A robot welding method and system based on parameterized programming

Abstract

The application discloses a kind of robot welding method and system based on parameterization programming, comprising: step 1: collecting workpiece surface data, extracting geometric dimension feature related to welding from workpiece surface data;Step 2: based on the geometric dimension feature extracted in step 1, path planning is carried out, and welding path is obtained;Step 3: based on welding path and welding parameter, multi-arm collaborative welding is carried out, in the welding process, welding path is tracked in real time, and key parameters in the welding process are collected, whether the path error is less than the set threshold is judged, if less than the set threshold, whether the current welding quality is up to standard is further judged, if up to standard, then end welding;If not up to standard, then according to the key parameters in the welding process, adjust welding parameter, execute step 3;If not less than the set threshold, then according to path error, welding path is corrected, and step 3 is executed.

Description

Technical Field

[0001] This invention relates to the field of robotic welding technology, and in particular to a robotic welding method and system based on parametric programming. Background Technology

[0002] In the field of robotic welding, parametric programming welding systems automatically generate welding program code based on parameters such as workpiece geometry, material properties, and welding process requirements. In traditional programming methods, manual teaching requires operators to manually guide the robot to complete tasks. With frequent changes in product size and shape, each new task requires re-teaching, significantly increasing teaching time and complexity, leading to reduced production efficiency. Offline programming requires simulating the welding process on a computer using specialized programming tools and writing complete program code; this is inefficient and lacks real-time environmental awareness, making it difficult for the program to adapt to dynamic production needs. In contrast, parametric programming, through its highly automated approach, not only shortens programming time but also reduces the possibility of human error, improving work efficiency.

[0003] However, existing robotic welding technology cannot automatically adjust parameters based on welding quality. When welding defects occur, current robotic welding technology cannot automatically adjust key parameters in a timely manner, making it difficult to meet the demands for high-precision, high-efficiency, and high-quality welding. Existing trajectory planning primarily focuses on welding efficiency and reachability, neglecting welding deformation, which is a crucial criterion for product qualification. Summary of the Invention

[0004] Purpose of the invention: To address the problem that existing welding systems cannot automatically adjust parameters according to welding quality in dynamic production environments, and the problem that existing trajectory planning often focuses on welding efficiency and reachability without considering welding deformation, this invention proposes a robot parametric programming welding method and system with multi-objective path planning, multi-modal quality perception, and autonomous parameter adjustment.

[0005] Technical solution: A robot welding method based on parametric programming, comprising the following steps:

[0006] Step 1: Collect workpiece surface data and extract welding-related geometric features from the workpiece surface data;

[0007] Step 2: Based on the geometric features extracted in Step 1, perform path planning to obtain the welding path;

[0008] Step 3: Based on the welding path and welding parameters, perform multi-arm collaborative welding. During the welding process, the welding path is tracked in real time, and key parameters are collected to determine whether the path error is less than a set threshold. If it is less than the set threshold, the current welding quality is further assessed. If it meets the standard, the welding ends. If it does not meet the standard, the welding parameters are adjusted according to the key parameters during the welding process, and Step 3 is executed. If the error is not less than the set threshold, the welding path is corrected according to the path error, and Step 3 is executed.

[0009] In step 2, the path planning based on the geometric features extracted in step 1 to obtain the welding path specifically includes:

[0010] Step 1: Construct multiple objective functions, including: welding efficiency objective function, deformation objective function, path equalization objective function, and welding accessibility objective function; the welding efficiency objective function is to minimize the total welding time; the deformation objective function is to minimize the deformation on the welding path; the path equalization objective function is to balance the workload of each robotic arm; and the welding accessibility objective function is to ensure that each path point is within the reachability space.

[0011] Step 2: Employ a multi-objective particle swarm optimization algorithm. Each particle represents a path scheme, and each position of the particle represents a path. Through the global search capability of the particle swarm, the multi-objective function is balanced and optimized to obtain the optimal path, which is the welding path.

[0012] Furthermore, the objective function for welding efficiency is expressed as:

[0013]

[0014] In the formula, T i (p) represents the welding time, L i v represents the length of each path segment. i (p) represents the welding speed, and n represents the total number of path segments.

