An Adaptive Positioning Method and System for Robot Welding Based on Multi-Sensor Fusion

By using multi-sensor fusion technology, multi-source heterogeneous sensor data is acquired and processed, weights are dynamically allocated, and adaptive adjustments are made at different welding stages. This solves the problems of insufficient welding accuracy and real-time performance in existing technologies, and achieves more stable and accurate welding results.

CN123060201APending Publication Date: 2026-11-10GUANGXI INST OF IND TECHNICIANS (GUANGXI PETROCHEMICAL ADVANCED TECH SCHOOL) +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611245172.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-17
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing robot welding positioning technology based on multi-sensor fusion has shortcomings in dynamic weight allocation and processing of different welding stages, resulting in insufficient welding accuracy and real-time performance.

Method used

By acquiring multi-source heterogeneous sensor data streams, preprocessing them, calculating data validity and reliability indicators, dynamically allocating fusion weights, and adaptively adjusting them at different frequencies at different welding stages, and combining multi-sensor channel data for spatial fusion, adaptive positioning of the welding robot's end effector can be achieved.

Benefits of technology

It improves the accuracy of welding data acquisition and welding accuracy, avoids weight fluctuations caused by noise interference, and makes the fusion results more stable and reliable.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123060201A_ABST
    Figure CN123060201A_ABST
Patent Text Reader

Abstract

This invention discloses a robot welding adaptive positioning method and system based on multi-sensor fusion. It acquires multi-source heterogeneous sensor data streams through multiple sensor channels, performs time synchronization and spatial calibration on these data streams, and then compares and analyzes the data from each sensor channel after time synchronization and spatial calibration to determine data validity and reliability indicators. Based on the real-time data confidence scores of the sensor channels determined by the data validity and reliability indicators, a fusion weight is dynamically allocated to weighted fuse the data acquired from different sensor channels, obtaining an initial relative pose estimate of the welding robot's end effector relative to the target weld. Then, based on the initial relative pose estimate, the welding process is adaptively adjusted in multiple stages until welding is completed. This application improves the accuracy of data acquisition by fusing data from multiple sensor channels; furthermore, it improves the accuracy of welding through multi-stage adaptive adjustments to the welding process.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of robotic welding technology, and more specifically, to a robotic welding adaptive positioning method and system based on multi-sensor fusion. Background Technology

[0002] With the rapid development of manufacturing towards intelligence and automation, robotic welding technology has become an indispensable key technology in modern industrial production. Among these technologies, the accuracy and real-time performance of weld seam tracking are important indicators for evaluating welding quality. Although existing technologies for robotic welding positioning based on multi-sensor fusion exist, they still have shortcomings in areas such as optimizing dynamic weight allocation and handling different stages of the welding process. Summary of the Invention

[0003] To address at least one of the aforementioned technical problems, the present invention aims to provide a robot welding adaptive positioning method and system based on multi-sensor fusion, which can adaptively adjust to different stages of the welding process, thereby improving welding accuracy.

[0004] The first aspect of this invention provides a robot welding adaptive positioning method based on multi-sensor fusion, comprising: Step 1: Acquire multi-source heterogeneous sensor data streams, preprocess the multi-source heterogeneous sensor data streams to obtain time-synchronized and spatially calibrated data for each sensor channel; Step 2: For the data of each sensor channel after time synchronization and spatial calibration, calculate the data validity and data reliability index of each sensor channel at the current moment; Step 3: Based on the data validity and data reliability indicators, determine the comprehensive evaluation value of the corresponding sensing channel and dynamically allocate the fusion weights based on the real-time data confidence level; Step 4: Based on the real-time data confidence level, dynamically allocate fusion weights and spatially fuse the current time data of different sensor channels in the robot base coordinate system to obtain the initial relative pose estimate of the welding robot end effector relative to the target weld. Step 5: Based on the initial relative pose estimation value, adaptively adjust the welding positioning of the weld.

[0005] A second aspect of the present invention provides a robot welding adaptive positioning system based on multi-sensor fusion, comprising a memory and a processor. The memory stores a program for a robot welding adaptive positioning method based on multi-sensor fusion, and the processor, when executed, implements the robot welding adaptive positioning method based on multi-sensor fusion as described in any one of the above.

