Dispensing machine anti-piling motion control method

Abstract

The application discloses a motion control method for preventing glue stacking of a glue dispenser, and relates to the technical field of intelligent control.The method comprises the following steps: constructing a non-Newtonian fluid dynamic model through a power law equation, and generating a four-dimensional compensation coefficient matrix; predicting glue amount deviation through a convolutional neural network, calculating compensation glue amount based on linkage of a piezoelectric valve and a screw valve, obtaining a control instruction set; constructing a Jacobian matrix to calculate an end equivalent inertia force, and injecting a reverse compensation current through a servo driver to synchronously optimize a multi-axis motion trajectory, and re-planning a path through a curvature radius constraint at an inflection point; detecting glue line height through laser height measurement, triggering an online learning algorithm to update convolutional neural network weights, switching process parameter files, and controlling synchronization of a conveying belt beat and a glue dispensing gap time; monitoring glue line height, executing an abnormal glue stacking classification processing strategy, and recording fault data packets.The application realizes glue stacking prevention of the glue dispenser through lightweight convolutional neural network prediction and multi-axis motion trajectory optimization.

Description

A motion control method for preventing glue piling in a dispensing machine Technical Field

[0001] This invention relates to the field of intelligent control technology, and in particular to a motion control method for preventing glue buildup in a dispensing machine. Background Technology

[0002] In electronics manufacturing, precision assembly, and other fields, dispensing machines are widely used for precision coating and sealing operations. Current motion control methods for dispensing machines are typically based on traditional Newtonian fluid models, combining preset parameters and fixed trajectories for adhesive volume control. However, the adhesives used in practical applications (such as UV-cured adhesives and epoxy resins) are mostly non-Newtonian fluids, whose viscosity and flow characteristics dynamically change with operating conditions such as shear rate, temperature, and pressure. Existing technologies often rely on empirical formulas or static models for parameter compensation, lacking the ability to model the unsteady rheological properties of non-Newtonian fluids in real time. This results in a certain lag in the matching of adhesive volume to the motion trajectory during high-speed motion or complex trajectory operations.

[0003] This lag makes adhesive buildup (piling up) a common problem during dispensing, especially at inflection points, acceleration / deceleration phases, or when environmental parameters fluctuate. The adhesive, due to structural damage and inertial forces, cannot be compensated for in time, leading to uneven adhesive line height, overflow, or adhesive breakage, severely impacting product yield and production efficiency. Existing solutions largely rely on post-processing inspection or manual intervention, lacking proactive prediction and closed-loop control of dynamic fluid characteristics and motion inertia, making it difficult to achieve high-precision and high-stability anti-piling effects. Summary of the Invention

[0004] In view of the above-mentioned existing problems, the present invention provides a motion control method for preventing glue buildup in a dispensing machine to solve the problem of glue accumulation in the dispensing machine.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] In a first aspect, the present invention provides a motion control method for preventing glue piling in a dispensing machine, which includes: constructing a non-Newtonian fluid dynamic model based on the motion parameters of the dispensing machine and the physical property parameters of the glue through a power-law equation, generating a four-dimensional compensation coefficient matrix, and forming a set of reference parameters;

[0007] The system uses multiple sensors to collect operating condition data, packages it into a time-series data stream, and transmits it to an embedded FPGA for normalization processing to generate standardized feature vectors.

[0008] The glue quantity deviation is predicted by a convolutional neural network, and the glue quantity is compensated by the linkage between the piezoelectric valve and the screw valve to obtain the control command set.

[0009] The equivalent inertial force at the end is calculated by constructing the Jacobian matrix, and a reverse compensation current is injected through the servo driver to synchronously optimize the multi-axis motion trajectory. At the inflection point, the path is replanned by the radius of curvature constraint, and the compensation motor control command is output.

[0010] The height of the glue line is detected by laser height measurement, which triggers the online learning algorithm to update the weights of the convolutional neural network, switches the process parameter file, and controls the synchronization of the conveyor belt cycle and the glue dispensing interval time.

[0011] Monitor the glue line height, implement abnormal glue stacking classification and handling strategies, and record fault data packets.

[0012] As a preferred embodiment of the motion control method for preventing glue buildup in the dispensing machine described in this invention, the physical properties of the glue include the shear stress characteristics of the UV-curing adhesive and the epoxy resin adhesive, and the four-dimensional compensation coefficient matrix includes velocity, acceleration, temperature and air pressure.

