative and not restrictive, and it should not be assumed that these advantages, strengths, and effects are required of each embodiment of this application. In addition, the specific details disclosed above are merely illustrative and facilitating understanding, and are not restrictive. The above details do not limit this application to necessarily being implemented using the above specific details.

[0057] The flowcharts of the methods involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the flowcharts. As will be appreciated by those skilled in the art, these devices, apparatuses, equipment, and systems may be connected, arranged, or configured in any manner. Words such as "include," "comprise," "have," and the like are open-ended words, meaning "including but not limited to," and may be used interchangeably therewith. The words "or" and "and" used herein refer to the words "and / or" and may be used interchangeably therewith, unless the context clearly indicates otherwise. The word "such as" used herein refers to the phrase "such as but not limited to," and may be used interchangeably therewith.

[0058] It should also be noted that in the method of the present application, each step can be decomposed and / or recombined. Such decomposition and / or recombination should be regarded as equivalent solutions of the present application.

[0059] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present application. Therefore, the present application is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0060] The above description has been provided for the purpose of illustration and description. Furthermore, this description is not intended to limit the embodiments of the present application to the forms disclosed herein. Although a number of example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A control method for a fully automatic dispensing machine, characterized in that: include: After detecting the dispensing task start instruction, a camera trigger signal is issued; The industrial camera collects original image data containing the reference feature area of ​​the workpiece according to the camera trigger signal, and transmits the original image data containing the reference feature area of ​​the workpiece to the industrial computer; The industrial computer detects and locates key reference feature points in the original image data containing the workpiece reference feature area to obtain a position set of the key reference feature points in pixel coordinates, including: performing distortion correction, image noise reduction and normalization processing on the original image data containing the workpiece reference feature area to obtain a preprocessed image; using a convolutional neural network model to extract image features of the preprocessed image to obtain an image feature matrix; performing convergence optimization based on tensor feature association gradient and dynamic tensor field on the image feature matrix to obtain an optimized image feature matrix; inputting the optimized image feature matrix into a softmax function to determine the probability value of each pixel as a key point; and determining the position set of the key reference feature points in pixel coordinates based on the probability value of each pixel as a key point. Based on the coordinate system calibration parameters of the camera and the machine motion system, the key reference feature points are converted from the pixel coordinate system to the world coordinate system to obtain a position set of the key reference feature points in the world coordinate system; Comparing the position set of the key reference feature points in the world coordinate system with the ideal position set of the key reference feature points in the world coordinate system to obtain deviation transformation parameters; generating a dispensing drive instruction based on the deviation transformation parameters; The image feature matrix is ​​subjected to convergence optimization based on tensor feature associated gradient and dynamic tensor field to obtain an optimized image feature matrix, including: performing deconvolution operation on the image feature matrix to obtain a priori feature matrix in the gradient direction; deriving the differential field quantity distribution based on gradient response based on the image feature matrix and the priori feature matrix to obtain a gradient inverse response matrix; calculating the hyperspace superposition field of the gradient based on the image feature matrix and the priori feature matrix to obtain a hyperspace anchor matrix; using the hyperspace anchor matrix as a hyperspace anchor to normalize the constraint field intensity center convergence to obtain a canonical field intensity matrix; and using hyperparameters as weighting coefficients to fuse the gradient inverse response matrix and the canonical field intensity matrix to obtain the optimized image feature matrix.

2. The control method of the fully automatic dispensing machine according to claim 1, characterized in that: The original image data containing the workpiece reference feature area is transmitted to the industrial computer via a high-speed interface.

3. The control method of the fully automatic dispensing machine according to claim 2, characterized in that: The trained deep learning model is a key point detection model based on convolutional neural network.

4. The control method of the fully automatic dispensing machine according to claim 1, characterized in that: Comparing the position set of the key reference feature points in the world coordinate system with the ideal position set of the key reference feature points in the world coordinate system to obtain deviation transformation parameters, including: calculating the optimal 2D rigid body transformation matrix between the position set of the key reference feature points in the world coordinate system and the ideal position set of the key reference feature points in the world coordinate system, the optimal 2D rigid body transformation matrix including X-direction translation parameters, Y-direction translation parameters and rotation parameters around the Z axis; extracting the X-direction translation parameters, Y-direction translation parameters and rotation parameters around the Z axis from the optimal 2D rigid body transformation matrix as the deviation transformation parameters.

5. The control method of the fully automatic dispensing machine according to claim 4, characterized in that: Based on the deviation transformation parameters, a dispensing drive instruction is generated, including: obtaining preset dispensing path data; inputting the deviation transformation parameters and the preset dispensing path data into a motion controller, and the motion controller fine-tuning the preset dispensing path data based on the deviation transformation parameters to obtain dispensing path adjustment data; based on the dispensing path adjustment data, the dispensing drive instruction is generated.

6. A control system for a fully automatic glue dispensing machine, used to execute the control method for the fully automatic glue dispensing machine according to claim 1, characterized in that: include: The dispensing start trigger control module is used to send a camera trigger signal after detecting the dispensing task start instruction; An image acquisition and transmission module is used for the industrial camera to acquire original image data containing the workpiece reference feature area according to the camera trigger signal, and transmit the original image data containing the workpiece reference feature area to the industrial computer; an image feature point detection and positioning module is used for the industrial computer to detect and locate key reference feature points in the original image data containing the workpiece reference feature area to obtain a position set of the key reference feature points in pixel coordinates; A coordinate system conversion processing module is used to convert the key reference feature points from the pixel coordinate system to the world coordinate system based on the coordinate system calibration parameters of the camera and the machine motion system to obtain a position set of the key reference feature points in the world coordinate system; A feature point position comparison module is used to compare the position set of the key reference feature points in the world coordinate system with the ideal position set of the key reference feature points in the world coordinate system to obtain deviation transformation parameters; a dispensing drive instruction generation module is used to generate a dispensing drive instruction based on the deviation transformation parameters.

Images