Automobile production line gluing system based on visual inspection

Abstract

The invention discloses an automobile production line gluing system based on visual inspection, and relates to the technical field of industrial automation and intelligent control, and the automobile production line gluing system comprises a gluing data acquisition module, a gluing feature extraction module, a compensation correction module, a gluing decision module and a closed-loop management and control module; the gluing data acquisition module preprocesses assembly line gluing track images and glue gun nozzle three-dimensional coordinate data, the gluing feature extraction module constructs a gluing feature sequence table, the compensation correction module outputs glue outlet opening degree compensation amount and nozzle track correction amount, and the gluing decision-making module outputs glue gun pressure adjusting amount and nozzle movement speed. And the closed-loop management and control module detects pressure abnormity and speed stability in real time, optimizes control parameters and generates a process quality map. Through real-time visual detection and dynamic closed-loop control, the glue line width precision is improved, the track compensation response is shortened, the glue gun speed fluctuation coefficient and glue waste are reduced, the gluing qualification rate is improved, the production cost is reduced, and the automobile manufacturing and production requirements are met.

Description

Technical Field

[0001] This invention relates to the field of industrial automation and intelligent control technology, specifically to a vision-based adhesive application system for an automobile production line. Background Technology

[0002] Glue application on the automotive production line is a crucial step in the automotive manufacturing process. The production line encompasses welding, stamping, painting, and powertrain processes. Glue application is primarily carried out in the welding and painting workshops. Adhesives used in the welding workshop include edge adhesive, expanding adhesive, spot weld sealant, vibration damping sheets, and structural adhesive, used for door edging, vibration reduction, noise reduction, and waterproofing. The painting workshop uses weld sealant, PVC anti-stone chip coating, and cavity spray wax to ensure the vehicle body's sealing, stone chip protection, and corrosion resistance. With the development of automation, robotic glue application has become widely used, ensuring stable and reliable glue application quality, meeting the requirements for vehicle body structural strength, vibration reduction, noise reduction, rust prevention, and sealing, and improving the overall performance and safety of the vehicle.

[0003] To address the issues of glue line forming accuracy and process stability in automotive production line adhesive application, existing technologies employ manual experience-based adjustments and single-sensor static monitoring. However, this approach suffers from decoupling between visual inspection and motion control, leading to problems such as excessive glue line width fluctuations, delayed trajectory offset compensation, and inability to adapt process parameters to multi-model production. To resolve these issues, a vision-based adhesive application system for automotive production lines is proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a vision-based adhesive application system for automotive production lines to solve the problems mentioned in the background section.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a visual inspection-based automotive production line adhesive application system, including an adhesive application data acquisition module, an adhesive application feature extraction module, a compensation and correction module, an adhesive application decision module, and a closed-loop control module;

[0006] The glue application data acquisition module collects and preprocesses glue application trajectory image data and glue gun nozzle three-dimensional coordinate data in the automobile production line.

[0007] The glue application feature extraction module extracts features from the preprocessed glue application trajectory image data and the three-dimensional coordinate data of the glue gun nozzle to construct a glue application feature sequence table.

[0008] The compensation and correction module, in conjunction with the glue application feature sequence table, constructs a width compensation model and a trajectory correction model, and outputs the glue gun's nozzle opening compensation amount and nozzle trajectory correction amount.

[0009] The glue application decision module dynamically weights the glue outlet opening compensation amount and the nozzle trajectory correction amount, and outputs the glue gun pressure adjustment amount and the nozzle movement speed.

[0010] The closed-loop control module performs pressure anomaly detection, speed stability detection, and feedback control optimization based on the glue gun pressure adjustment and nozzle movement speed, and generates a process quality map and outputs corresponding control measures.

[0011] A further improvement to the technical solution of this invention lies in the following: the acquisition and preprocessing process of the adhesive application trajectory image data and the three-dimensional coordinate data of the glue gun nozzle in the automotive production line in the adhesive application data acquisition module includes:

[0012] A 20-megapixel industrial area array camera and a ring-shaped multi-angle adjustable LED light source are deployed 30 cm behind the end effector of the glue-applying robot. The glue-applying robot sends a trigger signal every 0.5 mm of movement, and the camera captures 1024×1024 pixel glue-applying trajectory image data at a rate of 30 frames per second.

[0013] A laser triangulation sensor is installed 10 centimeters directly above the glue gun nozzle. The coordinate system of the laser triangulation sensor is aligned with the base coordinate system of the glue dispensing robot using a six-point calibration method. The laser triangulation sensor collects three-dimensional coordinate data of the glue gun nozzle at a frequency of 1000 Hz.

[0014] The camera lens axis is at a 45-degree angle to the coating plane. The camera is hardwired to the motion controller of the coating robot and receives the pulse signal that triggers an exposure every 0.5 mm of robot movement. The LED light source has a wavelength range of 450-650 nm, an illuminance range of 5000-10000 lux, a light source incident angle of 15 degrees, and an XYZ three-axis measurement accuracy of ±0.01 mm.

