A multi-rate asynchronous sensor fusion welding control system and method

Abstract

The application discloses a kind of multi-rate asynchronous sensing fusion's welding control system and method, it is related to welding automation and intelligent control technical field, including visual sensor, torque sensor, thermal imaging sensor, multi-rate state estimation module, online parameter identification module, model predictive control module, welding robot motion controller and welding power supply.The state vector including fast variable and slow variable is constructed, and the asynchronous fusion of sensing data of different sampling rates is carried out using multi-rate extended Kalman filter, and the state estimation of welding process is realized;Recursion least square method is used to update model parameters online;Based on model predictive control, the trajectory of welding torch, welding speed, current and voltage are optimized, and the feedforward compensation is generated according to temperature field and thermal deformation state.The application improves the sensing data fusion precision and system adaptability, realizes the active compensation of thermal deformation, improves the welding seam tracking precision and welding quality stability.

Description

Technical Field

[0001] This invention relates to the field of welding automation and intelligent control technology, and in particular to a welding control system and method with multi-rate asynchronous sensing fusion. Background Technology

[0002] Welding, as an important processing technology in modern intelligent manufacturing, is widely used in shipbuilding, automobile manufacturing, rail transportation, aerospace, and pressure vessels. To improve welding accuracy and automation, existing welding systems are typically equipped with detection devices such as vision sensors, force sensors, and temperature sensors, and are combined with welding robots to achieve weld seam tracking and welding parameter control.

[0003] However, existing welding control technologies still have the following problems in practical applications: 1. During welding, it is typically necessary to simultaneously collect information such as weld position, molten pool condition, contact force, and temperature field. The sampling frequencies of different sensors vary significantly. For example, the sampling frequency of vision sensors is usually tens to hundreds of hertz, torque sensors can reach kilohertz levels, while thermal imaging sensors typically only have sampling frequencies of tens of hertz. Most existing fusion methods use a uniform frequency processing approach, requiring downsampling of high-frequency data and interpolation compensation for low-frequency data. This can easily lead to loss of effective information or introduce additional errors, making it difficult to accurately reflect the real-time status of the welding process. 2. During the welding process, factors such as differences in workpiece material, variations in plate thickness, fluctuations in assembly gaps, and changes in heat dissipation conditions can cause changes in model parameters such as thermal conductivity and thermal expansion coefficient. Most existing control systems use pre-established fixed models for control, making it impossible to adjust model parameters in real time according to actual working conditions. This can easily lead to model mismatch, affecting the accuracy of welding trajectory tracking and the stability of welding quality. 3. During the welding process, the workpiece undergoes thermal expansion and contraction deformation after being heated, causing the actual weld position to deviate from the preset trajectory. Existing technologies mainly rely on visual feedback or post-weld inspection results for compensation, which is essentially a reactive or passive compensation method. It cannot predict the development trend of thermal deformation in advance and adjust the welding trajectory and process parameters in a timely manner, thus easily causing problems such as weld misalignment, uneven penetration, and reduced welding quality.

[0004] Therefore, there is an urgent need for a welding adaptive control system and control method that can achieve multi-rate asynchronous sensor data fusion, online updating of welding process model parameters, and real-time prediction and active compensation of welding thermal deformation. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a welding control system and method with multi-rate asynchronous sensing fusion to solve the problems existing in the background art.