[0006] One or more technical solutions proposed in this application have at least the following technical effects: 1. Based on the current data validity index and the current data reliability index, and combined with the weight decay coefficient of the corresponding sensor in the most recent preset time period, a real-time data confidence dynamic allocation fusion weight is jointly constructed. Compared with the simple weight allocation based only on the quality parameters at the current moment in the prior art, this application fully considers the time dimension performance and historical reliability of the sensor, avoids the drastic fluctuation of weight caused by a single noise interference, and makes the fusion result more stable and reliable. 2. During the welding process, different sampling frequencies were designed from far to near. A lower frequency was used to save computing resources during the approaching stage, a higher frequency was used to ensure positioning accuracy during the neighborhood fine positioning stage, and a moderate frequency was used during the welding travel stage to balance real-time performance and stability. In summary, this invention improves the accuracy of data acquisition by fusing data from multiple sensor channels; furthermore, it improves the accuracy of welding by adaptively adjusting the welding process in multiple stages. Attached Figure Description

[0007] Figure 1 A flowchart of an adaptive positioning method for robot welding based on multi-sensor fusion according to the present invention is shown. Detailed Implementation

[0008] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0009] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0010] Figure 1 A flowchart of an adaptive positioning method for robot welding based on multi-sensor fusion according to the present invention is shown.

[0011] like Figure 1 As shown, this invention discloses an adaptive positioning method for robot welding based on multi-sensor fusion, comprising: Step 1: Acquire multi-source heterogeneous sensor data streams, preprocess the multi-source heterogeneous sensor data streams to obtain time-synchronized and spatially calibrated data for each sensor channel; Step 2: For the data of each sensor channel after time synchronization and spatial calibration, calculate the data validity and data reliability index of each sensor channel at the current moment; Step 3: Based on the data validity and data reliability indicators, determine the comprehensive evaluation value of the corresponding sensing channel and dynamically allocate the fusion weights based on the real-time data confidence level; Step 4: Based on the real-time data confidence level, dynamically allocate fusion weights and spatially fuse the current time data of different sensor channels in the robot base coordinate system to obtain the initial relative pose estimate of the welding robot end effector relative to the target weld. Step 5: Based on the initial relative pose estimation value, adaptively adjust the welding positioning of the weld.

[0012] According to an embodiment of the present invention, after determining the comprehensive evaluation value of the corresponding sensing channel, the sensing channels are prioritized and sorted. Sensing channels with a comprehensive evaluation value greater than or equal to a preset threshold are selected as valid sensing channels. The current time data of sensing channels with a comprehensive evaluation value less than the preset threshold are removed, and the corresponding sensing channels are determined to be invalid sensing channels.

[0013] According to an embodiment of the present invention, the step of obtaining data validity specifically includes: For laser vision sensors, the first effective component is determined based on the ratio of the continuous length of the effective laser stripes to the theoretical laser stripe length in the laser stripe image acquired at the current moment. The second effective component is determined based on the kurtosis coefficient of the gray-level histogram of the laser stripe image; The first validity component and the second validity component are weighted and summed to obtain the data validity index of the laser vision sensor; For the arc signal acquisition module, the seventh valid component is determined based on whether the waveforms of the welding current signal and welding voltage signal acquired at the current moment are within the preset normal waveform envelope range. The seventh validity component is used as the data validity index of the arc electrical signal acquisition module; The sensing channel includes a laser vision sensor and an arc electrical signal acquisition module.

[0014] It should be noted that the first effective component is the proportion of the continuous length of the laser stripe. If the laser stripe is blocked by splashes or the oxide scale on the weld surface causes uneven reflection, the stripe will break. The ratio of the effective length to the theoretical length reflects the image quality. The second effective component is the kurtosis coefficient of the grayscale histogram, which reflects the contrast distribution of the image. Overexposure or underexposure will cause the kurtosis to deviate from the normal range, indicating that the camera parameters need to be adjusted.

[0015] Furthermore, the sensing channel also includes a six-dimensional force / torque sensor and an ultrasonic ranging sensor. For the six-dimensional force / torque sensor, a third validity component is calculated based on whether each force / torque component is within a preset range at the current moment; a fourth validity component is calculated based on whether the absolute value of the rate of change of each force / torque component is lower than a preset rate of change threshold at the current moment; the third and fourth validity components are weighted and summed to obtain the data validity index of the six-dimensional force / torque sensor. For the ultrasonic ranging sensor array, a fifth validity component is calculated based on the ratio of the number of effective ultrasonic ranging sensors participating in the spherical intersection calculation to the total number of sensors at the current moment; a sixth validity component is calculated based on the ratio of the standard deviation to the mean of the distance values ​​measured by each effective ultrasonic ranging sensor; the fifth and sixth validity components are weighted and summed to obtain the data validity index of the ultrasonic ranging sensor array.