[0015] A spatial-color dual-domain filtering algorithm is used to eliminate noise in the glue coating trajectory image data. The spatial Gaussian kernel suppresses isolated noise points, and the color Gaussian kernel smooths gray-level abrupt changes. Polarization filtering and adaptive light intensity adjustment are combined to suppress metallic reflections. The image pixel coordinates are converted into millimeter-level physical coordinates in the glue coating robot's base coordinate system through the camera calibration matrix.

[0016] The motion noise in the three-dimensional coordinate data of the glue gun nozzle is smoothed by using a moving average filter. The window width is set to 10 sampling points. Abnormal fluctuation data is removed according to the criterion of 3 times the standard deviation. The motion speed compensation of the glue dispensing robot is added synchronously, and the output is a three-dimensional coordinate sequence of the glue gun nozzle with strictly matched timestamps.

[0017] A further improvement of the technical solution of the present invention is that: in the adhesive application feature extraction module, the process of extracting features from the preprocessed adhesive application trajectory image data includes:

[0018] An edge detection algorithm is used to scan the glue application trajectory image, locate the boundary point set on both sides of the glue line, traverse the edge point pair along the normal direction, obtain the distance between adjacent edge points and take the maximum value as the glue line width feature;

[0019] The equation of the glue line centerline is extracted by a straight line detection algorithm, and the vertical distance between the actual centerline and the theoretical reference point is used as the offset feature of the glue line center position.

[0020] A further improvement to the technical solution of this invention lies in the following: In the glue application feature extraction module, the process of extracting features from the preprocessed three-dimensional coordinate data of the glue gun nozzle and constructing a glue application feature sequence table includes:

[0021] The displacement changes of the three-dimensional coordinates of the glue gun nozzle at adjacent timestamps are statistically analyzed. Combined with the sampling interval, the instantaneous velocity is obtained. Based on the mean and standard deviation of 100 consecutive instantaneous velocities, the characteristics of the glue gun motion speed fluctuation coefficient are obtained.

[0022] The glue line width feature, glue line center position offset feature, and glue gun movement speed fluctuation coefficient feature are aligned with millisecond-level timestamps and normalized to construct a glue application feature sequence table.

[0023] A further improvement to the technical solution of this invention lies in the following: In the compensation correction module, the process of constructing a width compensation model by combining the glue application feature sequence table and outputting the glue gun's nozzle opening compensation amount includes:

[0024] 100,000 sets of historical data on the offset of the glue line center position and the fluctuation coefficient of the glue gun movement speed are integrated into an input vector. A width compensation model is constructed using a backpropagation neural network architecture improved by particle swarm optimization algorithm. The backpropagation neural network architecture includes an input layer, a hidden layer and an output layer.

[0025] The input layer contains two nodes, corresponding to the offset of the glue line center position and the fluctuation coefficient of the glue gun movement speed, respectively. The hidden layer controls the information flow through the internal forget gate and input gate. The memory unit stores the long-term dependency relationship between temporal features. The output layer uses a linear function to map the hidden state output at the last time step of the output gate to the glue gun nozzle opening compensation amount through a fully connected layer.

[0026] The neural network weights and biases represented by each particle are encoded as 22-dimensional particle vectors. The fitness function is defined as the reciprocal of the absolute value of the prediction error. The optimal weight combination is searched globally through iterative updates of particle velocity and position until the fitness function converges. The maximum number of iterations is set to 500.

[0027] The system takes into account the real-time offset of the center position of the glue line and the fluctuation coefficient of the glue gun's movement speed. The trained width compensation model outputs the glue gun's nozzle opening compensation amount and limits the nozzle opening compensation amount according to process constraints.

[0028] A further improvement to the technical solution of this invention lies in the following: In the compensation and correction module, the process of constructing a trajectory correction model by combining the glue application feature sequence table and outputting the nozzle trajectory correction amount of the glue gun includes:

[0029] Using a 10-millisecond time window, 100,000 sets of historical sample data, including the glue line width feature and the glue line center position offset feature, are combined into a two-dimensional input feature vector containing timestamps, and the input features are normalized.

[0030] The radial basis function kernel of the support vector machine regression algorithm is adopted. The nonlinear classification boundary is constructed by using the kernel function mapping. A regression function containing weights and biases is established. The optimal weights and biases are solved by the Lagrange dual method to obtain the set of support vectors and the corresponding Lagrange multipliers. The classification hyperplane is determined by the set of support vectors and the Lagrange multipliers. Constraints are set to minimize the regularization loss function. The classification distance is converted into a high-dimensional space mapping output by the exponential kernel function to construct a trajectory correction model. The constraints include slack variables and insensitive loss thresholds.