[0015] Furthermore, the objective function for the deformation amount is expressed as:

[0016]

[0017] In the formula, Let be the predicted deformation value of the i-th diameter segment; The predicted values ​​are obtained by using a trained physical information neural network model; the physical information neural network model takes the temperature field distribution and thermal stress field during the welding process as inputs and outputs the corresponding predicted values ​​of deformation.

[0018] Furthermore, the path-sharing objective function is expressed as:

[0019]

[0020] In the formula, I i Let be the load of the i-th path segment.

[0021] Furthermore, the objective function for weld accessibility is expressed as:

[0022]

[0023] Among them, 1 reachable (p i ) is the reachability judgment function, if path point p i If it is within the reachable space, the value is 1; otherwise, it is 0.

[0024] Furthermore, the specific operations in Step 2 include:

[0025] Each particle represents a path scheme, and each position of the particle represents a path, which consists of multiple path parameters. The particle swarm is initialized by randomly generating multiple path parameters, and each particle is assigned a random velocity. The path parameters include: geometric parameters, motion parameters, and cooperative parameters. The geometric parameters include: path point coordinates. The motion parameters include welding speed. The cooperative parameters include robotic arm allocation.

[0026] In each iteration, for each particle, four objective functions f1(p), f2(p), f3(p), and f4(p) for its corresponding path scheme are calculated. The calculated objective function values ​​are used to evaluate the quality of the path, and the particle's comprehensive objective function value f is calculated according to the following formula. total (p):

[0027] f total (p)=ω1f1(p)+ω2f2(p)+ω3f3(p)+ω4f4(p)

[0028] Where ω1, ω2, ω3, and ω4 are the weight coefficients of each objective function, and satisfy the following:

[0029] ω1+ω2+ω3+ω4=1

[0030] Minimizing the deformation is the most important optimization objective; therefore, ω2 > ω1, ω3, ω4.

[0031] In each iteration, the particle's velocity and position are updated according to the following formula:

[0032]

[0033] x i(t+1)=x i (t)+v i (t+1)

[0034] Among them, v i (t) is the velocity of the i-th particle at time t, x i (t) is the position of the particle at time t, p best It is the historical optimal position of the particle, g best It is the global optimal position, c1 and c2 are acceleration constants, r1 and r2 are random factors, and ω is the inertia weight;

[0035] Through multiple iterations, the particle swarm continuously updates its position and velocity, and after each iteration, it updates the global optimal solution g. best The algorithm continues to optimize until the preset stopping criteria are met, thus obtaining the optimal path.

[0036] Furthermore, the specific operations of adjusting welding parameters based on key parameters during the welding process include:

[0037] Key parameters in the welding process include: weld image, current data, and noise data;

[0038] Using convolutional neural networks to extract local features of welds from weld images;

[0039] Using a multilayer perceptron, predicted values ​​of welding parameters are obtained based on current and noise data;

[0040] The outputs of the convolutional neural network and the multilayer perceptron are fused together. The fused features are then input into a deep Q-network. The deep Q-network determines the welding parameters that need to be adjusted based on the fused features, and then adjusts the welding parameters accordingly.

[0041] Furthermore, it also includes:

[0042] Establish a self-updating welding parameter adjustment database;

[0043] After each welding process is completed, record the key parameters of the current welding process, as well as the corresponding welding parameters and welding defect types;

[0044] We used data from a self-updating welding parameter tuning database to dynamically train a convolutional neural network for welding defect detection.

[0045] The present invention also discloses a robot welding system based on parametric programming, including a multi-arm robot, a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the robot welding method based on parametric programming disclosed above.

[0046] Beneficial Effects: Compared with existing technologies, this invention uses a high-resolution vision sensor to identify weld seams in real time, extracting weld seam morphology and related geometric features, providing accurate data support for subsequent path planning. Then, with multiple objectives such as welding deformation, welding reachability, and shortest path as optimization directions, a multi-objective particle swarm optimization algorithm (MOPSO) is used to generate initial welding paths and motion parameters, ensuring that the optimized path does not produce excessive deformation while improving welding efficiency. This invention employs multi-robotic arm collaborative welding technology, rationally allocating welding tasks to avoid overloading a single robotic arm, thus improving welding efficiency and accuracy. During the welding process, this invention uses weld seam tracking and autonomous adjustment of pose parameters for correction, employing vision and laser sensors to monitor the welding path in real time, correcting and compensating for path errors to ensure precise execution of the welding path. Furthermore, this invention collects weld seam morphology data and current and voltage information during the welding process. By training a multimodal perception model, the system can analyze welding quality in real time and intelligently adjust welding parameters such as welding current and welding voltage based on a deep Q-network (DQN), thereby ensuring quality stability during the welding process. Furthermore, this invention establishes a self-updating parameter tuning database capable of real-time acquisition and storage of historical data and feedback results during the welding process. After each welding operation, relevant sensor data and welding quality information are recorded and used to update the database. This data serves as training samples for the parameter tuning model, continuously optimizing tuning accuracy through incremental training and online learning. This enhances the system's adaptability and welding quality under different production environments, thereby achieving more efficient and precise welding operations. Specifically, it offers the following advantages:

