An automobile island type welding numerical-physical consistency traceability and hierarchical compensation method and system

Abstract

The application discloses an island type welding digital-physical consistency traceability and hierarchical compensation method and system, belonging to the cross field of automobile intelligent manufacturing and digital twin technology, comprising: synchronously collecting automobile welding island physical entity and digital twin body data and performing space-time alignment to construct a four-dimensional comprehensive consistency index; decomposing total deviation traceability into four components of sensor deviation, model structure deviation, parameter drift deviation and time delay deviation, and identifying the contribution degree of each component based on energy proportion; calling a differentiated compensation sub-strategy for different components and adaptively adjusting the fusion weight; feeding back the compensated twin output to the first step for re-evaluation until the convergence criterion is met and a compensation traceability report is output. The application realizes consistency maintenance between the automobile island type welding digital twin and the physical entity, can significantly improve the beat stability and the production line adjustability, and is suitable for intelligent manufacturing and equipment operation and maintenance scenes of a body-in-white welding workshop.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of intelligent automotive manufacturing and digital twin technology, specifically involving a method and system for traceability and hierarchical compensation of consistency between real and digital data in automotive island welding assembly. Background Technology

[0002] Body-in-white welding is a core process in automobile manufacturing. For a long time, it has primarily employed a sequential assembly line welding model—breaking down hundreds to thousands of welding points into dozens of workstations, with workpieces flowing sequentially and completed station by station. While this model offers stable cycle times for mass production of a single model, as the automotive industry evolves towards "multi-model mixed-line production, small-batch customization, and rapid model changeover," the traditional model has gradually revealed significant problems such as rigid workstations, long production lines, high changeover costs, and insufficient flexibility. It struggles to meet the demands of rapid iteration and platform-co-production for new energy vehicles.

[0003] To overcome the aforementioned bottlenecks, SAIC-GM-Wuling Automobile Co., Ltd. proposed a new manufacturing organization paradigm: island manufacturing. Its core idea is to break away from the traditional linear process division and compactly integrate multiple pieces of equipment, clamping and logistics units around a single workstation. This allows processes that were originally scattered across multiple sequential workstations to be completed in parallel within a single "manufacturing island." Applying this paradigm to the body-in-white welding process results in automotive island welding: multiple welding robots, servo welding guns, reconfigurable positioning fixtures, and loading / unloading mechanisms are integrated into a single workstation, working collaboratively around the same body-in-white to achieve process compactness and vehicle flexibility.

[0004] However, while island-type welding offers flexibility, it also involves multiple robots working collaboratively and intensively in a compact space, with strong process coupling and heterogeneous signal sources. Traditional manual debugging and offline simulation are no longer sufficient to guarantee operational stability. Therefore, constructing a high-fidelity digital twin for the automotive welding island, and using this twin to predict weld quality, perform multi-robot collaborative virtual debugging, and optimize cycle time, has become a key technological path to support the engineering implementation of island-type welding. However, in automotive welding engineering practice, due to the following factors, multi-source deviations inevitably arise between the digital twin and the physical welding island:

[0005] First, the displacement sensors, force sensors, current transformers, and vision measurement devices deployed on the welding island have zero drift, temperature drift, and noise, which leads to measurement errors in the robot pose, welding clamp pressure, and vehicle body measurement points.

[0006] Second, the welding thermo-mechanical coupling mechanism model, robot dynamics model, and fixture stiffness model on which the digital twin is based contain simplifications, assumptions, and approximations, and there is a gap between the model structure and the real welding physical process.

[0007] Third, the electrodes wear down continuously during long-term welding, the fixture positioning pins and guide blocks become slightly loose, and the robot reducer and gear pair age and wear down, causing the key parameters of the physical entity to drift slowly. If the twin parameters are not updated in time, they will gradually become mismatched.

[0008] Fourth, data transmission, network communication, and computation delays exist between the robot controller, programmable logic controller (PLC), vision inspection system, and manufacturing execution system (MES), causing a misalignment of the states on both sides in the time dimension.