[0031] The normalized input features at the current moment are input into the trajectory correction model to calculate the nozzle trajectory correction amount of the glue gun. The calculation process is as follows:

[0032] ;

[0033] in, For the number of support vectors, Let Lagrange multipliers be the values ​​corresponding to the i-th support vector. The radial basis function is used to compute the current input feature vector. With the i-th support vector similarity, For bias terms;

[0034] The nozzle trajectory correction threshold range is set based on the maximum permissible offset of the adhesive dispensing robot's motion axis. millimeters, if millimeters, then set millimeters, if If it's in millimeters, then output directly. ,like millimeters, then set millimeters;

[0035] Abnormal events are marked using the offset threshold of the glue line center position and the out-of-tolerance range of the glue line width. The optimal kernel function parameters and penalty factor are determined by the grid search method, the loss function is minimized, and the parameters Lagrange multipliers and bias terms are fitted based on the validation set data.

[0036] A further improvement to the technical solution of this invention lies in the following: In the glue application decision module, the process of dynamically weighting the glue outlet opening compensation amount and the nozzle trajectory correction amount to output the glue gun pressure adjustment amount and the nozzle movement speed includes:

[0037] Define width compensation weight With trajectory correction weights Satisfying constraints Set initial weights Based on the width error and the offset of the glue line center position, a dynamic adjustment rule is set. When the absolute value of the deviation between the actual value and the theoretical value of the glue line width exceeds 0.1 mm, ,and When the offset of the center position of the adhesive line exceeds 0.15 mm, ;

[0038] The nozzle opening compensation amount output by the width compensation model Combined with width compensation weight Calculate the pressure adjustment of the glue gun. The nozzle trajectory correction amount output by the trajectory correction model Combined with trajectory correction weights Coefficient of motion speed of glue gun Calculate the nozzle movement speed The calculation process is as follows:

[0039] ;

[0040] ;

[0041] in, This is the current reference value for glue gun pressure. This is the current baseline value for the movement speed of the adhesive application robot;

[0042] The pressure adjustment of the glue gun is limited to 0.8-1.2 times the rated pressure adjustment, and the nozzle movement speed is limited to 0.5-1.5 times the rated nozzle movement speed. If the range is exceeded, the glue gun will be forcibly shut off.

[0043] The width compensation weight and trajectory correction weight are reset to their initial values ​​every 10 minutes. When a broken glue line or a clogged glue gun is detected, the weight adjustment is immediately frozen and an emergency stop is triggered.

[0044] A further improvement to the technical solution of this invention lies in the following: the process of detecting pressure anomalies and speed stability based on the glue gun pressure adjustment and nozzle movement speed in the closed-loop control module includes:

[0045] Based on a sliding window of 50 consecutive samples of glue gun pressure adjustment, the mean and standard deviation were statistically analyzed. A control limit of 3 times the standard deviation was set. A level one pressure warning was triggered when 5 consecutive samples exceeded 2 times the standard deviation. A level two shutdown was triggered when a single sample showed a glue gun pressure adjustment exceeding 3 times the standard deviation and a pressure mutation rate exceeding 20% / second. The slope of the linear regression of the pressure sequence was analyzed simultaneously. If the absolute value of the slope was greater than 0.05... If the pressure trend is abnormal within seconds, it will be triggered.

[0046] The glue gun speed fluctuation coefficient was statistically analyzed from 100 continuous nozzle motion speed samples. ,like If the nozzle movement speed is qualified, then the nozzle movement speed is deemed acceptable. If the nozzle movement speed is detected, a warning is issued. If the nozzle movement speed is abnormal, the absolute value of the difference between the movement speeds of adjacent nozzles is detected. If the difference exceeds 0.1 times the current glue-applying robot movement speed benchmark value for 3 consecutive times, the nozzle is determined to have a sudden speed change abnormality.

[0047] A further improvement to the technical solution of this invention lies in the following: the feedback control optimization process in the closed-loop control module, based on the glue gun pressure adjustment and nozzle movement speed, includes:

[0048] The weights of the width compensation model are adjusted by the frequency of abnormal stress events. The more frequent the abnormal stress events, the lower the corresponding weights become proportionally. At that time, the support vector machine penalty factor is increased;

[0049] Based on the deviation direction between the actual and theoretical pressure values, the width compensation weight is adjusted with a fixed step size, and the trajectory correction weight is updated synchronously.

[0050] A further improvement to the technical solution of this invention lies in the following: the process of drawing the process quality map and outputting corresponding control measures in the closed-loop control module includes:

[0051] A three-dimensional process quality map is constructed based on the glue gun pressure adjustment, nozzle movement speed, and glue line width deviation. The quality status is mapped by color. The pressure, speed, and width deviation data are stored in the glue application system of the automobile production line in real time at 10-millisecond intervals. The process quality map is refreshed every 5 minutes. A manual intervention interface is provided, allowing engineers to manually adjust the width compensation weight and trajectory correction weight.