[0047] (1) Improve welding accuracy: This invention reduces defects and ensures stable welding quality by real-time detection and adjustment of welding parameters;

[0048] (2) Reduced human intervention: The present invention automatically generates welding programs, reducing manual teaching, improving efficiency and reducing errors;

[0049] (3) Optimized path planning: This invention optimizes the welding path through multiple objectives, reduces deformation, balances the load on the robotic arm, and improves work efficiency;

[0050] (4) Dynamically adapt to demand: The present invention dynamically adjusts based on real-time feedback to quickly adapt to changes in workpieces and production environment. Attached Figure Description

[0051] Figure 1 This is a flowchart of a robot welding method based on parametric programming proposed in this invention;

[0052] Figure 2 This is a flowchart of the multi-objective path planning proposed in this invention;

[0053] Figure 3 Flowchart for welding quality perception and parameter adjustment. Detailed Implementation

[0054] The technical solution of this embodiment will now be further described in conjunction with the accompanying drawings and examples.

[0055] Example 1:

[0056] like Figure 1 As shown, this embodiment proposes a robot parametric programming welding method with multi-objective path planning, multi-modal quality perception, and autonomous parameter tuning. The aim is to achieve precise path planning, real-time monitoring and correction, and multi-modal perception and feedback adjustment of welding quality in multi-arm collaborative welding, thereby improving welding efficiency, reducing welding defects, and optimizing the welding process. It mainly includes the following steps:

[0057] Step 1: Use a high-resolution structured light camera to acquire weld morphology data of the workpiece, and extract weld, bevel and related geometric features from the weld morphology data of the workpiece to provide accurate workpiece information for subsequent welding path planning and control.

[0058] Step 2: A Physical Information Neural Network (PINN) model is used to predict deformation during the welding process. This PINN model combines the heat source input, temperature field distribution, thermal stress field, and the relationship between path points during the welding process. SYSWELD simulation results are used as training data to optimize the PINN model. Specifically, the SYSWELD simulation results provide simulation data such as the temperature field distribution and thermal stress field during the welding process. By inputting this SYSWELD simulation data into the PINN model, the model can learn the variation patterns of temperature and thermal stress during the welding process and their relationship with the deformation at path points.

[0059] A loss function is designed to measure the difference between the deformation predicted by the physical information neural network model and the actual deformation in the SYSWELD simulation data. By optimizing this loss function, the accuracy of the physical information neural network model can be improved. The loss function can be defined as:

[0060]

[0061] In the formula, d represents the predicted deformation. i This indicates the actual amount of deformation.

[0062] By using deformation prediction as a prerequisite for path planning, more accurate data support can be provided for subsequent multi-objective path optimization, and the overall quality and efficiency of welding tasks can be effectively improved.

[0063] Step 3: The Multi-Objective Particle Swarm Optimization (MOPSO) algorithm is used to perform multi-objective path planning for complex welding tasks. This comprehensively optimizes objectives such as welding deformation, welding reachability, and path efficiency to generate the initial welding path and motion parameters. The path planning process can be further divided into the following steps:

[0064] First, construct the objective functions, including the welding efficiency objective function, the deformation objective function, the path distribution objective function, and the welding accessibility objective function.

[0065] Welding efficiency objective function: The goal is to shorten the welding path as much as possible to reduce welding time and improve efficiency. Welding time T i (p) and the length L of each path segment i and velocity v i (p) related, represented as:

[0066]

[0067] By calculating the length of the path segment and the welding speed, the path and speed can be adjusted reasonably to reduce the total welding time.