[0006] This invention provides the following technical solution: a welding control system and method based on multi-rate asynchronous sensor fusion, comprising: A vision sensor is used to acquire positional features of the weld seam and molten pool morphology on the workpiece. Torque sensor is used to acquire contact force information between the welding torch and the workpiece; Thermal imaging sensors are used to acquire temperature field information in the welding area; The multi-rate state estimation module is connected to the vision sensor, torque sensor and thermal imaging sensor. It is used to construct a system state vector containing fast and slow variables, and asynchronously fuse sensor data with different sampling rates based on multi-rate extended Kalman filtering to obtain the welding process state estimation result. An online parameter identification module, connected to the multi-rate state estimation module, is used to identify welding process model parameters in real time based on state estimation results and sensor measurement results. The model prediction control module is connected to the online parameter identification module. It is used to establish a prediction model based on the identified model parameters, and to perform collaborative optimization using welding torch trajectory correction, welding speed, welding current and welding voltage as optimization variables. It also generates feedforward compensation control quantity based on the predicted thermal deformation. A welding robot, wherein the execution end of the welding robot is equipped with a welding torch, wherein the welding robot is connected to the model prediction control module, and is used to control the movement of the welding torch along the weld seam on the workpiece according to the welding torch trajectory correction amount and the welding speed; The welding power source is connected to the model prediction and control module and is used to adjust the welding process parameters according to the welding current control quantity and the welding voltage control quantity.

[0007] Preferably, the fast variables include the welding torch tip position, the welding torch tip speed, and the contact force between the welding torch and the workpiece, and the slow variables include the workpiece temperature field state, the thermal deformation state, and the welding process model parameters.

[0008] Preferably, the multi-rate state estimation module uses the sensor with the highest sampling frequency as the reference frequency to perform the prediction step, and triggers the corresponding measurement update step when the corresponding sensor generates new measurement data.

[0009] Preferably, the position observation model corresponding to the visual sensor is used to extract weld edge features and molten pool geometric features, the observation model corresponding to the torque sensor is used to obtain the contact force state, and the observation model corresponding to the thermal imaging sensor is used to obtain the temperature field state and thermal deformation state.

[0010] Preferably, the online parameter identification module uses recursive least squares or recursive least squares with forgetting factor to update at least one of the thermal conductivity coefficient, thermal expansion coefficient and heat dissipation parameters in real time.

[0011] Preferably, the online parameter identification module transmits the real-time updated model parameters to the model prediction control module to update the prediction model.

[0012] Preferably, the model predictive control module constructs an objective function that includes trajectory tracking error, contact force error, temperature field error, and a penalty term for the rate of change of control input, and solves for the optimal control quantity in the prediction time domain.

[0013] Preferably, the model prediction control module predicts the future thermal deformation trend based on the temperature field state and thermal deformation state, and adjusts the welding torch trajectory before thermal deformation occurs to achieve thermal deformation feedforward compensation.

[0014] Preferably, the vision sensor includes one or more of an industrial camera, a binocular vision camera, and an event camera.

[0015] A welding control method based on multi-rate asynchronous sensor fusion includes the following steps: S1: Acquire welding process data collected by vision sensors, torque sensors, and thermal imaging sensors; S2: Construct a system state vector containing fast and slow variables; S3: Multi-rate extended Kalman filtering is used to asynchronously fuse sensor data at different sampling rates to obtain welding process state estimation results; S4: Identify welding process model parameters in real time based on state estimation results; S5: Establish a model predictive control model based on the identified model parameters; S6: Solve the problem by using welding torch trajectory correction, welding speed, welding current, and welding voltage as joint optimization variables; S7: Generate thermal deformation feedforward compensation based on temperature field and thermal deformation prediction results; S8: Control the welding robot and welding power source to complete the welding operation based on the optimization results.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention constructs a system state vector containing fast and slow variables, and uses multi-rate extended Kalman filtering to fuse asynchronous data collected by visual sensors, torque sensors, and thermal imaging sensors. This allows sensors with different sampling frequencies to participate in state updates according to their respective sampling rhythms, without the need to unify all sensor data to the same sampling frequency for processing. This not only avoids the information loss caused by downsampling of high-speed sensor data, but also avoids the estimation error caused by interpolation of low-speed sensor data, thereby improving the accuracy and real-time performance of welding process state estimation. 2. This invention, by setting up an online parameter identification module, continuously updates model parameters such as thermal conductivity coefficient, thermal expansion coefficient, and heat dissipation parameters based on real-time state estimation results and sensor measurement results during the welding process. The updated parameters are then fed back to the model predictive control module in real time, enabling the predictive model to dynamically adapt to complex working conditions such as changes in workpiece material, plate thickness, and heat dissipation conditions. This reduces the impact of model mismatch on control performance, thereby improving the adaptability of the welding control system to changes in working conditions and the stability of welding quality. 3. This invention utilizes temperature field information obtained from thermal imaging sensors and thermal deformation state obtained from state estimation to predict future thermal deformation trends during model predictive control. It also uses welding torch trajectory correction, welding speed, welding current, and welding voltage as joint optimization variables for collaborative optimization. Before thermal deformation occurs, trajectory compensation and process parameter adjustment are generated in advance to achieve active prediction and feedforward compensation of thermal deformation. This effectively reduces weld seam offset and welding quality fluctuations caused by thermal deformation during welding, and improves weld seam tracking accuracy and welding consistency. 4. This invention organically combines multi-rate asynchronous state estimation, online parameter identification, and thermal deformation prediction and compensation to construct a closed-loop control system integrating perception, estimation, prediction, decision-making, and execution. This system achieves synergistic optimization of welding trajectory control and welding process parameter control, further improving welding quality, control accuracy, and system robustness while ensuring real-time control. It is applicable to various welding scenarios such as arc welding, laser welding, and robotic automatic welding. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall system structure of the present invention.