[0052] When the glue line width exceeds the tolerance, the root cause probability is obtained by combining pressure, speed and environmental data, triggering a first-level warning and a second-level shutdown, and taking corresponding control measures. The control measures include triggering an audible and visual alarm for the first-level warning, issuing warning information and maintaining the current control parameters; cutting off the glue gun pressure supply for the second-level shutdown, slowing down and stopping the robot, generating a maintenance work order and marking the root cause type including pressure change and speed fluctuation.

[0053] Due to the adoption of the above technical solution, the technical progress achieved by this invention compared to the prior art is as follows:

[0054] 1. This invention provides a vision-based automotive production line adhesive application system. By acquiring real-time data on the three-dimensional coordinates of the glue gun nozzle and the adhesive application trajectory image, and combining a dynamic weighted decision model and a closed-loop feedback mechanism, the system improves the accuracy of glue line width deviation control, reduces trajectory offset compensation response time, and significantly improves the consistency and pass rate of the adhesive application process.

[0055] 2. This invention provides a vision-based automotive production line adhesive application system. It adopts a neural network and support vector machine regression model improved by particle swarm optimization algorithm to adaptively adjust the glue gun pressure and movement speed, reduce the glue gun speed fluctuation coefficient and glue waste rate, significantly save production costs and improve equipment operating efficiency.

[0056] 3. This invention provides a vision-based automotive production line adhesive coating system. By using a three-dimensional process quality map and a multi-parameter root cause tracing algorithm, it improves the accuracy of abnormal event location. Combined with hierarchical control measures, it shortens the fault diagnosis time, ensures the continuous and stable operation of the production line, and meets the stringent requirements of high cycle time and zero defects in automotive manufacturing. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0058] Figure 1 This is a block diagram of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] Examples, such as Figure 1 As shown, the present invention provides a vision-based detection-based glue application system for automobile production lines, including a glue application data acquisition module, a glue application feature extraction module, a compensation and correction module, a glue application decision module, and a closed-loop control module.

[0061] The glue application data acquisition module collects and preprocesses glue application trajectory image data and glue gun nozzle 3D coordinate data from the automotive production line. A 20-megapixel industrial area array camera and a ring-shaped multi-angle adjustable LED light source are deployed 30 cm behind the glue application robot's end effector. The glue application robot sends a trigger signal every 0.5 mm of movement, and the camera captures 1024×1024 pixel glue application trajectory image data at a rate of 30 frames per second. A laser triangulation rangefinder is installed 10 cm directly above the glue gun nozzle. The coordinate system of the laser triangulation rangefinder is aligned with the base coordinate system of the glue application robot using a six-point calibration method. The laser triangulation rangefinder collects the 3D coordinate data of the glue gun nozzle at a frequency of 1000 Hz. The camera lens axis is at a 45-degree angle to the glue application plane. The camera is hardwired to the glue application robot's motion controller, receiving trigger signals every 0.5 mm of robot movement. A pulse signal for a single exposure is emitted. The LED light source has a wavelength range of 450-650 nm, an illuminance range of 5000-10000 lux, a light source incident angle of 15 degrees, and an XYZ three-axis measurement accuracy of ±0.01 mm. A spatial-color dual-domain filtering algorithm is used to eliminate noise in the glue application trajectory image data. The spatial Gaussian kernel suppresses isolated noise points, and the color Gaussian kernel smooths gray-scale abrupt changes. Polarization filtering and adaptive light intensity adjustment are combined to suppress metallic reflections. The image pixel coordinates are converted into millimeter-level physical coordinates in the glue application robot's base coordinate system through the camera calibration matrix. A moving average filter is used to smooth the motion noise in the glue gun nozzle's three-dimensional coordinate data. The window width is set to 10 sampling points. Abnormal fluctuation data is removed according to the 3-times standard deviation criterion. The glue application robot's motion speed compensation is superimposed synchronously, and the output is a glue gun nozzle three-dimensional coordinate sequence with strictly matched timestamps.

[0062] The glue application feature extraction module extracts features from the preprocessed glue application trajectory image data and the three-dimensional coordinate data of the glue gun nozzle, constructs a glue application feature sequence table, uses an edge detection algorithm to scan the glue application trajectory image, locates the boundary point set on both sides of the glue line, traverses the edge point pairs along the normal direction, obtains the distance between adjacent edge points and takes the maximum value as the glue line width feature, extracts the glue line centerline equation through a straight line detection algorithm, takes the vertical distance between the actual centerline and the theoretical reference point as the glue line center position offset feature, counts the displacement change of the three-dimensional coordinates of the glue gun nozzle at adjacent timestamps, and obtains the instantaneous speed by combining the sampling interval. Based on the mean and standard deviation of 100 consecutive instantaneous speeds, the glue gun movement speed fluctuation coefficient feature is obtained. The glue line width feature, glue line center position offset feature and glue gun movement speed fluctuation coefficient feature are aligned by millisecond-level timestamps and normalized to construct the glue application feature sequence table.