[0013] As a preferred embodiment of the motion control method for preventing glue buildup in the dispensing machine described in this invention, the convolutional neural network is a lightweight residual network, the input layer contains time-series data of a four-dimensional compensation coefficient matrix, the hidden layer uses a depthwise separable convolutional kernel, and the output layer generates deviation weights through a sigmoid activation function; the online learning algorithm uses a stochastic gradient descent method with a driving factor, and the update cycle is determined by a fixed time and acceleration.

[0014] As a preferred embodiment of the motion control method for preventing glue buildup in the dispensing machine described in this invention, the optimized multi-axis motion trajectory includes cubic spline interpolation path smoothing and dynamic curvature radius adjustment, and the insertion of transition fillets according to the corner angle.

[0015] As a preferred embodiment of the motion control method for preventing glue buildup in the dispensing machine described in this invention, the switching process parameter file refers to sending equipment status codes and allocating dispensing areas for piezoelectric valves and screw valves during multi-station collaborative operation.

[0016] As a preferred embodiment of the motion control method for preventing glue buildup in the dispensing machine described in this invention, the non-Newtonian fluid dynamic model adds a thixotropic differential term to the power-law model, incorporates the glue structure destruction effect caused by acceleration into the non-Newtonian fluid dynamic model, and performs unsteady fluid control. The parameters of the non-Newtonian fluid dynamic model are calibrated and updated online through feedback from laser height measurement and glue pressure sensor.

[0017] As a preferred embodiment of the motion control method for preventing glue buildup in the dispensing machine described in this invention, the abnormal glue buildup graded processing strategy is divided into primary processing and secondary processing. A first threshold and a second threshold are set, where the first threshold is less than the second threshold. When the glue line height is greater than the first threshold but less than or equal to the second threshold, primary processing is used; when the glue line height is greater than the second threshold, secondary processing is used; and when the glue line height is less than or equal to the first threshold, no processing is performed. The primary processing reduces the motion speed and initiates vacuum glue suction. The secondary processing triggers an emergency stop protocol and generates a fault data packet.

[0018] As a preferred embodiment of the motion control method for preventing glue buildup in the dispensing machine described in this invention, the convolutional neural network is deployed on an edge computing unit and achieves clock synchronization and jitter control with the motion controller through a communication bus.

[0019] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the motion control method for preventing glue piling in a dispensing machine as described in the first aspect of the present invention.

[0020] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the motion control method for preventing glue piling in a dispensing machine as described in the first aspect of the present invention.

[0021] The beneficial effects of this invention are as follows: By combining a non-Newtonian fluid dynamic model with a four-dimensional compensation coefficient matrix, the rheological properties of the adhesive are accurately characterized and parameters are calibrated in real time; a lightweight convolutional neural network is used to predict adhesive volume deviation, and edge computing and online learning algorithms are combined to achieve dynamic optimization of control commands; the smoothness and synchronization of multi-axis motion trajectories are optimized through Jacobian matrix calculation and servo drive compensation, and the curvature radius constraint and cubic spline interpolation are used to effectively suppress corner adhesive buildup; at the same time, based on laser height measurement and graded anomaly handling strategies, real-time monitoring and graded response to adhesive buildup risks are achieved, improving the accuracy of adhesive line height control and production stability. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 is an overall architecture diagram of the motion control method for preventing glue piling in the dispensing machine in Example 1.

[0024] Figure 2 is a flowchart of the non-Newtonian fluid model construction process in Example 1.

[0025] Figure 3 is a flowchart of the prediction and control instruction generation process of the convolutional neural network in Example 1.

[0026] Figure 4 is a flowchart of motion optimization and anomaly handling in Example 1. Detailed Implementation

[0027] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0028] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0029] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0030] Example 1, referring to Figures 1-4, is the first embodiment of the present invention. This embodiment provides a motion control method for preventing glue piling in a dispensing machine, including the following steps:

[0031] S1: Based on the motion parameters of the dispensing machine and the physical properties of the adhesive, a non-Newtonian fluid dynamic model is constructed through the power-law equation, generating a four-dimensional compensation coefficient matrix and forming a set of reference parameters.

[0032] Specifically, it includes the following steps:

[0033] S1.1: Collect motion parameters of the dispensing machine, physical properties of the adhesive, and environmental parameters.

[0034] Specifically, the motion parameters of a dispensing machine include speed and acceleration.

[0035] Speed ​​refers to the instantaneous movement speed of the dispensing head in the XYZ three-axis direction (unit: mm / s), which is fed back in real time by the servo motor encoder.

[0036] Acceleration refers to the instantaneous acceleration of the dispensing head (unit: mm / s²). 2 The speed signal is obtained through differential calculation.

[0037] The physical properties of adhesives include consistency coefficient, flow index, and thixotropic coefficient.