[0016] It should be noted that the central controller reads the six component values ​​output by the six-dimensional force / torque sensor in real time. For each component, a unique range is set at the factory or during the calibration stage. When all components fall within the preset measurement range, the corresponding third validity component is set to 1; otherwise, it is set to zero. If the absolute value of the rate of change of all six components is lower than their respective thresholds, the fourth validity component is set to 1. If the absolute value of the rate of change of any component exceeds the threshold, the change is deemed unreasonable, and the corresponding fourth validity component is set to zero. The ultrasonic ranging sensor array consists of at least three ultrasonic sensors fixed around the workstation. The fifth validity component is the ratio of the number of effective ultrasonic ranging sensors participating in the spherical intersection calculation at the current moment to the total number of sensors. The higher the ratio, the more sensors can be used for spherical intersection positioning, and the more sufficient the geometric constraints of the positioning result. The sixth validity component evaluates the dispersion between the distance values ​​measured by all effective ultrasonic sensors, quantified by the ratio of the standard deviation to the mean.

[0017] It should be noted that before welding begins or during the process parameter setting stage, the central controller retrieves the corresponding standard current-voltage waveform envelope template from the process database based on the selected welding process type. This template defines the upper and lower bounds of the allowable current and voltage envelopes for each phase of the welding cycle (usually centered on the theoretical waveform, with a certain tolerance percentage extended above and below). If the real-time acquired current and voltage values ​​are both within their respective envelopes, the seventh effective component takes the higher value; if any point exceeds the envelope range (e.g., a sudden current spike or a voltage drop), the arc signal is determined to be abnormal at that moment, and the seventh component decreases. Alternatively, the seventh effective component can be quantized through binarization. For example, if the waveforms of the welding current and welding voltage signals acquired at the current moment are both within the preset normal waveform envelope range, the seventh effective component is set to 1; otherwise, it is set to zero.

[0018] According to an embodiment of the present invention, the step of obtaining the data reliability index specifically includes: The central controller is set to maintain a historical data queue of length N for each sensing channel. The historical data queue stores the measurement data of the sensing channel in the most recent N sampling times. For the measurement data at the current moment, the preset central controller determines the predicted value at the current moment and the uncertainty range of the predicted value based on the historical data queue of the sensing channel and the autoregressive prediction model. The absolute deviation is obtained by calculating the difference between the absolute deviation of the current measurement data and the predicted value. By comparing and analyzing the absolute deviation with the half-width of the uncertainty interval of the predicted value, the first reliability component of the sensing channel at the current moment is obtained. The preset central controller calculates the moving average and moving standard deviation of the measurement data of the sensor channel over the most recent N sampling times based on the historical data queue of the sensor channel, and obtains the deviation between the measurement data at the current time and the moving average. The deviation is compared with a multiple of the sliding standard deviation to obtain the second reliability component of the sensing channel at the current moment; The first reliability component and the second reliability component are weighted and summed to obtain the data reliability index of the sensing channel at the current moment.

[0019] It should be noted that the central controller maintains the following historical data queue for each sensor channel: This stores the measurement data of the sensor channel over the most recent N sampling times, where N is a positive integer greater than or equal to 10. For the measurement data at the current time t, the central controller uses a p-order autoregressive model to calculate the predicted value for the current time, the mathematical expression of which is: ,in This is the model's prediction of the measured data at the current time t. denoted as the constant mean term of the historical data for this sensor channel, and p is the autoregressive order of the model, indicating how many past moments of historical data are used to predict the current value. Let i represent the i-th order autoregressive coefficient, where i belongs to p. This represents the actual historical measurement data of the sensing channel at the i-th time in the past. This represents the random error term at the current moment.

[0020] According to an embodiment of the present invention, it further includes: When the data reliability index of a certain sensing channel is lower than the preset reliability threshold for M consecutive sampling times, the preset central controller sends a self-test command to the acquisition front end of the sensing channel, triggering the sensing channel to execute the built-in self-test program; where M is a positive integer greater than or equal to 3.