[0063] The compensation and correction module, combined with the glue application feature sequence list, constructs a width compensation model and a trajectory correction model, outputting the glue gun's nozzle opening compensation and nozzle trajectory correction. It integrates 100,000 sets of historical data on the glue line center position offset and glue gun speed fluctuation coefficient into an input vector. A backpropagation neural network architecture improved by particle swarm optimization is used to construct the width compensation model. This architecture includes an input layer, hidden layers, and an output layer. The input layer contains two nodes, corresponding to the glue line center position offset and the glue gun speed fluctuation coefficient, respectively. The hidden layer controls the information flow through an internal forget gate and input gate, and memory units store long-term dependencies between temporal features. The output layer uses a linear function to map the hidden state output at the last time step of the output gate to the glue gun's nozzle opening compensation via a fully connected layer. The neural network weights and biases represented by each particle are encoded as 22-dimensional particle vectors. The fitness function is defined as the reciprocal of the absolute value of the prediction error. The optimal weight combination is searched globally through iterative updates of particle velocity and position until the fitness function converges. The maximum number of iterations is set to 500. The system takes the current offset of the glue line center position and the fluctuation coefficient of the glue gun's movement speed as input. The trained width compensation model outputs the glue gun's dispensing nozzle opening compensation amount. Based on process constraints, the dispensing nozzle opening compensation amount is limited. Using a 10-millisecond time window, 100,000 sets of historical sample data representing the glue line width and center position offset features are combined into a two-dimensional input feature vector containing timestamps. The input features are then normalized, and the radial basis function of the support vector machine regression algorithm is used to construct a nonlinear classification boundary through kernel function mapping. A regression function containing weights and biases is established. The optimal weights and biases are solved using the Lagrange duality method to obtain the support vector set and corresponding Lagrange multipliers. The classification hyperplane is jointly determined by the support vector set and the Lagrange multipliers. Constraints are set to minimize the regularized loss function. The classification distance is converted into a high-dimensional space mapping output using an exponential kernel function to construct a trajectory correction model. The constraints include slack variables and an insensitive loss threshold. The normalized input features at the current time are input into the trajectory correction model to calculate the nozzle trajectory correction amount of the glue gun. The calculation process is as follows:

[0064] ;

[0065] in, For the number of support vectors, Let Lagrange multipliers be the values ​​corresponding to the i-th support vector. The radial basis function is used to compute the current input feature vector. With the i-th support vector similarity, As an offset term, the nozzle trajectory correction threshold range is set according to the maximum permissible offset of the adhesive dispensing robot's motion axis. millimeters, if millimeters, then set millimeters, if If it's in millimeters, then output directly. ,like millimeters, then set Millimeters, using the offset threshold of the glue line center position and the glue line width out-of-tolerance range to mark abnormal events, determine the optimal kernel function parameters and penalty factor through grid search method, minimize the loss function, and fit the parameters Lagrange multipliers and bias terms based on the validation set data;

[0066] The glue application decision module dynamically weights the glue outlet opening compensation and nozzle trajectory correction, outputting the glue gun pressure adjustment and nozzle movement speed, and defining the width compensation weight. With trajectory correction weights Satisfying constraints Set initial weights Based on the width error and the offset of the glue line center position, a dynamic adjustment rule is set. When the absolute value of the deviation between the actual value and the theoretical value of the glue line width exceeds 0.1 mm, ,and When the offset of the center position of the adhesive line exceeds 0.15 mm, The amount of nozzle opening compensation output based on the width compensation model. Combined with width compensation weight Calculate the pressure adjustment of the glue gun. The nozzle trajectory correction amount output by the trajectory correction model Combined with trajectory correction weights Coefficient of motion speed of glue gun Calculate the nozzle movement speed The calculation process is as follows:

[0067] ;

[0068] ;

[0069] in, This is the current reference value for glue gun pressure. The current glue-applying robot's movement speed is set as the baseline value. The glue gun pressure adjustment is limited to 0.8-1.2 times the rated glue gun pressure adjustment, and the nozzle movement speed is limited to 0.5-1.5 times the rated nozzle movement speed. If the range is exceeded, the robot will be forcibly shut down. The width compensation weight and trajectory correction weight will be reset to their initial values ​​every 10 minutes. When glue line breakage and glue gun blockage are detected, the weight adjustment will be frozen immediately and an emergency stop will be triggered.