[0038] Consistency coefficient refers to the material constant in the power-law equation, characterizing the viscosity of the adhesive (unit: Pa·s). n The result was obtained by fitting the measured data of the electronic scale for UV glue / epoxy resin.

[0039] The flow index is a non-Newtonian fluid characteristic parameter; a value less than 1 indicates shear thinning, and a value greater than 1 indicates shear thickening.

[0040] Thixotropic coefficient describes the response characteristics of adhesive structure to acceleration failure (unit: Pa·s). 2 The speed is dynamically adjusted based on the acceleration of the linear motor.

[0041] Environmental parameters include temperature and air pressure.

[0042] Temperature refers to the real-time temperature (unit: °C) near the dispensing head, which is collected by a thermocouple sensor.

[0043] Air pressure refers to the driving air pressure of the dispensing valve (unit: kPa), which is measured by an SMC pressure sensor.

[0044] S1.2: Construct a dynamic model of a non-Newtonian fluid.

[0045] Specifically, based on the traditional power-law model, a thixotropic differential term is added to characterize the effect of acceleration on the rheological properties of the glue. The expression for the non-Newtonian fluid dynamic model is as follows:

[0046]

[0047] Where τ is the shear stress, which determines the flow resistance of the adhesive, and K is the consistency coefficient. For shear rate, (v is the velocity, h is the gap between the dispensing needle and the substrate, which is fixed at 0.1 mm), n is the flow index, η is the thixotropic coefficient, d is the differential sign, and t is the time window length.

[0048] Furthermore, rheological tests were performed on each type of adhesive (UV adhesive / epoxy resin), and K and n were fitted using the least squares method. Every 5 minutes, the adhesive line height and actual adhesive pressure fed back by the laser height measurement unit and adhesive pressure sensor were used to inversely correct K and n.

[0049] S1.3: Generate a four-dimensional compensation coefficient matrix.

[0050] It should be understood that the four-dimensional compensation coefficient matrix includes velocity, acceleration, temperature, and air pressure.

[0051] S1.4: Form a set of reference parameters.

[0052] Specifically, the calibrated K, n, η and the four-dimensional compensation coefficient matrix are stored in the non-volatile memory (NAND Flash) of the dispensing machine controller to form a reference parameter set.

[0053] Ideally, a non-Newtonian fluid dynamic model is established by combining power-law equations with thixotropic effects to accurately describe the rheological properties of the adhesive under different velocity and acceleration conditions. Real-time data collection is performed on the dispensing head motion state, adhesive consistency coefficient, flow index, and ambient temperature and pressure. Based on laser height measurement and adhesive pressure feedback, the rheological parameters are dynamically corrected. A four-dimensional compensation coefficient matrix is ​​generated, using velocity, acceleration, temperature, and pressure as variables to optimize the dispensing flow compensation strategy. The baseline parameter set is stored in the dispensing machine controller, enabling precise control and environmental adaptation. This effectively improves dispensing consistency, reduces process errors, and is suitable for high-precision electronic packaging and micro-assembly manufacturing, improving production yield and stability.

[0054] S2: Use multi-source sensors to collect operating condition data, package it into a time-series data stream, and transmit it to the embedded FPGA for normalization processing to generate standardized feature vectors.

[0055] Specifically, the steps include: S2.1: Collect operating condition data using multi-source sensors.

[0056] Specifically, the operating data includes the real-time position of the XYZ axes, end-effector jitter data, speed, acceleration, adhesive pressure, and adhesive temperature.

[0057] The end-point jitter data was collected by configuring a plasma station IMU sensor (±16g accelerometer).

[0058] It should be noted that the multi-source sensors receive the same external clock signal (frequency 1kHz) through the FPGA's GPIO pins to achieve timestamp alignment.

[0059] S2.2: Package the operating condition data into a time-series data stream.

[0060] Specifically, the operating data is encapsulated into data frames within each sampling period (1ms), and the FPGA's built-in dual-port RAM (capacity 8KB) is used to buffer 800 consecutive data frames (800 frames), which are managed in a FIFO queue.

[0061] S2.3: Data processing is performed on the embedded FPGA.

[0062] Specifically, the Hampel filter is used to remove outliers with a window length of 50ms (50 frames) and the outlier removal threshold is set to 3 times the median absolute deviation (MAD).

[0063] Wavelet thresholding denoising was performed on the acceleration and adhesive compression signals (Daubechies4 wavelet basis, soft thresholding).

[0064] Z-score standardization was used for laser altimetry data, while piecewise linear normalization was used for dynamic range compression of other data.

[0065] S2.4: Extract and generate standardized feature vectors.