[0009] If the aforementioned deviations are not compensated for in a timely manner, they will gradually accumulate and amplify, eventually leading to "distortion" of the twin, rendering key applications such as solder joint location prediction, electrode life estimation, and cycle time simulation completely ineffective. Existing technologies mainly include the following methods:

[0010] (1) The state estimation method based on Kalman filtering and its extended form corrects the twin state by weighted fusion of physical observations and twin predictions. However, it lacks detailed distinction of the physical source of the deviation. The compensation process is a "black box" adjustment with poor interpretability. Moreover, the compensation parameters are statically fixed and difficult to adapt to vehicle model switching and multi-condition changes. (2) The method based on manual threshold alarm and offline parameter calibration requires engineers to manually correct the model when the deviation exceeds the preset threshold. The response is lagging, relies on experience, and cannot close the loop in real time, making it difficult to meet the requirements of modern high-speed welding. The real-time consistency requirement of the production line; (3) The end-to-end deviation prediction method based on deep learning trains the neural network to directly learn the deviation pattern and make compensation. It has high accuracy under static working conditions, but lacks interpretability, is sensitive to the distribution of training data, and has weak generalization ability when switching models or new workpieces are put on the line; (4) The deviation quantification method based on a single consistency index evaluates the degree of deviation and triggers compensation, such as geometric distance or timing error. However, a single index cannot characterize the consistency differences of the welding island in multiple dimensions such as geometric pose, physical process, behavioral rhythm, and timing synchronization.

[0011] In summary, the existing technologies have the following common technical problems:

[0012] (1) The consistency measurement dimension is singular, making it difficult to fully depict the consistency between the digital twin and the physical entity of the automotive welding island in multiple attribute dimensions such as geometry, physics, behavior, and time sequence;

[0013] (2) The deviation compensation adopts a uniform processing mode, without subdividing and tracing the physical source of the deviation for diagnosis. The compensation process lacks interpretability and is not targeted enough for different types of deviations.

[0014] (3) The compensation strategies mostly use static parameters, which cannot adapt to the changes in deviation characteristics caused by the drift of working conditions such as vehicle model switching, electrode wear, and shift differences, and there is a risk of "overcompensation" or "undercompensation".

[0015] (4) Lacking a closed-loop convergence mechanism and a cross-workpiece knowledge accumulation mechanism, the compensation process is considered complete after a single compensation, making it difficult to guarantee the stability, convergence and optimality of the compensation process. Summary of the Invention

[0016] To address the aforementioned technical issues, this invention provides a method and system for tracing and hierarchical compensation of consistency between digital twins and physical entities in automotive island welding systems. Employing a multi-dimensional approach, it utilizes deviation tracing, hierarchical compensation, and closed-loop convergence to achieve comprehensive perception, interpretable tracing, precise hierarchical compensation, and stable convergence of deviations between the digital twin and physical entities of the automotive welding island. This enhances the consistency maintenance capabilities and engineering usability of the digital twin welding system.

[0017] To achieve the above objectives, the present invention adopts the following technical solution:

[0018] A method for tracing and hierarchical compensation of the consistency of welding data in automotive island-type structures includes the following steps:

[0019] S1. Multi-dimensional measurement: Synchronously collect measured state data of the physical entity of the automotive island welding assembly. Simulation output data of the corresponding digital twin The measured state data includes at least the welding robot tool center point pose, welding clamp electrode pressure and current, key measurement point positions of the body-in-white, fixture positioning pin status, and process cycle time stamps. The measured state data and simulation output data are synchronized with time stamps and registered in a spatial coordinate system to obtain spatiotemporally aligned data pairs. , Simulation output data after timestamp synchronization and spatial coordinate system registration; based on this spatiotemporally aligned data pair, consistency sub-indices are constructed in the dimensions of geometric pose, physical process, behavioral beat, and temporal synchronization. , , and A comprehensive consistency index is obtained through scenario-adaptive weighted fusion. ;

[0020] S2, Deviation Source Decomposition: Calculate the total deviation of the spatiotemporally aligned data pairs described in S1. The total deviation is decomposed into sensor deviation components. Model structure deviation components Parameter drift deviation components and time delay deviation components There are four types of components, and the contribution of each deviation component is identified based on its energy proportion. This leads to explainable results for tracing the sources of bias.