[0068] Deformation Objective Function: During path planning, based on the trained PINN model, deformation is predicted for each point on the path. The deformation predicted by PINN is used as part of the objective function to construct the deformation objective function, which aims to minimize the deformation on the welding path, thereby improving welding quality. This function is expressed as:

[0069]

[0070] in, Let L be the predicted deformation value of the i-th diameter segment. i Let be the length of this path segment. By minimizing this objective function, the amount of deformation during the overall welding process can be reduced.

[0071] Path distribution objective function: To avoid any single robotic arm from undertaking too many tasks, a task distribution strategy ensures a balanced workload for each robotic arm, preventing some arms from being overworked or idle, thereby improving the efficiency of the entire welding process. The path distribution objective function is expressed as:

[0072]

[0073] Among them, I i Let n be the load of the i-th path segment, and n be the number of path segments.

[0074] Welding accessibility objective function: Considering that some areas may be inaccessible to the robotic arm during the welding process, or that there may be motion interference between robotic arms, this embodiment ensures that each path point is within the reachable space through path planning. Unreachable areas will be manually repaired. This objective function is used to avoid blind spots in the robotic arm's movement and ensure path accessibility.

[0075]

[0076] Among them, 1 reachable (p i ) is the reachability judgment function, if path point p i If it is within the reachable space, the value is 1; otherwise, it is 0.

[0077] Secondly, a multi-objective particle swarm optimization algorithm (MOPSO) is employed, leveraging the global search capability of the particle swarm to achieve balanced optimization of multiple objective functions. For example... Figure 2 As shown.

[0078] Initialize the particle swarm. Each particle represents a path scheme, and each position of the particle represents a path, which consists of multiple path parameters. The particle swarm is initialized by randomly generating multiple path parameters, and each particle is assigned a random velocity. The path parameters include: geometric parameters, motion parameters, and cooperative parameters. The geometric parameters include: path point coordinates x, y, z. The motion parameters include welding speed. The cooperative parameters include robotic arm allocation.

[0079] In each iteration, the objective function value corresponding to the path scheme of each particle is evaluated. For each particle, four objective functions f1(p), f2(p), f3(p), and f4(p) for its corresponding path are calculated. The calculated objective function values ​​are used to evaluate the quality of the path and to calculate the particle's comprehensive objective function value f. total (p):

[0080] f total (p)=ω1f1(p)+ω2f2(p)+ω3f3(p)+ω4f4(p)

[0081] Where ω1, ω2, ω3, and ω4 are the weight coefficients of each objective function, satisfying:

[0082] ω1+ω2+ω3+ω4=1

[0083] In each iteration, the particle's velocity and position are adjusted based on its historical best solution p. best and the global optimal solution g best Update. The update rule, combining inertia weights and acceleration constants with historical experience and global information, ensures that the particle swarm can effectively search for the optimal solution. The particle velocity and position update formulas are as follows:

[0084]

[0085] x i (t+1)=x i (t)+v i (t+1)

[0086] Among them, v i (t) is the velocity of the i-th particle at time t, x i (t) is the position of the particle, p best It is the historical optimal position of the particle, g best ω is the global optimal position, c1 and c2 are acceleration constants, r1 and r2 are random factors, and ω is the inertia weight.

[0087] Through multiple iterations, the particle swarm continuously updates its position and velocity, gradually converging to an optimal path. After each iteration, the global optimal solution g is updated. best And continue to optimize until the preset stopping criteria are met.

[0088] In each iteration, the MOPSO algorithm calculates the comprehensive objective function value for each particle and selects the optimal path using a weighted method.

[0089] Minimizing deformation is the most important optimization objective, as large deformation will affect the welding quality, so it is given the highest weight.

[0090] ω2=0.5

[0091] Then, weights are assigned to the other objectives as follows:

[0092] Welding efficiency: While welding efficiency is a key objective in path planning, it cannot completely dominate the optimization process. Therefore, it is typically assigned a relatively low weight to avoid sacrificing quality for time.

[0093] ω1=0.2

[0094] Path equalization: To avoid overloading certain path segments, appropriate weights are assigned to them.