[0070] The closed-loop control module performs pressure anomaly detection, speed stability detection, and feedback control optimization based on the glue gun pressure adjustment and nozzle movement speed. It also generates a process quality graph and outputs corresponding control measures. Based on a sliding window of 50 consecutive glue gun pressure adjustment samples, it calculates the mean and standard deviation, setting a control limit of 3 times the standard deviation. A level one pressure warning is triggered when five consecutive samples exceed 2 times the standard deviation. A level two shutdown is triggered when a single sample shows a glue gun pressure adjustment exceeding 3 times the standard deviation and a pressure mutation rate exceeding 20% / second. Simultaneously, it analyzes the slope of the linear regression of the pressure sequence. If the absolute value of the slope is greater than 0.05... If the pressure trend is abnormal within seconds, the glue gun speed fluctuation coefficient is calculated from 100 consecutive nozzle movement speed samples. ,like If the nozzle movement speed is qualified, then the nozzle movement speed is deemed acceptable. If the nozzle movement speed is detected, a warning is issued. If the nozzle movement speed is abnormal, it is determined that the absolute value of the speed difference between adjacent nozzles exceeds 0.1 times the current glue-applying robot's speed benchmark value for three consecutive times. This indicates a sudden speed change anomaly. The width compensation model weight is adjusted based on the frequency of pressure anomaly events; the more pressure anomalies occur, the lower the corresponding weight becomes. Simultaneously, the support vector machine penalty factor is incremented, and the width compensation weight is adjusted with a fixed step size according to the deviation direction between the actual and theoretical pressure values. The trajectory correction weight is updated synchronously. A three-dimensional process quality map is constructed based on the glue gun pressure adjustment, nozzle movement speed, and glue line width deviation. The quality status is mapped by color, and the pressure, speed, and width deviation data are stored in the automotive production line glue application system in real time at 10-millisecond intervals. The process quality map is refreshed every 5 minutes. A manual intervention interface is provided, allowing engineers to manually adjust the width compensation weight and trajectory correction weight. When the glue line width exceeds the tolerance, the root cause probability is obtained by comprehensively considering the pressure, speed, and environmental data, triggering a first-level warning and a second-level shutdown, and taking corresponding control measures. The control measures include triggering an audible and visual alarm for the first-level warning, issuing warning information, and maintaining the current control parameters; cutting off the glue gun pressure supply for the second-level shutdown, slowing down and stopping the robot, generating a maintenance work order, and marking the root cause type including pressure mutation and speed fluctuation.

[0071] First, the glue application trajectory image and the three-dimensional coordinate data of the glue gun nozzle are acquired in real time by the glue application data acquisition module. After noise reduction filtering and coordinate alignment processing, the data is transmitted to the glue application feature extraction module. Next, the feature extraction module analyzes the glue line width, position offset, and speed fluctuation coefficient, and constructs a normalized feature sequence list to be input into the compensation and correction module. Subsequently, the compensation and correction module generates the glue nozzle opening compensation amount through the width compensation model, and calculates the nozzle trajectory correction amount using the trajectory correction model. The two parameters are transmitted to the glue application decision module. This module fuses the two parameters based on a dynamic weighted algorithm and outputs glue gun pressure adjustment commands and movement speed commands in real time. Finally, the closed-loop control module monitors the pressure and speed stability of the execution effect, provides visual feedback through a three-dimensional process quality map, and simultaneously triggers anomaly classification control and model parameter self-optimization.

[0072] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A vision-based adhesive application system for an automotive production line, characterized in that: It includes a glue application data acquisition module, a glue application feature extraction module, a compensation and correction module, a glue application decision module, and a closed-loop control module; The glue application data acquisition module collects and preprocesses glue application trajectory image data and glue gun nozzle three-dimensional coordinate data in the automobile production line. The glue application feature extraction module extracts features from the preprocessed glue application trajectory image data and the three-dimensional coordinate data of the glue gun nozzle to construct a glue application feature sequence table. The compensation and correction module, in conjunction with the glue application feature sequence table, constructs a width compensation model and a trajectory correction model, and outputs the glue gun's nozzle opening compensation amount and nozzle trajectory correction amount. The glue application decision module dynamically weights the glue outlet opening compensation amount and the nozzle trajectory correction amount, and outputs the glue gun pressure adjustment amount and the nozzle movement speed. The closed-loop control module performs pressure anomaly detection, speed stability detection, and feedback control optimization based on the glue gun pressure adjustment and nozzle movement speed, and generates a process quality map and outputs corresponding control measures.