[0021] S3. Adaptive hierarchical compensation: Based on the deviation components and their contribution rates identified in step S2. The system invokes a compensation sub-strategy that matches the deviation component to implement tiered compensation, and adaptively adjusts the fusion weight of each sub-strategy based on the welding condition characteristics and the change in the overall consistency index. The compensated digital twin output ;

[0022] S4. Closed-loop iterative convergence: Output the compensated digital twin. Feedback is sent to step S1 to recalculate the overall consistency index. ;like If the preset convergence criterion is met, the compensation process ends and the final twin state and compensation source tracing report are output; otherwise, the process ends. The new simulation output is returned to step S2 for the next iteration, until the convergence criterion is met or the maximum number of iterations is reached. .

[0023] A system for traceability and hierarchical compensation of numerical consistency in automotive island-type welding assembly, the system being used to implement the method described, and the system comprising five core modules:

[0024] The data acquisition and spatiotemporal alignment module is used to acquire measured data of the physical entity of the automotive welding island and simulation output data of the digital twin, and to perform timestamp synchronization and spatial coordinate system registration to obtain spatiotemporal aligned data pairs.

[0025] The multi-dimensional quantity module is connected to the data acquisition and spatiotemporal alignment module. It is used to construct consistency sub-indicators based on the spatiotemporal alignment data pair in the dimensions of geometric pose, physical process, behavioral beat and temporal synchronization, and obtain a comprehensive consistency index through scene adaptive weighted fusion.

[0026] The deviation source decomposition module, connected to the multi-dimensional quantity module, is used to calculate the total deviation of the spatiotemporal aligned data pair, decompose the total deviation source into four components: sensor deviation, model structure deviation, parameter drift deviation, and time delay deviation, identify the contribution of each component, and output interpretable source decomposition results.

[0027] An adaptive hierarchical compensation module, connected to the deviation source decomposition module, is used to call the corresponding compensation sub-strategy for each deviation component and adaptively adjust the fusion weight, and output the compensated digital twin state.

[0028] The closed-loop iterative convergence module is connected to both the adaptive hierarchical compensation module and the multi-dimensional quantity module. It is used to evaluate the compensation effect and trigger the iterative process until the preset convergence criterion is met, and to generate a compensation source tracing report that includes the contribution of the deviation component, the fusion weight of each compensation sub-strategy, and the trajectory of the consistency index change.

[0029] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0030] First, multi-dimensional comprehensive perception: construct a four-dimensional consistency measurement system of geometric pose, physical process, behavioral rhythm, and temporal synchronization to comprehensively and objectively reflect the consistency status between the welding island twin and the physical entity;

[0031] Second, explainable deviation diagnosis: each type of deviation has a clear physical source explanation, each compensation has a traceable basis, and the output compensation traceability report significantly improves the debuggability, auditability and quality traceability of the digital twin welding system;

[0032] Third, adaptive precision compensation: Differentiated compensation sub-strategies are adopted for deviations from different sources. Through adaptive weight dynamic adjustment, static compensation is avoided from failing under vehicle model switching and electrode wear conditions, which greatly improves robustness and accuracy.

[0033] Fourth, closed-loop convergence and knowledge accumulation: introduce closed-loop iteration, multiple convergence criteria and historical experience library to ensure that the compensation process is stable and controllable, avoid overcompensation, undercompensation or oscillation, and support knowledge reuse across workpieces and shifts to accelerate the convergence process of new model launch. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the overall process of the multi-dimensional quantity and traceability hierarchical compensation method for automotive island welding digital twins described in this invention. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0036] like Figure 1 As shown, the present invention provides a method for traceability and hierarchical compensation of the consistency of welding data in automotive island structures, comprising the following steps:

[0037] S1. Multi-dimensional measurement: Synchronously collect measured state data of the physical entity of the automotive island welding assembly. Simulation output data of the corresponding digital twin The measured state data includes at least the welding robot's TCP pose, welding clamp electrode pressure and current, key measurement point positions on the body-in-white, fixture positioning pin status, and process cycle timestamps. The measured state data and simulation output data are synchronized with their timestamps and registered in a spatial coordinate system to obtain spatiotemporally aligned data pairs. , Simulation output data after timestamp synchronization and spatial coordinate system registration; based on this spatiotemporally aligned data pair, consistency sub-indices are constructed in the dimensions of geometric pose, physical process, behavioral beat, and temporal synchronization. , , and A comprehensive consistency index is obtained through scenario-adaptive weighted fusion. ;