[0021] According to an embodiment of the present invention, the step of obtaining the real-time data confidence level specifically includes: The product of the data validity index and the data reliability index of each sensing channel at the current moment is used as the basic confidence level of that sensing channel at the current moment. The weighted moving average of the basic confidence level of the current sensing channel over the most recent L sampling times is used as the historical confidence level trend value of the sensing channel; The time interval between the moment when the sensing channel was most recently determined to be an invalid sensing channel and the moment when it is currently determined is taken as the continuous effective duration of the sensing channel. The confidence recovery coefficient of the sensing channel is calculated based on the continuous effective duration. The real-time data confidence level of the sensing channel at the current moment is obtained by multiplying the current baseline confidence level, the historical confidence trend value, and the confidence recovery coefficient.

[0022] It should be noted that sampling times closer to the current time are assigned a higher weighting coefficient, and the longer the continuous effective duration, the higher the confidence recovery coefficient and the closer it is to 1. The confidence recovery coefficient of the sensing channel can be obtained by setting a duration interval and normalizing the continuous effective duration based on the set duration interval.

[0023] Furthermore, the dynamic fusion weight of each sensor channel is the ratio of the real-time data confidence of that sensor channel to the sum of the real-time data confidence of all sensor channels. When the real-time data confidence of a certain sensor channel is lower than the preset lower confidence threshold, the dynamic fusion weight of that sensor channel is forcibly set to zero, and the current data of that sensor channel does not participate in the weighted fusion calculation.

[0024] According to an embodiment of the present invention, the step of adaptively adjusting the welding positioning of the weld seam based on the initial relative pose estimation value specifically includes: During the process of the robot end effector moving towards the starting welding point, the update data of the initial relative pose estimation value output in step four is acquired in real time at the first correction frequency, and the motion trajectory of the end effector is dynamically corrected for the first time based on the updated data. Once the end effector reaches the preset neighborhood of the solder joint location, it switches to the second correction frequency to obtain the updated data of the initial relative pose estimation value output in step four in real time, and performs a second dynamic correction on the pose of the end effector based on the updated data. The second correction frequency is higher than the first correction frequency. After the welding robot begins to move along the weld seam, it switches to the third correction frequency to obtain the updated data of the initial relative pose estimation value output in step four in real time, and performs a third dynamic correction on the welding trajectory of the end effector based on the updated data. The third correction frequency is lower than the second correction frequency but higher than the first correction frequency until the welding task is completed.

[0025] According to an embodiment of the present invention, the first dynamic correction step specifically includes: The initial relative pose estimate at the current moment is compared with the initial relative pose estimate at the previous moment, and the rate of change vector of the initial relative pose estimate in three-dimensional space is calculated. Based on the change rate vector, determine whether the current motion trend of the end effector is consistent with the preset approach path direction. If it is consistent, maintain the current motion control parameters unchanged. If it is inconsistent, calculate the deviation angle between the current actual motion direction of the end effector and the preset approach path direction. Based on the deviation angle, a lateral correction amount and a longitudinal correction amount are generated. The lateral correction amount and the longitudinal correction amount are superimposed on the current motion control command of the end effector, so that the actual motion direction of the preset end effector gradually approaches the preset approach path direction.

[0026] It should be noted that the central controller acquires the fused pose estimate at the first correction frequency and calculates its rate of change vector, i.e., the actual movement direction and velocity of the end effector. Simultaneously, it presets an ideal approach path from the current position to the welding start point (the preset approach path). The actual movement direction is compared with the movement direction of the ideal approach path (the preset approach path direction) to obtain the deviation angle. The purpose of the lateral correction is to correct the lateral deviation of the end effector in the direction perpendicular to the path, "pulling" it back onto the preset path. The magnitude of the lateral correction is proportional to the weighted sum of the current lateral deviation velocity and the sine of the deviation angle. The purpose of the longitudinal correction is not only to adjust the forward speed along the path direction, but more importantly, to compensate for the "path length loss" caused by lateral correction, ensuring that the end effector does not excessively lag or advance beyond the expected longitudinal position while performing lateral correction, ultimately achieving synchronous matching of position and velocity within the weld start point neighborhood.