[0018] Figure 2 This is a schematic diagram illustrating the principle of the multi-rate asynchronous state estimator of the present invention.

[0019] Figure 3 This is a schematic diagram of the hierarchical structure of the system state vector of the present invention.

[0020] Figure 4 This is a flowchart of the multi-rate extended Kalman filter of the present invention.

[0021] Figure 5 This is a schematic diagram illustrating the working principle of the online model parameter identification module of the present invention.

[0022] Figure 6 This is a schematic diagram of the model predictive control collaborative optimization of the present invention.

[0023] Figure 7 This is a schematic diagram showing the effect comparison of the thermal deformation feedforward compensation of the present invention.

[0024] The attached figures are labeled as follows: 1. Vision sensor; 2. Torque sensor; 3. Thermal imaging sensor; 4. Multi-rate state estimation module; 5. Online parameter identification module; 6. Model predictive control module; 7. Welding robot; 8. Welding power source; 9. Welding torch; 10. Workpiece; 11. Weld seam. Detailed Implementation

[0025] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby providing a clearer and more explicit definition of the scope of protection of the present invention.

[0026] This invention provides a welding control system and method based on multi-rate asynchronous sensor fusion, such as... Figure 1-7 As shown, it includes a vision sensor 1, a torque sensor 2, a thermal imaging sensor 3, a multi-rate state estimation module 4, an online parameter identification module 5, a model prediction control module 6, a welding robot 7, a welding power source 8, a welding torch 9, a workpiece 10, and a weld seam 11.

[0027] The vision sensor 1 is used to acquire in real time the positional features of the weld 11 on the surface of the workpiece 10 and the morphological information of the molten pool during the welding process. Specifically, the vision sensor 1 can be one or more of an industrial camera, a binocular vision camera, a structured light vision device, or an event camera. It extracts feature parameters such as the weld centerline position, weld edge contour, molten pool width, molten pool length, and molten pool area through image processing algorithms, providing a basis for weld tracking and molten pool condition monitoring.

[0028] The torque sensor 2 is installed at the end of the welding torch 9 or between the welding robot end effector and the welding torch 9, and is used to measure the contact force information between the welding torch 9 and the workpiece 10 in real time. The contact force information includes force components and torque components along three coordinate directions. By monitoring the changes in contact force, the changes in welding torch posture, workpiece assembly errors, and mechanical disturbances generated during the welding process can be reflected.

[0029] The thermal imaging sensor 3 is positioned around the welding area to acquire infrared thermal image information of the welding area. The data collected by the thermal imaging sensor 3 includes the temperature of the molten pool, the temperature of the heat-affected zone, and the temperature field distribution information on the surface of the workpiece 10. By continuously monitoring the temperature change process, it can reflect the heat transfer and the trend of thermal deformation.