2. The vision-based adhesive coating system for automobile production lines according to claim 1, characterized in that: The glue application data acquisition module includes the following processes for acquiring and preprocessing glue application trajectory image data and glue gun nozzle three-dimensional coordinate data in the automotive production line: A 20-megapixel industrial area array camera and a ring-shaped multi-angle adjustable LED light source are deployed 30 cm behind the end effector of the glue-applying robot. The glue-applying robot sends a trigger signal every 0.5 mm of movement, and the camera captures 1024×1024 pixel glue-applying trajectory image data at a rate of 30 frames per second. A laser triangulation sensor is installed 10 centimeters directly above the glue gun nozzle. The coordinate system of the laser triangulation sensor is aligned with the base coordinate system of the glue dispensing robot using a six-point calibration method. The laser triangulation sensor collects three-dimensional coordinate data of the glue gun nozzle at a frequency of 1000 Hz. A spatial-color dual-domain filtering algorithm is used to eliminate noise in the glue coating trajectory image data. The spatial Gaussian kernel suppresses isolated noise points, and the color Gaussian kernel smooths gray-level abrupt changes. Polarization filtering and adaptive light intensity adjustment are combined to suppress metallic reflections. The image pixel coordinates are converted into millimeter-level physical coordinates in the glue coating robot's base coordinate system through the camera calibration matrix. The motion noise in the three-dimensional coordinate data of the glue gun nozzle is smoothed by using a moving average filter. The window width is set to 10 sampling points. Abnormal fluctuation data is removed according to the criterion of 3 times the standard deviation. The motion speed compensation of the glue dispensing robot is added synchronously, and the output is a three-dimensional coordinate sequence of the glue gun nozzle with strictly matched timestamps.

3. The vision-based adhesive coating system for automobile production lines according to claim 2, characterized in that: The glue application feature extraction module includes the following process for extracting features from the preprocessed glue application trajectory image data: An edge detection algorithm is used to scan the glue application trajectory image, locate the boundary point set on both sides of the glue line, traverse the edge point pair along the normal direction, obtain the distance between adjacent edge points and take the maximum value as the glue line width feature; The equation of the glue line centerline is extracted by a straight line detection algorithm, and the vertical distance between the actual centerline and the theoretical reference point is used as the offset feature of the glue line center position.

4. The vision-based adhesive coating system for automobile production lines according to claim 3, characterized in that: The glue application feature extraction module extracts features from the preprocessed 3D coordinate data of the glue gun nozzle and constructs a glue application feature sequence table, including: The displacement changes of the three-dimensional coordinates of the glue gun nozzle at adjacent timestamps are statistically analyzed. Combined with the sampling interval, the instantaneous velocity is obtained. Based on the mean and standard deviation of 100 consecutive instantaneous velocities, the characteristics of the glue gun motion speed fluctuation coefficient are obtained. The glue line width feature, glue line center position offset feature, and glue gun movement speed fluctuation coefficient feature are aligned with millisecond-level timestamps and normalized to construct a glue application feature sequence table.

5. The vision-based adhesive coating system for automobile production lines according to claim 4, characterized in that: In the compensation and correction module, the process of constructing a width compensation model by combining the glue application feature sequence table and outputting the glue gun nozzle opening compensation amount includes: 100,000 sets of historical data on the offset of the glue line center position and the fluctuation coefficient of the glue gun movement speed are integrated into an input vector. A width compensation model is constructed using a backpropagation neural network architecture improved by particle swarm optimization algorithm. The backpropagation neural network architecture includes an input layer, a hidden layer and an output layer. The input layer contains two nodes, corresponding to the offset of the glue line center position and the fluctuation coefficient of the glue gun movement speed, respectively. The hidden layer controls the information flow through the internal forget gate and input gate. The memory unit stores the long-term dependency relationship between temporal features. The output layer uses a linear function to map the hidden state output at the last time step of the output gate to the glue gun nozzle opening compensation amount through a fully connected layer. The neural network weights and biases represented by each particle are encoded as 22-dimensional particle vectors. The fitness function is defined as the reciprocal of the absolute value of the prediction error. The optimal weight combination is searched globally through iterative updates of particle velocity and position until the fitness function converges. The maximum number of iterations is set to 500. The system takes into account the real-time offset of the center position of the glue line and the fluctuation coefficient of the glue gun's movement speed. The trained width compensation model outputs the glue gun's nozzle opening compensation amount and limits the nozzle opening compensation amount according to process constraints.

6. The vision-based adhesive coating system for an automotive production line according to claim 5, characterized in that: In the compensation and correction module, the process of constructing a trajectory correction model by combining the glue application feature sequence table and outputting the nozzle trajectory correction amount of the glue gun includes: Using a 10-millisecond time window, 100,000 sets of historical sample data, including the glue line width feature and the glue line center position offset feature, are combined into a two-dimensional input feature vector containing timestamps, and the input features are normalized. The radial basis function kernel of the support vector machine regression algorithm is adopted. The nonlinear classification boundary is constructed by using the kernel function mapping. A regression function containing weights and biases is established. The optimal weights and biases are solved by the Lagrange dual method to obtain the set of support vectors and the corresponding Lagrange multipliers. The classification hyperplane is determined by the set of support vectors and the Lagrange multipliers. Constraints are set to minimize the regularization loss function. The classification distance is converted into a high-dimensional space mapping output by the exponential kernel function to construct a trajectory correction model. The constraints include slack variables and insensitive loss thresholds. The normalized input features at the current moment are input into the trajectory correction model to calculate the nozzle trajectory correction amount of the glue gun. The nozzle trajectory correction threshold range is set according to the maximum allowable offset of the glue-applying robot's motion axis. millimeters, if millimeters, then set millimeters, if If it's in millimeters, then output directly. ,like millimeters, then set millimeters; Abnormal events are marked using the offset threshold of the glue line center position and the out-of-tolerance range of the glue line width. The optimal kernel function parameters and penalty factor are determined by the grid search method, the loss function is minimized, and the parameters Lagrange multipliers and bias terms are fitted based on the validation set data.