[0027] Furthermore, by setting a first deviation angle threshold, adjustments caused by small directional deviations can be reduced. For example, if the deviation angle is less than the preset first deviation angle threshold, the actual movement direction of the current end effector is not adjusted; if the deviation angle is greater than or equal to the preset first deviation angle threshold, a lateral correction and a longitudinal correction are generated and superimposed on the current motion control command of the end effector.

[0028] According to an embodiment of the present invention, the second dynamic correction step specifically includes: Taking the starting point of the solder joint as the target point, the central controller calculates the position deviation vector between the current position of the end effector and the starting point of the solder joint, as well as the attitude deviation quaternion between the current attitude of the end effector and the target attitude, based on the initial relative pose estimation value at the current moment. A position adjustment command is generated based on the position deviation vector, and an attitude adjustment command is generated based on the attitude deviation quaternion. The position adjustment command and the attitude adjustment command are then superimposed on the motion control command of the end effector. After each superposition is completed, the central controller re-acquires the updated data of the relative pose estimation value output in step four, and determines whether the magnitude of the updated position deviation vector is less than the preset position accuracy threshold and whether the norm of the updated attitude deviation quaternion is less than the preset attitude accuracy threshold. If both conditions are met, the superposition is terminated and the current pose is determined as the accurate pose of the starting point. If not, the next superposition is performed until the termination condition is met or the number of superpositions reaches the preset maximum number of superpositions.

[0029] It should be noted that, for example, if the radius of the preset neighborhood range is R=10mm, when the end effector has entered the neighborhood of the starting weld point, the current position of the end effector is (1000.5, 500.3, 200.1)mm, the target starting weld point is (1000.0, 500.0, 200.0)mm, the position deviation vector is (0.5, 0.3, 0.1)mm, the module length is 0.59mm, the attitude deviation quaternion is (0.01, 0.02, -0.01, 0.9998), and the norm is approximately 0.024; for example, the position accuracy threshold is set to 0.1mm, and the attitude accuracy threshold is set to 0.01. After the first iteration, the positional deviation decreased to (0.2, 0.1, 0.05) and the mold length was 0.23 mm, which still did not meet the standard. After the second iteration, it decreased to (0.08, 0.04, 0.02) and the mold length was 0.09 mm. At the same time, the attitude deviation norm decreased to 0.008, which met the threshold. The iteration was terminated, and the current pose was confirmed as the accurate pose of the starting point for welding. Welding can then begin.

[0030] According to an embodiment of the present invention, the third dynamic correction step specifically includes: The central controller calculates in real time the lateral and height deviations of the current position of the preset end effector relative to the center line of the preset welding trajectory based on the updated data of the initial relative pose estimation value. The lateral deviation is input to the first PID controller, which outputs the lateral correction amount; the height deviation is input to the second PID controller, which outputs the height correction amount. The lateral and height correction values ​​are superimposed on the weld seam tracking control command of the welding robot, driving the end effector to correct the lateral and height deviations in real time during the welding process until the welding is completed.

[0031] It should be noted that during the welding process, the end needs to move along the weld seam trajectory while continuously correcting the lateral and height deviations. The lateral and height deviations are input into two independent PID controllers, and the output correction values ​​are superimposed on the weld seam tracking command to ensure the quality of weld formation.

[0032] Furthermore, the central controller calculates the actual welding heat input at the current moment based on the real-time welding current signal and real-time welding voltage signal acquired by the arc electrical signal acquisition module. The actual welding heat input is compared with the preset target welding heat input. A welding heat input compensation is generated based on the comparison result. The welding heat input compensation is then added to the output control parameters of the welding power supply until welding is completed.

[0033] It should be noted that the welding heat input (current × voltage) is also calculated in real time using the electric arc signal, compared with the preset target value, and the heat input compensation is generated to adjust the welding power output in order to improve the weld formation quality.

[0034] Furthermore, during the welding process of the welding robot moving along the weld seam, when the data validity index or data reliability index of the arc electrical signal acquisition module shows a downward trend but both are still higher than the preset effective threshold, the central controller automatically increases the dynamic fusion weight of the laser vision sensor and the six-dimensional force / torque sensor, and correspondingly decreases the dynamic fusion weight of the arc electrical signal acquisition module; when it is determined that either of the two indicators is lower than the preset effective threshold, the fusion weight of the arc electrical signal acquisition module is forcibly set to zero, and the fusion weight of the remaining effective sensing channels is renormalized to maintain the accuracy and stability of welding positioning until the welding task is completed.