[0030] It should be noted that the sampling frequencies of the visual sensor 1, torque sensor 2, and thermal imaging sensor 3 typically differ significantly. For example, the sampling frequency of the visual sensor 1 is generally 50–200 Hz, the sampling frequency of the torque sensor 2 is generally 1000–5000 Hz, and the sampling frequency of the thermal imaging sensor 3 is generally 20–100 Hz. Existing technologies usually employ a uniform sampling frequency processing method, requiring downsampling for high-frequency data and interpolation compensation for low-frequency data, which can easily lead to information loss and estimation errors. This invention improves the accuracy of state estimation by constructing a multi-rate asynchronous state estimation mechanism, enabling various sensors to participate in state updates according to their own sampling rhythm.

[0031] In this embodiment, the welding process state is divided into fast variables and slow variables, and a system state vector is constructed: ; Among them, X fast X represents the set of fast variable states; slow This represents the set of slow variable states.

[0032] Specifically, fast variables include: welding torch tip position, welding torch tip speed, and contact force between the welding torch and the workpiece. Slow variables include: workpiece temperature field state, thermal deformation state, and welding process model parameters.

[0033] Because the welding torch movement and contact force change rapidly, their time constants are typically 1 to 10 ms; while the temperature field changes and thermal deformation changes slowly, their time constants are typically 0.1 s to several seconds. Therefore, adopting a hierarchical modeling approach for fast and slow variables can improve the efficiency of state estimation.

[0034] Furthermore, the multi-rate state estimation module 4 uses the multi-rate extended Kalman filter algorithm for state estimation.

[0035] The prediction step is performed using the sensor with the highest sampling frequency as the reference frequency, for example, using the 1000Hz sampling frequency of torque sensor 2 as the prediction clock, and is executed in each control cycle: ; Where f is the nonlinear state transition function of the welding process (including simplified models of rigid body kinematics, contact mechanics, and thermodynamics), u is the control input (welding speed, current, voltage, etc.), F is the Jacobian of the state transition matrix, and Q is the process noise covariance.

[0036] The state transition function includes the welding torch kinematics model, the contact mechanics model, and the heat conduction model.

[0037] Furthermore, when any sensor generates new measurement data, a measurement update step is immediately triggered.

[0038] Position observation model corresponding to visual sensor 1: h vision (x) is used to extract the weld edge location and molten pool geometry parameters; The observation model corresponding to torque sensor 2: h force (x) is used to obtain the contact force state; The observation model corresponding to thermal imaging sensor 3: h thermal (x) is used to obtain the temperature field state and thermal deformation state.

[0039] Measurement updates are performed using the extended Kalman filter formula: ; Where K is the Kalman gain matrix, H is the observation matrix, R is the measurement noise covariance matrix, z is the sensor measurement value, and I is the identity matrix.

[0040] Since the measurement times of each sensor are different, the multi-rate state estimation module 4 maintains an event time queue and performs corresponding observation model updates whenever a new measurement event arrives, thereby achieving asynchronous fusion.

[0041] The above structure avoids information loss caused by downsampling of high-frequency sensor data and interpolation of low-frequency data, thereby improving the accuracy of welding process state estimation.

[0042] The online parameter identification module 5 is used to solve the problem of drift in welding process model parameters as the working conditions change.

[0043] During the welding process, the workpiece material, workpiece thickness, ambient temperature, and heat dissipation conditions may all change, resulting in changes in the thermal conductivity coefficient, thermal expansion coefficient, and heat dissipation parameters.

[0044] Therefore, this embodiment uses the recursive least squares method for online parameter identification.

[0045] First, based on the current state estimation results and control inputs, the model predicts the output: ; Where: g(·) represents the welding process model, θ This represents the vector of parameters to be identified.

[0046] Then, the prediction error is calculated based on the actual measured values: ; Where: y meas This represents the actual measured value.

[0047] Then update the model parameters: ; in: θ new Indicates the updated parameters, θ old Indicates the parameters before the update, K rls This represents the recursive least-squares gain matrix.