7. The vision-based adhesive application system for an automotive production line according to claim 6, characterized in that: In the glue application decision module, the process of dynamically weighting the glue outlet opening compensation and nozzle trajectory correction to output the glue gun pressure adjustment and nozzle movement speed includes: Define width compensation weight With trajectory correction weights Satisfying constraints Set initial weights Based on the width error and the offset of the glue line center position, a dynamic adjustment rule is set. When the absolute value of the deviation between the actual value and the theoretical value of the glue line width exceeds 0.1 mm, ,and When the offset of the center position of the adhesive line exceeds 0.15 mm, ; The nozzle opening compensation amount output by the width compensation model Combined with width compensation weight Calculate the pressure adjustment of the glue gun. The nozzle trajectory correction amount output by the trajectory correction model Combined with trajectory correction weights Coefficient of motion speed of glue gun Calculate the nozzle movement speed ; The pressure adjustment of the glue gun is limited to 0.8-1.2 times the rated pressure adjustment, and the nozzle movement speed is limited to 0.5-1.5 times the rated nozzle movement speed. If the range is exceeded, the glue gun will be forcibly shut off. The width compensation weight and trajectory correction weight are reset to their initial values ​​every 10 minutes. When a broken glue line or a clogged glue gun is detected, the weight adjustment is immediately frozen and an emergency stop is triggered.

8. The vision-based adhesive application system for an automotive production line according to claim 7, characterized in that: In the closed-loop control module, the process of detecting pressure anomalies and speed stability based on the glue gun pressure adjustment and nozzle movement speed includes: Based on a sliding window of 50 consecutive samples of glue gun pressure adjustment, the mean and standard deviation were statistically analyzed. A control limit of 3 times the standard deviation was set. A level one pressure warning was triggered when 5 consecutive samples exceeded 2 times the standard deviation. A level two shutdown was triggered when a single sample showed a glue gun pressure adjustment exceeding 3 times the standard deviation and a pressure mutation rate exceeding 20% / second. The slope of the linear regression of the pressure sequence was analyzed simultaneously. If the absolute value of the slope was greater than 0.05... If the pressure trend is abnormal within seconds, it will be triggered. The glue gun speed fluctuation coefficient was statistically analyzed from 100 continuous nozzle motion speed samples. ,like If the nozzle movement speed is qualified, then the nozzle movement speed is deemed to be qualified. If the nozzle movement speed is detected, a warning is issued. If the nozzle movement speed is abnormal, the absolute value of the difference between the movement speeds of adjacent nozzles is detected. If the difference exceeds 0.1 times the current glue-applying robot movement speed benchmark value for 3 consecutive times, the nozzle is determined to have a sudden speed change abnormality.

9. A vision-based adhesive coating system for an automotive production line according to claim 8, characterized in that: In the closed-loop control module, the process of feedback control optimization based on the glue gun pressure adjustment and nozzle movement speed includes: The weights of the width compensation model are adjusted by the frequency of abnormal stress events. The more frequent the abnormal stress events, the lower the corresponding weights become proportionally. At that time, the support vector machine penalty factor is increased; Based on the deviation direction between the actual and theoretical pressure values, the width compensation weight is adjusted with a fixed step size, and the trajectory correction weight is updated synchronously.

10. A vision-based adhesive coating system for an automotive production line according to claim 9, characterized in that: In the closed-loop control module, the process of drawing process quality charts and outputting corresponding control measures includes: A three-dimensional process quality map is constructed based on the glue gun pressure adjustment, nozzle movement speed, and glue line width deviation. The quality status is mapped by color. The pressure, speed, and width deviation data are stored in the glue application system of the automobile production line in real time at 10-millisecond intervals. The process quality map is refreshed every 5 minutes. A manual intervention interface is provided, allowing engineers to manually adjust the width compensation weight and trajectory correction weight. When the glue line width exceeds the tolerance, the root cause probability is obtained by combining pressure, speed and environmental data, triggering a first-level warning and a second-level shutdown, and taking corresponding control measures. The control measures include triggering an audible and visual alarm for the first-level warning, issuing warning information and maintaining the current control parameters; cutting off the glue gun pressure supply for the second-level shutdown, slowing down and stopping the robot, generating a maintenance work order and marking the root cause type including pressure change and speed fluctuation.

Images