[0066] Specifically, the mean, variance, and peak factor of the statistical working condition data are used to extract time-domain features.

[0067] Furthermore, an FFT (256-point, Hanning window) is performed on the acceleration signal to extract the energy percentage of the top ten harmonic components as frequency domain features.

[0068] The time-domain features and frequency-domain features are integrated into a standardized feature vector.

[0069] Ideally, key parameters such as XYZ axis position, end effector jitter, speed, acceleration, adhesive pressure, and adhesive temperature are acquired through an IMU, high-precision encoder, and pressure sensor, with time alignment ensured by an FPGA-synchronized clock signal. Sampled data is encapsulated as a time-series data stream with a 1ms cycle and stored in a dual-port RAM buffer. Outliers are removed using Hampel filtering, and wavelet denoising is applied to the adhesive pressure and acceleration signals. Z-score normalization and piecewise linear normalization are employed to improve data consistency. Finally, time-domain and frequency-domain features are extracted through time-domain statistics and frequency-domain analysis. This effectively improves the accuracy of status monitoring during the dispensing process, enhances adaptability to complex working conditions, and provides efficient data support for subsequent intelligent control and quality optimization.

[0070] S3: Predict glue quantity deviation through convolutional neural network, calculate and compensate glue quantity based on the linkage of piezoelectric valve and screw valve, and obtain control command set.

[0071] Specifically, it includes the following steps:

[0072] S3.1: Construct a lightweight residual convolutional neural network.

[0073] Specifically, a lightweight residual convolutional neural network model is constructed using a convolutional neural network (CNN) architecture.

[0074] The standardized feature vectors are segmented by a sliding window with a window length of 50ms and a step size of 1ms to form a time series input. The input dimension is the time series window length (fixed at 50 frames, corresponding to 50ms of historical data) multiplied by the feature dimension (time series data containing the four-dimensional compensation coefficient matrix).

[0075] The hidden layers consist of depthwise separable convolutional layers and global average pooling layers. The first depthwise separable convolutional layer has a 3×1 kernel size (time-axis convolution, not crossing feature dimensions), 32 output channels, and uses ReLU activation. The second convolutional layer has a 3×1 kernel size (feature-axis convolution, not crossing the time axis), 64 output channels, and also uses ReLU activation. Skip connections are introduced after every two convolutional layers, using 1×1 convolutions to adjust the number of channels to match the dimensionality and suppress gradient vanishing. The global average pooling layer compresses the time dimension to 1, resulting in an output dimension of 64.

[0076] The fully connected layer maps 64-dimensional features to a 3-dimensional output, including piezoelectric valve compensation, screw valve compensation, and overall deviation weights. The piezoelectric valve compensation and screw valve compensation use the Sigmoid activation function, with an output range of 0 to 1, corresponding to the normalized compensation of the valve opening. The overall deviation weights are weighted using the Softmax activation function, ensuring that the sum of the weights equals 1.

[0077] S3.2: Perform glue quantity deviation prediction and compensation calculation.

[0078] Specifically, the expression for the deviation between the predicted theoretical amount of adhesive and the actual amount of adhesive is:

[0079]

[0080] Where ΔQ is the glue quantity deviation, ρ is the glue density, r is the inner diameter of the dispensing needle, v is the dispensing head movement speed, Q1 is the piezoelectric valve compensation amount, Q2 is the screw valve compensation amount, and Q theory Q is the theoretical amount of adhesive. real This represents the actual amount of adhesive.

[0081] The predicted ΔQ is decomposed into piezoelectric valve compensation Q1 and screw valve compensation Q2, and allocated using a comprehensive deviation weighting method, expressed as:

[0082] Q1 = w1·ΔQ;

[0083] Q2 = w2·ΔQ;

[0084] Where w1 is the compensation weight of the piezoelectric valve and w2 is the compensation weight of the screw valve.

[0085] The piezoelectric valve compensation weight w1 is used for high-frequency response, prioritizing compensation for high-frequency disturbances; the screw valve compensation weight w2 is used for steady-state accuracy, compensating for low-frequency cumulative errors. The sum of w1 and w2 equals 1.

[0086] Furthermore, the expression for the piezoelectric valve compensation amount is:

[0087]

[0088] Where f1 is the piezoelectric valve driving frequency, d1 is the piezoelectric valve orifice diameter, P is the adhesive pressure, and P0 is the ambient air pressure.

[0089] The expression for the screw valve compensation amount is:

[0090]

[0091] Where n2 is the screw speed, D is the screw diameter, L is the screw pitch, η2 is the apparent viscosity of the adhesive, and η ref For viscosity calibration, ΔP is the pressure difference between the valve inlet and outlet, and U is the valve flow resistance coefficient.