[0048] In a preferred embodiment, a recursive least squares method with a forgetting factor can also be used to improve the system's ability to track time-varying parameters.

[0049] The updated model parameters are transmitted in real time to the model prediction control module 6 for updating the prediction model.

[0050] Online identification enables the predictive model to maintain consistency with the actual welding process, thereby improving control accuracy.

[0051] The model prediction control module 6 establishes a prediction model based on the state estimation results and the online updated model parameters.

[0052] Optimization variables in the prediction time domain N p Includes: welding torch position correction amount Δ P Welding speed v Welding current I Welding voltage U The objective function to be optimized is: ; Where: p ref For the target trajectory, p actual For predicting trajectory, F ref For the target contact force, F actual To predict contact force, T ref For the target temperature field, T actual For predicting the temperature field, Δu is the change in control input; Q1, Q2, Q3 are the error weight matrices; R is the control weight matrix. N p For predicting the time domain, N c To control the time domain.

[0053] Simultaneously, actuator constraints and welding quality constraints are set. Actuator constraints include: maximum torch speed, maximum torch acceleration, and welding power output range. Welding quality constraints include: molten pool temperature range, heat input range, and weld formation requirements.

[0054] Preferably: N p =10~20; N c =3~5; The control cycle is set to 5-10ms.

[0055] The solution can be obtained using a sequential quadratic programming algorithm or a real-time iterative Newton's method.

[0056] Furthermore, the model prediction control module 6 uses the temperature field state and thermal deformation state obtained from the state estimation to predict the future thermal deformation trend.

[0057] When the prediction results indicate that the workpiece will undergo thermal deformation at a future moment, the model prediction control module 6 generates the welding torch trajectory compensation amount in advance and adjusts the welding speed, current and voltage simultaneously.

[0058] For example, when it is predicted that the workpiece will undergo thermal deformation to the left, the model prediction control module 6 controls the welding torch to shift to the right by a certain distance in advance for pre-compensation, so as to ensure that the actual weld trajectory remains near the target position.

[0059] At the same time, by reducing the welding current or increasing the welding speed, the heat input per unit length is reduced, thus suppressing further expansion of thermal deformation.

[0060] This enables a shift from traditional passive correction to proactive predictive compensation.

[0061] A welding control method based on multi-rate asynchronous sensor fusion includes the following steps: S1. Welding process data acquisition After welding begins, vision sensor 1, torque sensor 2, and thermal imaging sensor 3 are put into operation simultaneously. Among them, vision sensor 1 acquires images of the weld seam 11 and the molten pool on the surface of workpiece 10 in real time, and extracts feature parameters such as the weld center position, weld edge contour, molten pool width, molten pool length, and molten pool area through image processing algorithms.

[0062] The torque sensor 2 detects the contact force information between the welding torch 9 and the workpiece 10 in real time, including force components in three directions and torque components, which are used to reflect changes in the welding torch posture and mechanical disturbances.

[0063] The thermal imaging sensor 3 acquires infrared thermal images of the welding area in real time and obtains information on the temperature of the molten pool, the temperature of the heat-affected zone, and the temperature field distribution on the surface of the workpiece 10.

[0064] The collected multi-source data are sent to the multi-rate state estimation module 4 respectively.

[0065] S2: Multi-rate asynchronous state estimation Multi-rate state estimation module 4 first establishes the welding process state vector: ; Subsequently, a multi-rate extended Kalman filter algorithm is used for state estimation; the system uses torque sensor 2 with the highest sampling frequency as the prediction clock, and performs state prediction in each control cycle. ; When the vision sensor 1, torque sensor 2, or thermal imaging sensor 3 generates new measurement data, an immediate measurement update is performed: ; Data fusion from sensors with different sampling frequencies is achieved through asynchronous updates, thereby obtaining real-time welding status estimation results.

[0066] S3: Online Model Parameter Identification After receiving the state estimation results, the online parameter identification module 5 updates the welding process model parameters in real time.