[0095] ω3=0.1

[0096] Calculate reachability: To ensure path reachability, assign appropriate weights:

[0097] ω4=0.2

[0098] Step 4: Through multi-arm collaborative welding, welding tasks are rationally allocated to avoid overloading a single robotic arm and ensure overall welding efficiency. During the welding process, vision and laser sensors are used for real-time path tracking and automatic correction of path errors. Specifically, the following steps are taken: The goal of welding path tracking is to ensure that the robot welds along a predetermined path and to ensure the accuracy of the welding trajectory. The robot's current position is set as P. current (t) = (x(t), y(t), z(t)), and the target path point is P. target (t)=(x target (t),y target (t),z target (t)), where t is time.

[0099] The path error e(t) is defined as the difference between the target position and the current robot position:

[0100] e(t) = P target (t)-P current (t)

[0101] To achieve accurate path tracking, error-based control algorithms are used to correct the path. For example, PID control is used to calculate the error correction amount:

[0102]

[0103] Among them, K p K i and K d These are the proportional, integral, and derivative coefficients. Adjusting these parameters allows the robot to move more precisely along a predetermined path. ΔP(t) represents the path correction, and e(t) is the path error. The goal is to minimize this error, ensuring the welding path matches the predetermined path.

[0104] When a robot is performing a welding task, if the path deviates from the predetermined trajectory due to errors or disturbances (such as inaccurate robot movement or welding heat effects), the path correction mechanism can adjust the path in real time based on feedback information to ensure the accuracy of the welding process.

[0105] The offset of each path point of the robot is set as ΔP(t). The path correction model adjusts the correction value by calculating the real-time deviation of the current path point and the robot's motion error. The specific path correction function is as follows:

[0106] ΔP(t)=P target (t)-P current (t)+f correction (t)

[0107] Among them, f correction (t) is the correction function, which takes into account factors such as path deviation, welding position error, and environmental disturbances. The design of the correction function is based on real-time monitoring data and feedforward control to ensure the optimization of robot motion and welding accuracy.

[0108] During the welding process, to ensure high precision of the welding path, it is necessary to correct the path in real time to prevent the accumulation of path errors caused by robot motion errors, environmental factors, etc. Through real-time monitoring and correction methods, corrections are made according to the preset path error to ensure that the robot always stays near the optimal path during the actual welding process.

[0109] Step 5: As Figure 3 As shown, multimodal sensing technology is employed, utilizing multiple sensors such as vision, current, and noise to acquire key parameters during the welding process. The data is preprocessed, and welding defect detection is pre-trained using convolutional neural networks (CNN) and multilayer perceptrons (MLP). Intelligent adjustment of welding parameters is achieved based on deep Q-networks (DQN). A self-updating database mechanism is established, and dynamic training for welding defect detection is conducted. Specific strategies are developed for different welding defects such as porosity, cracks, and slag inclusions, thereby continuously optimizing the accuracy of welding quality parameter tuning. Specific operations include:

[0110] Preprocessing of visual sensor data includes:

[0111] Denoising: Gaussian filtering algorithm is used to denoise the weld seam image to smooth the image and remove external interference (such as sensor noise, illumination changes, etc.).

[0112] The Gaussian filtering formula is:

[0113]

[0114] Where G(x,y) is the Gaussian kernel and σ is the standard deviation.

[0115] Edge detection: The Canny edge detection algorithm is used to extract the edge features of the weld seam, enhancing key information in the image.

[0116] Preprocessing of current and noise data includes:

[0117] Normalization: Normalize the current and noise data to convert input data with different dimensions into the same range, so that different features have the same weight in influencing the model.

[0118] Smoothing: Fluctuations in current and noise data are eliminated through moving average filtering, improving data stability.

[0119] A convolutional neural network (CNN) is trained on weld seam images to learn local features within the weld seam and identify potential defects. The CNN automatically extracts hierarchical features from the image, especially when dealing with different welding defects.

[0120] Convolutional Neural Networks (CNNs) use the cross-entropy loss function for classification tasks, as shown in the following formula:

[0121]

[0122] Among them, y i It is the actual label. It represents the predicted probability, and N is the number of categories.

[0123] The multilayer perceptron (MLP) model is used to perform regression training on sensor data such as current and noise to predict the most suitable welding parameters (such as current and voltage).

[0124] The multilayer perceptron model uses the mean squared error loss function (MSE) for regression tasks, as shown in the following formula:

[0125]

[0126] Among them, y pred For the target welding parameters, y true N is the number of samples used to predict welding parameters.