[0092] S3.3: Generate control instruction set.

[0093] Specifically, based on the predicted piezoelectric valve compensation amount Q1 and screw valve compensation amount Q2, and combined with the expressions for piezoelectric valve compensation amount and screw valve compensation amount, the piezoelectric valve drive frequency f1 and screw speed n2 are calculated.

[0094] The piezoelectric valve drive frequency, screw speed, screw valve compensation weight, piezoelectric valve compensation weight, and timestamp are encapsulated into a control instruction set and sent to the piezoelectric valve controller (PZT Driver) and screw valve servo driver at a frequency of 1kHz via the EtherCAT bus.

[0095] S3.4: Set valve switching priority.

[0096] It should be noted that when the acceleration is greater than 2 (high acceleration disturbance), w1 is forced to be 1, and only the voltage valve is used for compensation. When the temperature is greater than 40°C (high temperature causes the adhesive viscosity to decrease), Q2 is limited to less than 0.8 × theoretical adhesive amount to avoid overflow.

[0097] A preferred lightweight residual convolutional neural network model employs a temporal window input, combining depthwise separable convolutions and skip connections to achieve efficient feature extraction. By predicting the deviation between theoretical and actual glue volume, the compensation amounts for the piezoelectric valve and screw valve are calculated, and a high-frequency and low-frequency error allocation strategy is used to optimize the compensation effect. The piezoelectric valve compensates for high-frequency disturbances, while the screw valve is used for steady-state accuracy correction. A control instruction set is generated by combining the screw valve compensation amount expressions, including the piezoelectric valve drive frequency, screw speed, and weight parameters, and is transmitted in real-time via the EtherCAT bus. Valve switching priorities are set, and the compensation strategy is dynamically adjusted under high acceleration disturbances or high-temperature environments. This method improves the accuracy and stability of the dispensing machine, reduces glue volume deviation, and enhances production consistency.

[0098] S4: Construct the Jacobian matrix to calculate the equivalent inertial force at the end, and inject reverse compensation current through the servo driver to synchronously optimize the multi-axis motion trajectory. At the inflection point, replan the path through the radius of curvature constraint and output the compensation motor control command.

[0099] Specifically, it includes the following steps:

[0100] S4.1: Construct the Jacobian matrix for the robotic arm.

[0101] Specifically, based on the geometric parameters of the multi-axis (XYZ axes) of the dispensing machine, a linear transformation matrix mapping the joint velocity to the end effector velocity is constructed, and the Jacobian matrix is ​​expressed as follows:

[0102]

[0103] Where J is the Jacobian matrix, x is the X-axis position of the end effector in the base coordinate system, y is the Y-axis position of the end effector in the base coordinate system, z is the Z-axis position of the end effector in the base coordinate system, q1 is the X-axis joint angle vector, q2 is the Y-axis joint angle vector, and q3 is the Z-axis joint angle vector.

[0104] The Jacobian matrix is ​​updated every 1ms based on the current joint angles q1, q2, and q3.

[0105] S4.2: Calculate the equivalent inertial force at the end.

[0106] Specifically, the equivalent inertial force caused by the movement of the robotic arm is calculated using the following expression:

[0107]

[0108] Where F is the equivalent inertial force, T is the matrix transpose operation, and M(a) is the mass matrix, representing the joint inertia. For joint acceleration, G(a) is the Coriolis force, and G(a) is the gravitational term.

[0109] S4.3: Injection of reverse current compensation.

[0110] Specifically, the equivalent inertial force is converted into joint torque, expressed as follows:

[0111] μ = J T F;

[0112] Where μ is the joint torque.

[0113] And based on the torque constant of the servo motor, the compensation current is calculated, and the expression is:

[0114]

[0115] Where I is the compensation current and H is the torque constant of the servo motor.

[0116] I is written to the servo driver's current loop at a frequency of 1kHz via the EtherCAT bus and superimposed on the original control signal. The compensation current is limited to no more than 30% of the motor's peak current to prevent overload.

[0117] S4.4: Optimize multi-axis motion trajectories.

[0118] Specifically, cubic spline interpolation is used for path smoothing. The discrete points P = [x, y, z] of the path are arranged in time series. Piecewise cubic polynomials are used for continuous acceleration constraints.

[0119] Furthermore, dynamic curvature radius adjustment is performed. Differential geometric analysis is conducted on the discrete points of the path to calculate the curvature, expressed as:

[0120]

[0121] Where κ is the curvature. Let P be the first derivative of the path, and let P be the second derivative of the path.