[0035] A second aspect of the present invention provides a robot welding adaptive positioning system based on multi-sensor fusion, comprising a memory and a processor. The memory stores a program for a robot welding adaptive positioning method based on multi-sensor fusion, and the processor, when executed, implements the robot welding adaptive positioning method based on multi-sensor fusion as described in any one of the above.

[0036] This invention discloses a robot welding adaptive positioning method and system based on multi-sensor fusion. By fusing data from multiple sensor channels, the accuracy of data acquisition is improved. In addition, the accuracy of welding is improved by adaptively adjusting the welding process in multiple stages.

[0037] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A robot welding adaptive positioning method based on multi-sensor fusion, characterized in that, include: Step 1: Acquire multi-source heterogeneous sensor data streams, preprocess the multi-source heterogeneous sensor data streams to obtain time-synchronized and spatially calibrated data for each sensor channel; Step 2: For the data of each sensor channel after time synchronization and spatial calibration, calculate the data validity and data reliability index of each sensor channel at the current moment; Step 3: Based on the data validity and data reliability indicators, determine the comprehensive evaluation value of the corresponding sensing channel and dynamically allocate the fusion weights based on the real-time data confidence level; Step 4: Based on the real-time data confidence level, dynamically allocate fusion weights and spatially fuse the current time data of different sensor channels in the robot base coordinate system to obtain the initial relative pose estimate of the welding robot end effector relative to the target weld. Step 5: Based on the initial relative pose estimation value, adaptively adjust the welding positioning of the weld.

2. The adaptive positioning method for robot welding based on multi-sensor fusion according to claim 1, characterized in that, The steps for obtaining data validity specifically include: For laser vision sensors, the first effective component is determined based on the ratio of the continuous length of the effective laser stripes to the theoretical laser stripe length in the laser stripe image acquired at the current moment. The second effective component is determined based on the kurtosis coefficient of the gray-level histogram of the laser stripe image; The first validity component and the second validity component are weighted and summed to obtain the data validity index of the laser vision sensor; For the arc signal acquisition module, the seventh valid component is determined based on whether the waveforms of the welding current signal and welding voltage signal acquired at the current moment are within the preset normal waveform envelope range. The seventh validity component is used as the data validity index of the arc electrical signal acquisition module; The sensing channel includes a laser vision sensor and an arc electrical signal acquisition module.

3. The adaptive positioning method for robot welding based on multi-sensor fusion according to claim 1, characterized in that, The steps for obtaining the data reliability indicators specifically include: The central controller is set to maintain a historical data queue of length N for each sensing channel. The historical data queue stores the measurement data of the sensing channel in the most recent N sampling times. For the measurement data at the current moment, the preset central controller determines the predicted value at the current moment and the uncertainty range of the predicted value based on the historical data queue of the sensing channel and the autoregressive prediction model. The absolute deviation is obtained by calculating the difference between the absolute deviation of the current measurement data and the predicted value. By comparing and analyzing the absolute deviation with the half-width of the uncertainty interval of the predicted value, the first reliability component of the sensing channel at the current moment is obtained. The preset central controller calculates the moving average and moving standard deviation of the measurement data of the sensor channel over the most recent N sampling times based on the historical data queue of the sensor channel, and obtains the deviation between the measurement data at the current time and the moving average. The deviation is compared with a multiple of the sliding standard deviation to obtain the second reliability component of the sensing channel at the current moment; The first reliability component and the second reliability component are weighted and summed to obtain the data reliability index of the sensing channel at the current moment.

4. The adaptive positioning method for robot welding based on multi-sensor fusion according to claim 3, characterized in that, Also includes: When the data reliability index of a certain sensing channel is lower than the preset reliability threshold for M consecutive sampling times, the preset central controller sends a self-test command to the acquisition front end of the sensing channel, triggering the sensing channel to execute the built-in self-test program; where M is a positive integer greater than or equal to 3.