[0127] The outputs of the Convolutional Neural Network (CNN) and the Multilayer Perceptron (MLP) are fused to form a feature vector. Specifically, the CNN and MLP extract image features and numerical features, respectively, generating two independent feature vectors. These two vectors are then concatenated into a single long vector, which is then fed into a Deep Q-Network (DQN).

[0128] Deep Q-Networks (DQNs) determine the parameters (current, voltage, etc.) that need to be adjusted based on fused features. Through a reinforcement learning framework, welding parameters are adjusted based on feedback after each welding operation, reducing defects and improving welding quality.

[0129] The Deep Q Network (DQN) will now be explained in more detail.

[0130] State Space: The current state space is formed by input sensor data (such as weld images, current, noise, etc.).

[0131] Action Space: The welding parameters that the system selects and adjusts based on the current state. The action space includes welding current, voltage, noise, etc.

[0132] Reward Function: The reward and penalty design is based on the feedback of welding quality. Positive rewards are given when there are few welding defects or the weld quality is good, and negative rewards are given otherwise.

[0133] Q-value update formula:

[0134]

[0135] Among them, s t As the current state, a t For the current action, r t For the reward, γ is the discount factor, and α is the learning rate.

[0136] A self-updating welding parameter adjustment database is established, including recording sensor data and welding results (such as defect type and adjusted welding parameters) into the database after each welding process. The database automatically stores historical data and feedback results of the welding process, gradually forming a training set with sufficient samples for future parameter adjustments during welding processes.

[0137] This embodiment specifies corresponding parameter adjustment strategies based on different types of welding defects to ensure stable welding quality and reduce the occurrence of defects; including:

[0138] Porosity: Porosity typically appears as small holes on the weld surface and may be caused by overheating of the molten pool or insufficient gas escape. We monitor porosity on the weld surface using visual sensors and analyze the causes based on current data. To address porosity, we reduce welding current and speed, increase molten pool fluidity, facilitate gas escape, and prevent porosity formation.

[0139] Cracks: Crack formation is usually related to temperature changes and thermal stress. Visual sensors can detect morphological changes on the weld surface. To reduce crack occurrence, welding current, voltage, and welding speed can be lowered, thereby reducing heat input and mitigating thermal stress during cooling, thus preventing crack formation.

[0140] Slag inclusions: Slag inclusions are usually caused by insufficient fluidity of the weld pool during welding, resulting in the failure to remove impurities in a timely manner. Slag inclusions on the weld surface are identified using visual sensors. To avoid slag inclusions, the welding current and voltage are increased to enhance weld pool fluidity, ensuring the removal of impurities during welding, and the wire feed rate is appropriately increased.

[0141] Incomplete penetration: Incomplete penetration is usually caused by insufficient welding current. Visual sensors are used to determine whether the weld has fully fused to the base material, identifying incomplete penetration issues. If incomplete penetration occurs, the welding current should be increased and the welding speed reduced to ensure sufficient melting of the weld area and avoid this problem.

[0142] Spatter: Spatter is usually caused by an unstable arc or excessive welding current. Arc stability and spatter patterns are analyzed using current and vision sensors. If spatter occurs, reducing the welding current and increasing the welding speed stabilizes the arc, reduces spatter, and ensures welding stability.

[0143] The current and voltage parameters involved in the aforementioned defects are adjusted through a deep Q-network, which can respond in real time to defects occurring during the welding process and precisely adjust welding parameters, thereby ensuring the stability and efficiency of welding quality.

[0144] This embodiment integrates the outputs of convolutional neural networks (CNN), multilayer perceptrons (MLP), and other sensor data. The system can effectively identify and perceive welding defects, comprehensively perceive welding quality through multimodal data such as vision, noise, and current, identify potential defects in real time, and ensure the stability and reliability of the welding process.

[0145] As the database continuously accumulates new data, the model will undergo incremental training or online learning based on the new samples to continuously optimize the accuracy of parameter tuning. Furthermore, data from each welding operation will be used to update the model, which will then be retrained based on the new data to adapt to new welding conditions and further optimize parameter tuning.