[0067] First, calculate the predicted output based on the current state estimate and control input: ; Then obtain the actual measurement output: y pred ; Calculate the model error: ; The model parameters are then updated using a recursive least squares algorithm. ; The updated parameters are fed back to the model prediction and control module 6 in real time.

[0068] Online identification can compensate for model mismatch problems caused by material differences, plate thickness variations, and environmental changes.

[0069] S4: Establish a prediction model; The model prediction control module 6 receives the state estimation results and the updated model parameters.

[0070] A prediction model was constructed using the kinematics model of the welding torch, the contact mechanics model, and the heat conduction model.

[0071] The predictive model can predict changes in welding torch position, contact force, temperature field, and thermal deformation over several future control cycles. This provides a foundation for subsequent optimization calculations. S5: Joint Optimization Solution Model predictive control module 6 uses welding torch trajectory correction amount Δp Welding speed v Welding current I and welding voltage U are used as optimization objects, and an objective function is constructed as follows: ; The objective function is minimized using a sequential quadratic programming algorithm or a real-time iterative Newton's method. This yields the optimal control sequence for future control cycles.

[0072] S6: Thermal Deformation Prediction and Feedforward Compensation The model prediction control module 6 uses a prediction model to predict future thermal deformation trends.

[0073] When the prediction result indicates that workpiece 10 will undergo thermal deformation at a future time, the amount of thermal deformation is first calculated. Δd, Then, a compensation trajectory is generated based on the direction of thermal deformation. Δp = f(Δd) For example, when thermal deformation of the workpiece to the left is predicted, the control system pre-emptively shifts the welding torch to the right by the corresponding compensation amount. Simultaneously, based on the prediction results, it adjusts the welding speed, welding current, and welding voltage to reduce heat input and suppress further thermal deformation.

[0074] S7: Welding torch motion control The model prediction control module 6 sends the optimization results to the welding robot motion controller 7.

[0075] The welding robot motion controller 7 generates robot motion commands based on trajectory correction and welding speed. It controls the welding torch 9 to move along the weld seam 11 on the workpiece 10 according to the target trajectory. Simultaneously, it corrects the welding torch posture in real time to ensure the torch is always in the optimal welding position.

[0076] S8: Welding process parameter control The model prediction control module 6 synchronously sends process control commands to the welding power source 8.

[0077] The welding power source 8 adjusts the welding current, welding voltage, and heat input power in real time according to control commands, thereby maintaining the molten pool temperature within the target range.

[0078] During the welding process, steps S1 to S8 are continuously executed in a loop, allowing visual, force, and thermal information to continuously participate in state estimation and control optimization, ultimately forming a closed-loop control system that integrates perception, estimation, identification, prediction, control, and execution.

[0079] Through the above-described process, this invention can improve the utilization rate of multi-source sensor information, improve the accuracy of welding condition estimation, enhance the system's adaptability to changes in working conditions, and achieve active prediction and compensation of thermal deformation, thereby improving weld tracking accuracy and welding quality stability.

[0080] Several points should be noted: First, in the description of this application, it should be noted that, unless otherwise specified and limited, the terms "installation", "connection" and "linkage" should be interpreted broadly, and can be mechanical or electrical connection, or internal connection between two components, or direct connection. "Up", "down", "left", "right", etc. are only used to indicate relative positional relationship. When the absolute position of the described object changes, the relative positional relationship may change.

[0081] The above description is only a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. Any equivalent modifications or changes made by those skilled in the art based on the content disclosed in the present invention should be included within the scope of protection set forth in the claims.