5. The adaptive positioning method for robot welding based on multi-sensor fusion according to claim 1, characterized in that, The steps for obtaining the real-time data confidence level specifically include: The product of the data validity index and the data reliability index of each sensing channel at the current moment is used as the basic confidence level of that sensing channel at the current moment; The weighted moving average of the basic confidence level of the current sensing channel over the most recent L sampling times is used as the historical confidence level trend value of the sensing channel; The time interval between the moment when the sensing channel was most recently determined to be an invalid sensing channel and the moment when it is currently determined is taken as the continuous effective duration of the sensing channel. The confidence recovery coefficient of the sensing channel is calculated based on the continuous effective duration. The real-time data confidence level of the sensing channel at the current moment is obtained by multiplying the current baseline confidence level, the historical confidence trend value, and the confidence recovery coefficient.

6. The adaptive positioning method for robot welding based on multi-sensor fusion according to claim 1, characterized in that, The step of adaptively adjusting the welding positioning of the weld seam based on the initial relative pose estimation value specifically includes: During the process of the robot end effector moving towards the starting welding point, the update data of the initial relative pose estimation value output in step four is acquired in real time at the first correction frequency, and the motion trajectory of the end effector is dynamically corrected for the first time based on the updated data. Once the end effector reaches the preset neighborhood of the solder joint location, it switches to the second correction frequency to obtain the updated data of the initial relative pose estimation value output in step four in real time, and performs a second dynamic correction on the pose of the end effector based on the updated data. The second correction frequency is higher than the first correction frequency. After the welding robot begins to move along the weld seam, it switches to the third correction frequency to obtain the updated data of the initial relative pose estimation value output in step four in real time, and performs a third dynamic correction on the welding trajectory of the end effector based on the updated data. The third correction frequency is lower than the second correction frequency but higher than the first correction frequency until the welding task is completed.

7. The adaptive positioning method for robot welding based on multi-sensor fusion according to claim 6, characterized in that, The steps of the first dynamic correction specifically include: The initial relative pose estimate at the current moment is compared with the initial relative pose estimate at the previous moment, and the rate of change vector of the initial relative pose estimate in three-dimensional space is calculated. Based on the change rate vector, determine whether the current motion trend of the end effector is consistent with the preset approach path direction. If it is consistent, maintain the current motion control parameters unchanged. If it is inconsistent, calculate the deviation angle between the current actual motion direction of the end effector and the preset approach path direction. Based on the deviation angle, a lateral correction amount and a longitudinal correction amount are generated. The lateral correction amount and the longitudinal correction amount are superimposed on the current motion control command of the end effector, so that the actual motion direction of the preset end effector gradually approaches the preset approach path direction.

8. The adaptive positioning method for robot welding based on multi-sensor fusion according to claim 6, characterized in that, The second dynamic correction step specifically includes: Taking the starting point of the solder joint as the target point, the central controller calculates the position deviation vector between the current position of the end effector and the starting point of the solder joint, as well as the attitude deviation quaternion between the current attitude of the end effector and the target attitude, based on the initial relative pose estimation value at the current moment. A position adjustment command is generated based on the position deviation vector, and an attitude adjustment command is generated based on the attitude deviation quaternion. The position adjustment command and the attitude adjustment command are then superimposed on the motion control command of the end effector. After each superposition is completed, the central controller re-acquires the updated data of the relative pose estimation value output in step four, and determines whether the magnitude of the updated position deviation vector is less than the preset position accuracy threshold and whether the norm of the updated attitude deviation quaternion is less than the preset attitude accuracy threshold. If both conditions are met, the superposition is terminated and the current pose is determined as the accurate pose of the starting point. If not, the next superposition is performed until the termination condition is met or the number of superpositions reaches the preset maximum number of superpositions.

9. The adaptive positioning method for robot welding based on multi-sensor fusion according to claim 6, characterized in that, The third dynamic correction step specifically includes: The central controller calculates in real time the lateral and height deviations of the current position of the preset end effector relative to the center line of the preset welding trajectory based on the updated data of the initial relative pose estimation value. The lateral deviation is input to the first PID controller, which outputs the lateral correction amount; the height deviation is input to the second PID controller, which outputs the height correction amount. The lateral and height correction values ​​are superimposed on the weld seam tracking control command of the welding robot, driving the end effector to correct the lateral and height deviations in real time during the welding process until the welding is completed.

10. A robot welding adaptive positioning system based on multi-sensor fusion, characterized in that, The device includes a memory and a processor. The memory stores a program for an adaptive positioning method for robot welding based on multi-sensor fusion. When the processor executes the program, it implements the adaptive positioning method for robot welding based on multi-sensor fusion as described in any one of claims 1 to 9.