[0122] Set the minimum radius of curvature to 2mm. If 1 / κ is less than the minimum radius of curvature, insert a transition fillet.

[0123] S4.5: Perform corner path replanning.

[0124] It should be understood that when the angle between adjacent path segments is greater than 90°, it is considered a sharp turn. The center coordinates of the circle are determined based on the radius of the fillet, the inserted arc path is discretized into new path points, and a smooth trajectory is generated through cubic spline interpolation.

[0125] S4.6: Output compensation motor control command.

[0126] Specifically, the data protocol uses EtherCAT slave messages with a period of 1ms and jitter of less than 10μs.

[0127] The control cycles of each axis servo drive are aligned using the IEEE 1588 Precision Time Protocol (PTP). Fifth-order polynomial interpolation is employed to ensure continuous position, speed, and acceleration of the end effector.

[0128] Ideally, a Jacobian matrix is ​​constructed based on the robotic arm's geometric parameters and updated at a frequency of 1 kHz. The end-effector inertial force is calculated using the mass matrix, Coriolis force, and gravity term, and converted into joint torque. Compensation current is calculated using the servo motor torque constant and injected into the motor current loop in real time to improve trajectory following accuracy. Cubic spline interpolation is used to smooth the path, and the trajectory is optimized based on curvature radius constraints. Sharp turns are carefully planned with the insertion of circular transition segments to ensure trajectory smoothness. Compensation motor control commands are generated and transmitted via EtherCAT bus, ensuring a 1 ms control cycle synchronization and jitter less than 10 μs. This improves trajectory smoothness, reduces end-effector inertial error, and enhances dispensing accuracy and dynamic stability.

[0129] S5: The height of the glue line is detected by laser height measurement, which triggers the online learning algorithm to update the weights of the convolutional neural network, switches the process parameter file, and controls the synchronization of the conveyor belt cycle and the glue dispensing interval time.

[0130] Specifically, it includes the following steps:

[0131] S5.1: Perform laser height measurement and real-time detection of adhesive line height.

[0132] Specifically, a line-scanning laser triangulation height gauge (0.01mm resolution, 10kHz sampling frequency) is used, installed 5mm behind the dispensing head at a 45° angle to the dispensing path. The glue line height is acquired every 1ms and transmitted to the FPGA via an RS-422 interface.

[0133] A sliding window Hampel filter (window length 50ms, threshold 3 times MAD) is applied to remove outliers, and the reference height is adjusted in real time based on the substrate flatness. The theoretical adhesive line height is calculated by combining the theoretical adhesive amount, and the adhesive line height deviation is calculated by the difference between the actual adhesive line height and the theoretical adhesive line height.

[0134] S5.2: Trigger online learning and Convolutional Neural Network (CNN) weight updates based on the detection results.

[0135] Specifically, online learning is triggered when the deviation of the adhesive line height exceeds 0.05mm within 10 consecutive sampling periods, and the cumulative acceleration change exceeds 50m / s² every five minutes. 2 Force an update at any time.

[0136] The loss function expression for online learning algorithms is:

[0137]

[0138] Where L is the loss function of the online learning algorithm, R is the total number of samples, i is the sample index, and ΔH i The deviation in adhesive line height, ΔHpred,i λ represents the deviation of the glue line height predicted by the CNN, λ is the L2 regularization coefficient, and W is the CNN weight parameter.

[0139] The updated weights are written to the DDR4 memory of the edge computing unit via the PCIe interface to keep the CNN inference uninterrupted.

[0140] S5.3: Dynamically switch process parameter files.

[0141] It should be noted that the process parameter file includes rheological parameters such as consistency coefficient, flow index and thixotropic coefficient of UV adhesive and epoxy resin adhesive, as well as different operating conditions according to temperature-pressure combinations.

[0142] The system switches between piezoelectric valves and screw valves based on status codes generated according to the glue stack height and temperature. The dispensing areas of the piezoelectric valves and screw valves are assigned through the EtherCAT CoE (CANopen over EtherCAT) protocol. The piezoelectric valves are responsible for high-frequency, low-flow areas (such as corners), while the screw valves are responsible for high-flow straight sections.

[0143] Two sets of parameter files are preloaded in the FPGA, and the switching is completed within 1μs via hardware interrupt, achieving seamless switching.

[0144] S5.4: Synchronous control of conveyor belt cycle time.

[0145] It should be understood that after the dispensing head completes dispensing in the current area, time is required to allow for movement across areas and for the piezoelectric valve or screw valve to close completely. The conveyor belt's movement speed must ensure that, within the dispensing interval, its movement distance aligns with the position of the next dispensing area.