Claims

1. A welding control system with multi-rate asynchronous sensor fusion, characterized in that, include: A vision sensor is used to acquire positional features of the weld seam and molten pool morphology on the workpiece. Torque sensor is used to acquire contact force information between the welding torch and the workpiece; Thermal imaging sensors are used to acquire temperature field information in the welding area; The multi-rate state estimation module is connected to the vision sensor, torque sensor and thermal imaging sensor. It is used to construct a system state vector containing fast and slow variables, and asynchronously fuse sensor data with different sampling rates based on multi-rate extended Kalman filtering to obtain the welding process state estimation result. An online parameter identification module, connected to the multi-rate state estimation module, is used to identify welding process model parameters in real time based on state estimation results and sensor measurement results. The model prediction control module is connected to the online parameter identification module. It is used to establish a prediction model based on the identified model parameters, and to perform collaborative optimization using welding torch trajectory correction, welding speed, welding current and welding voltage as optimization variables. It also generates feedforward compensation control quantity based on the predicted thermal deformation. A welding robot, wherein the execution end of the welding robot is equipped with a welding torch, wherein the welding robot is connected to the model prediction control module, and is used to control the movement of the welding torch along the weld seam on the workpiece according to the welding torch trajectory correction amount and the welding speed; The welding power source is connected to the model prediction and control module and is used to adjust the welding process parameters according to the welding current control quantity and the welding voltage control quantity.

2. The welding control system based on multi-rate asynchronous sensor fusion according to claim 1, characterized in that: The fast variables include the welding torch tip position, welding torch tip speed, and contact force between the welding torch and the workpiece; the slow variables include the workpiece temperature field state, thermal deformation state, and welding process model parameters.

3. The welding control system based on multi-rate asynchronous sensor fusion according to claim 1, characterized in that: The multi-rate state estimation module uses the sensor with the highest sampling frequency as the reference frequency to perform the prediction step, and triggers the corresponding measurement update step when the corresponding sensor generates new measurement data.

4. The welding control system with multi-rate asynchronous sensor fusion according to claim 3, characterized in that: The position observation model corresponding to the visual sensor is used to extract weld edge features and molten pool geometric features, the observation model corresponding to the torque sensor is used to obtain the contact force state, and the observation model corresponding to the thermal imaging sensor is used to obtain the temperature field state and thermal deformation state.

5. A welding control system based on multi-rate asynchronous sensor fusion according to claim 1, characterized in that: The online parameter identification module uses recursive least squares or recursive least squares with forgetting factor to update at least one of the thermal conductivity coefficient, thermal expansion coefficient and heat dissipation parameters in real time.

6. The welding control system based on multi-rate asynchronous sensor fusion according to claim 1, characterized in that: The online parameter identification module transmits the updated model parameters to the model prediction control module in real time to update the prediction model.

7. A welding control system based on multi-rate asynchronous sensor fusion according to claim 1, characterized in that: The model predictive control module constructs an objective function that includes trajectory tracking error, contact force error, temperature field error, and a penalty term for the rate of change of control input, and solves for the optimal control quantity in the prediction time domain.

8. A welding control system based on multi-rate asynchronous sensor fusion according to claim 7, characterized in that: The model prediction and control module predicts the future thermal deformation trend based on the temperature field state and thermal deformation state, and adjusts the welding torch trajectory before thermal deformation occurs to achieve thermal deformation feedforward compensation.

9. A welding control system based on multi-rate asynchronous sensor fusion according to claim 1, characterized in that: The vision sensor includes one or more of an industrial camera, a binocular vision camera, and an event camera.

10. A welding control method based on multi-rate asynchronous sensor fusion, wherein the welding control system based on multi-rate asynchronous sensor fusion as described in any one of claims 1-9 is characterized in that, Includes the following steps: S1: Acquire welding process data collected by vision sensors, torque sensors, and thermal imaging sensors; S2: Construct a system state vector containing fast and slow variables; S3: Multi-rate extended Kalman filtering is used to asynchronously fuse sensor data at different sampling rates to obtain welding process state estimation results; S4: Identify welding process model parameters in real time based on state estimation results; S5: Establish a model predictive control model based on the identified model parameters; S6: Solve the problem by using welding torch trajectory correction, welding speed, welding current, and welding voltage as joint optimization variables; S7: Generate thermal deformation feedforward compensation based on temperature field and thermal deformation prediction results; S8: Control the welding robot and welding power source to complete the welding operation based on the optimization results.

Images