[0146] Example 2:

[0147] This embodiment proposes a robot welding system based on parametric programming, which mainly includes a multi-arm collaborative welding platform, a multi-objective path planning module, a welding quality multimodal perception and feedback adjustment system, a path real-time monitoring and correction module, a control system and an integration unit.

[0148] The multi-robotic arm collaborative welding platform includes multiple robotic arms, a welding power source, an automatic wire feeder, and a welding path tracking system. Each robotic arm comprises multiple multi-degree-of-freedom robots capable of collaborative operation. By precisely controlling the motion trajectory of each robotic arm, efficient welding tasks can be achieved. The automatic wire feeder, through coordinated control with the welding robots and the power system, automatically adjusts the wire feeding speed and welding current to ensure welding quality. In this embodiment, the multi-robotic arm collaborative welding platform can adopt a modular design, supporting the expansion and customization of multiple robotic arms, and can flexibly configure different numbers of robotic arms according to production needs.

[0149] The multi-objective path planning module, based on deep reinforcement learning and multi-objective optimization algorithms, comprehensively considers objectives such as welding deformation, welding reachability, path efficiency, and shortest path to generate optimized welding paths and motion parameters. Specifically, this module includes a welding task allocation module, a path generation module, and a motion control module. The welding task allocation module automatically and rationally distributes welding tasks to each robotic arm based on the workpiece's geometric features and welding requirements, avoiding overload or uneven load distribution to any single robotic arm. The path generation module optimizes the path based on the welding task requirements using a multi-objective particle swarm optimization (MOPSO) algorithm to generate the optimal path scheme. This ensures minimal welding path deformation, even task distribution to each robotic arm, avoidance of motion interference, and path minimization, thereby improving welding efficiency and accuracy. The motion control module is responsible for controlling the robotic arm's movement, achieving precise welding through accurate trajectory control algorithms (such as path compensation and dynamic adjustment).

[0150] The welding quality multimodal sensing and feedback adjustment system integrates vision sensors, current sensors, and welding noise sensors to monitor welding quality in real time. Specifically, it analyzes the collected real-time data to identify welding defects (such as cracks and porosity) and automatically adjusts welding parameters such as welding current and voltage based on the defect type. This includes: a weld morphology monitoring module, which uses vision sensors to detect the weld morphology in real time, analyze the presence of defects, and adjust welding parameters based on the analysis results; and a current and voltage adjustment module, which monitors welding current and voltage in real time and intelligently adjusts power parameters using a deep Q-network (DQN) based on changes in weld quality during the welding process to ensure the stability of welding quality.

[0151] The real-time path monitoring and correction module acquires welding path data in real time through vision and laser sensors. Using a feedback algorithm, it automatically detects path deviations and corrects them, ensuring real-time welding path tracking and consistency with the predetermined trajectory. This ensures precise execution of the welding path and reduces path errors. Specifically, it includes a path tracking and correction module, which acquires welding path data in real time using vision and laser sensors and corrects trajectory deviations using a feedback algorithm; and a position detection and compensation module, which detects weld offset based on sensor feedback data and automatically adjusts the welding path through a feedback compensation mechanism to ensure accurate weld position. The control system coordinates with various subsystems (welding robot, welding power supply, sensor modules, etc.) through an integrated unit, managing system operations and data flow. This includes an operating terminal and a human-machine interface. The operating terminal allows users to input welding process parameters such as welding current, welding speed, voltage, welding arc length, and wire feed speed, monitoring the welding process in real time and dynamically adjusting welding parameters based on system feedback data to ensure welding accuracy and quality.

[0152] In this embodiment, a self-updating parameter tuning database is also included, which records historical data and feedback results in real time during the welding process. Through incremental training and online learning, the parameter tuning accuracy is continuously optimized, the adaptability of the welding process is improved, and the system's ability to adapt to different production environments and welding tasks is enhanced.