[0146] Synchronization is achieved through PID control. Error calculation is used to compare the difference between the actual cycle time and the theoretical cycle time in real time; proportional adjustment is then performed: if the current cycle time is too slow, the conveyor belt motor speed is increased proportionally; conversely, it is decelerated; historical errors are accumulated through integral adjustment to eliminate long-term deviations; and future error trends are predicted using derivative adjustment to adjust the speed in advance and avoid overshoot.

[0147] Based on the path planning module's prediction of the length and location of the next three dispensing areas, the conveyor belt speed is adjusted in advance. For example, if the spacing between subsequent dispensing areas increases, the conveyor belt is accelerated in advance to avoid the dispensing head waiting.

[0148] When multiple dispensing heads work together, the working area of ​​each dispensing head is dynamically allocated according to the conveyor belt speed and the distance between workstations to avoid overlapping areas or missed dispensing points. The interval time of all dispensing heads is uniformly based on the slowest movement to ensure that the conveyor belt speed meets the synchronization requirements of all workstations.

[0149] Ideally, a high-precision laser height gauge is used to detect the glue line height at a 1ms cycle, combined with Hampel filtering to eliminate outliers and dynamically adjust the reference height. When the glue line height deviation is detected to be continuously exceeding the limit, online learning is triggered to optimize the lightweight residual convolutional neural network model, ensuring stable dispensing accuracy. Based on the glue rheological properties and temperature-pressure conditions, process parameter files are dynamically switched to achieve seamless coordination between piezoelectric valves and screw valves, improving dispensing adaptability under complex working conditions. PID control is used to dynamically adjust the conveyor belt cycle time to ensure precise synchronization between the dispensing head and the conveying system. Through error calculation, advance prediction, and multi-dispensing head collaborative optimization, dispensing waiting time is reduced, and production efficiency is improved.

[0150] S6: Monitor glue line height, implement abnormal glue stacking classification handling strategy, and record fault data packets.

[0151] Specifically, it includes the following steps:

[0152] S6.1: Set the abnormal judgment threshold based on the theoretical glue line height.

[0153] Specifically, the first threshold S1 is set to the theoretical glue line height + 0.1mm (allowed process fluctuation range), and the second threshold S2 is set to the theoretical glue line height + 0.3mm (critical value for severe glue buildup).

[0154] If the deviation of the glue line height ΔH≤S1, it is a normal state and no processing is triggered.

[0155] If S1 < ΔH ≤ S2, trigger Level 1 processing.

[0156] If ΔH > S2, trigger secondary processing.

[0157] S6.2: Set a graded handling strategy for abnormal glue buildup.

[0158] Specifically, the first-level processing refers to slowing down the flow and vacuuming the adhesive.

[0159] The dispensing head speed is linearly reduced to 50% of its original speed, and the conveyor belt speed is reduced simultaneously to ensure that the dispensing head and the substrate are in synchronized relative positions. At the same time, the solenoid valve is controlled to start the vacuum pump for vacuum adhesive suction.

[0160] Level 2 processing refers to emergency stop and fault isolation.

[0161] The safety PLC sends an STO (Safe Torque Off) signal, cutting off the power to the servo motor and glue valve within 0.5ms, locking the robotic arm in its current position. The piezoelectric valve and screw valve are closed, while the glue return valve is activated (to prevent glue from hardening and clogging).

[0162] S6.3: Record fault data packets.

[0163] It should be understood that the fault data packet records relevant information for this fault handling, including the device ID, adhesive type, current process document, and handling log. The data packet is written to the FPGA's NAND Flash in binary format and uploaded to the cloud database via the MQTT protocol.

[0164] For cases requiring secondary processing, the operator is notified to clear the fault.

[0165] Ideally, anomaly detection thresholds should be set to differentiate between normal, Level 1, and Level 2 anomalies. When the glue line height deviation exceeds the set first threshold but is not severe, Level 1 processing is triggered, reducing the speed of the dispensing head and conveyor belt, and activating vacuum glue suction to reduce the risk of glue accumulation. If the height deviation exceeds the second threshold, Level 2 processing is triggered, immediately stopping equipment operation, cutting off power to the servo motor and glue valve, and simultaneously activating the glue return valve to prevent glue blockage. All anomalies are recorded as fault data packets, including equipment parameters, glue type, and processing logs, and uploaded to a cloud database to ensure traceability. Level 2 processing will notify the operator to intervene and prevent the accumulation of production anomalies.

[0166] This embodiment also provides a computer device applicable to the motion control method for preventing glue buildup in a dispensing machine, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the motion control method for preventing glue buildup in a dispensing machine as proposed in the above embodiment.

[0167] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0168] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the motion control method for preventing glue buildup in a dispensing machine as described in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0169] In summary, this invention achieves the following: a non-Newtonian fluid dynamic model combined with a four-dimensional compensation coefficient matrix accurately characterizes the rheological properties of the adhesive and calibrates parameters in real time; a lightweight convolutional neural network is used to predict adhesive volume deviation, combined with edge computing and online learning algorithms to achieve dynamic optimization of control commands; Jacobi matrix calculation and servo drive compensation optimize the smoothness and synchronization of multi-axis motion trajectories, and curvature radius constraints and cubic spline interpolation effectively suppress corner adhesive buildup; simultaneously, based on laser height measurement and a graded anomaly handling strategy, real-time monitoring and graded response to adhesive buildup risks are achieved, improving the accuracy of adhesive line height control and production stability.

[0170] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A motion control method for preventing glue piling in a dispensing machine, characterized in that: This includes constructing a non-Newtonian fluid dynamic model based on the motion parameters of the dispensing machine and the physical properties of the adhesive using power-law equations, generating a four-dimensional compensation coefficient matrix, and forming a benchmark parameter set; collecting working condition data using multi-source sensors, packaging it into a time-series data stream, transmitting it to an embedded FPGA for normalization processing, and generating standardized feature vectors. The glue quantity deviation is predicted by a convolutional neural network, and the glue quantity compensation is calculated based on the linkage of piezoelectric valve and screw valve to obtain the control command set. The Jacobian matrix is ​​constructed to calculate the equivalent inertial force at the end, and the reverse compensation current is injected through the servo driver to simultaneously optimize the multi-axis motion trajectory. At the inflection point, the path is replanned by the radius of curvature constraint, and the compensation motor control command is output. The height of the adhesive line is detected by laser height measurement, triggering an online learning algorithm to update the weights of the convolutional neural network, switching the process parameter file, and controlling the synchronization of the conveyor belt cycle and dispensing interval time. The adhesive line height is monitored, an abnormal adhesive buildup classification handling strategy is implemented, and fault data packets are recorded. The adhesive physical properties include the shear stress characteristics of the UV-cured adhesive and epoxy resin adhesive, and the four-dimensional compensation coefficient matrix includes velocity, acceleration, temperature, and air pressure. The non-Newtonian fluid dynamic model adds a thixotropic differential term to the power-law model, incorporating the adhesive structure destruction effect caused by acceleration into the non-Newtonian fluid dynamic model for unsteady fluid control. The parameters of the non-Newtonian fluid dynamic model are calibrated and updated online based on feedback from the laser height measurement and adhesive pressure sensor.

2. The motion control method for preventing glue piling in a dispensing machine as described in claim 1, characterized in that: The convolutional neural network is a lightweight residual network. The input layer contains a four-dimensional compensation coefficient matrix of time-series data, the hidden layer uses depthwise separable convolutional kernels, and the output layer generates bias weights through the Sigmoid activation function. The online learning algorithm uses stochastic gradient descent with a dynamic factor, and the update cycle is determined by fixed time and acceleration.

3. The motion control method for preventing glue piling in a dispensing machine as described in claim 1, characterized in that: The optimized multi-axis motion trajectory includes cubic spline interpolation path smoothing and dynamic curvature radius adjustment, and the insertion of transition fillets according to the corner angle.

4. The motion control method for preventing glue buildup in a dispensing machine as described in claim 1, characterized in that: The switching process parameter file refers to the sending of equipment status codes, which allocates the dispensing areas of piezoelectric valves and screw valves during multi-station collaborative operations.

5. The motion control method for preventing glue piling in a dispensing machine as described in claim 1, characterized in that: The abnormal glue stacking graded processing strategy is divided into primary processing and secondary processing. A first threshold and a second threshold are set. The first threshold is less than the second threshold. When the glue line height is greater than the first threshold but less than or equal to the second threshold, primary processing is used. When the glue line height is greater than the second threshold, secondary processing is used. When the glue line height is less than or equal to the first threshold, no processing is performed. The primary processing reduces the movement speed and starts vacuum glue suction. The secondary processing triggers an emergency stop protocol and generates a fault data packet.

6. The motion control method for preventing glue buildup in a dispensing machine as described in claim 2, characterized in that: The convolutional neural network is deployed on the edge computing unit and achieves clock synchronization and jitter control with the motion controller through a communication bus.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the motion control method for preventing glue piling in a dispensing machine as described in any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the motion control method for preventing glue piling in a dispensing machine as described in any one of claims 1 to 6.

Images