Claims

1. A robot welding method based on parametric programming, characterized in that: Includes the following steps: Step 1: Collect workpiece surface data and extract welding-related geometric features from the workpiece surface data; Step 2: Based on the geometric features extracted in Step 1, perform path planning to obtain the welding path; Step 3: Based on the welding path and welding parameters, perform multi-arm collaborative welding. During the welding process, the welding path is tracked in real time, and key parameters are collected to determine whether the path error is less than a set threshold. If it is less than the set threshold, the current welding quality is further assessed. If it meets the standard, the welding ends. If it does not meet the standard, the welding parameters are adjusted according to the key parameters during the welding process, and Step 3 is executed. If the error is not less than the set threshold, the welding path is corrected according to the path error, and Step 3 is executed. In step 2, the path planning based on the geometric features extracted in step 1 to obtain the welding path specifically includes: Step 1: Construct multiple objective functions, including: welding efficiency objective function, deformation objective function, path equalization objective function, and welding accessibility objective function; the welding efficiency objective function is to minimize the total welding time; the deformation objective function is to minimize the deformation on the welding path; the path equalization objective function is to balance the workload of each robotic arm; and the welding accessibility objective function is to ensure that each path point is within the reachability space. Step 2: Employ a multi-objective particle swarm optimization algorithm. Each particle represents a path scheme, and each position of the particle represents a path. Through the global search capability of the particle swarm, the multi-objective function is balanced and optimized to obtain the optimal path, which is the welding path. The aforementioned adjustment of welding parameters based on key parameters during the welding process includes the following specific operations: Key parameters in the welding process include: weld image, current data, and noise data; Using convolutional neural networks to extract local features of welds from weld images; Using a multilayer perceptron, predicted values ​​of welding parameters are obtained based on current and noise data; The outputs of the convolutional neural network and the multilayer perceptron are fused together. The fused features are then input into a deep Q-network. The deep Q-network determines the welding parameters that need to be adjusted based on the fused features, and then adjusts the welding parameters accordingly.

2. The robot welding method based on parametric programming according to claim 1, characterized in that: The objective function for welding efficiency is expressed as: ； In the formula, Indicates welding time. This indicates the length of each path segment. This represents the welding speed, and n represents the total number of path segments.

3. The robot welding method based on parametric programming according to claim 2, characterized in that: The objective function for the deformation amount is expressed as: ； In the formula, Let be the predicted deformation value of the i-th diameter segment; The predicted values ​​are obtained by using a trained physical information neural network model; the physical information neural network model takes the temperature field distribution and thermal stress field during the welding process as inputs and outputs the corresponding predicted values ​​of deformation.

4. The robot welding method based on parametric programming according to claim 3, characterized in that: The objective function for the path equalization is expressed as: ； In the formula, Let be the load of the i-th path segment.

5. The robot welding method based on parametric programming according to claim 4, characterized in that: The objective function for welding accessibility is expressed as: ； in, It is an reachability judgment function; if the path point If it is within the reachable space, the value is 1; otherwise, it is 0.

6. The robot welding method based on parametric programming according to claim 5, characterized in that: The specific operations in Step 2 include: Each particle represents a path scheme, and each position of the particle represents a path, which consists of multiple path parameters. The particle swarm is initialized by randomly generating multiple path parameters, and each particle is assigned a random velocity. The path parameters include: geometric parameters, motion parameters, and cooperative parameters. The geometric parameters include: path point coordinates. The motion parameters include welding speed. The cooperative parameters include robotic arm allocation. In each iteration, for each particle, four objective functions for its corresponding path scheme are calculated. The calculated objective function value will be used to evaluate the quality of the path, and the comprehensive objective function value of the particle will be calculated according to the following formula. : ； in, Let be the weight coefficients of each objective function, and satisfy: ； Minimizing the deformation is taken as the most important optimization objective; therefore... ; In each iteration, the particle's velocity and position are updated according to the following formula: ； ； in, It is the velocity of the i-th particle at time t. It is the position of the particle at time t. It is the best position in the history of the particle. It is the globally optimal position. , It is the acceleration constant. , It is a random factor. It is inertial weight; Through multiple iterations, the particle swarm continuously updates its position and velocity, and after each iteration, it updates the global optimal solution. The algorithm continues to optimize until the preset stopping criteria are met, thus obtaining the optimal path.

7. The robot welding method based on parametric programming according to claim 1, characterized in that: Also includes: Establish a self-updating welding parameter adjustment database; After each welding process is completed, record the key parameters of the current welding process, as well as the corresponding welding parameters and welding defect types; We used data from a self-updating welding parameter tuning database to dynamically train a convolutional neural network for welding defect detection.

8. A robotic welding system based on parametric programming, characterized in that: The invention includes a multi-arm robot, a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the robot welding method based on parametric programming as described in any one of claims 1 to 7.

Images