[0038] Furthermore, timestamp synchronization is achieved using the Precision Time Protocol (PTP), with an alignment accuracy better than 1ms; the spatial coordinate system registration is achieved through the following coordinate transformation relationship:

[0039] (1)

[0040] in, Let be a rotation matrix. As translation vectors, both are calibration feature points deployed within the welding island at the fixture reference surface, robot flange, and body-in-white reference point system (RPS) points, and are solved using the least squares method.

[0041] Furthermore, the fusion formula for the overall consistency index is:

[0042] (2)

[0043] in, The geometric pose consistency sub-index is calculated based on the normalized reciprocal of the Euclidean distance between the robot tool centerpoint (TCP) pose and the vehicle body measurement point. It is a sub-index of physical process consistency, calculated based on the reciprocal of the normalized root mean square error of physical quantities such as welding current, electrode pressure, electrode temperature, and weld nugget stress; The behavior rhythm consistency sub-index is calculated based on the dynamic time warping (DTW) distance between the welding action sequence and the upper and lower part action sequences; The temporal synchronization consistency sub-index is calculated based on the cross-correlation coefficient of the state time series; weights The overall consistency index is dynamically allocated according to the welding conditions and process standards, so that the consistency dimension that is sensitive to the scenario has a higher responsiveness. j is the weight index.

[0044] Furthermore, the geometric pose consistency sub-index This is used to measure the spatial consistency of geometric quantities such as the TCP pose of a welding robot, RPS measurement points on the body-in-white, and weld point positions. Let the physically measured position of the k-th geometric feature point within the sampling window N be denoted as . The twin output position is ,common If there are 1 feature point, then:

[0045] (3)

[0046] in, It is the Euclidean norm. To normalize the reference distance, the upper limit of the weld joint tolerance is taken; when all feature points completely coincide... The greater the deviation, The closer it gets to 0. For attitude components, the Euclidean distance can be extended to pose error:

[0047] (4)

[0048] in, , These are the physical and twin attitude rotation matrices, respectively. Let Frobenius norm be the rotation error. For attitude-position weighting coefficients.

[0049] Furthermore, the physical process consistency sub-index Used to measure the consistency of physical quantities such as welding current, electrode pressure, electrode temperature, and weld nugget stress. There are M physical quantity channels, and the sequence of physical quantities for the m-th physical quantity channel within window N is as follows: Twin output is Then the normalized root mean square error of each channel is:

[0050] (5)

[0051] The physical consistency sub-index is defined as the reciprocal of the weighted average of the NRMSE of each channel:

[0052] (6)

[0053] in, Assign importance weights to each physical channel. It is the maximum value of the m-th physical quantity channel within the sampling window. It is the minimum value of the m-th physical quantity channel within the sampling window.

[0054] Furthermore, the behavioral beat consistency sub-index Used to measure the consistency of welding action sequences and loading / unloading action sequences in terms of cycle time. Let the physical side action sequence be... I represents the number of physical action sequences, and the twin-side action sequences are... J represents the number of action sequences on the twin side. Each entry includes the action type and duration. (Based on DTW distance) The difference between two sequences is measured and calculated as shown in equation (7):

[0055] (7)

[0056] Where i and j are the indices of the physical-side action sequence and the twin-side action sequence, respectively, π is the alignment path that satisfies the boundary, continuity, and monotonicity constraints, and the local cost function is... Take the weighted sum of the penalty for inconsistent action types and the difference in beat duration:

[0057] (8)

[0058] in , The duration of the action. For nominal beat, As a type-duration tradeoff factor, Indicates the type of the i-th action on the physical side. This indicates the type of the j-th action on the twin side. The final behavioral consistency sub-index is:

[0059] (9)

[0060] in, To ensure the reference normalized distance .

[0061] Furthermore, the timing synchronization consistency sub-indicator Measure the synchronicity of physical and twin state time series on the time axis. For critical state signals and Define the zero-lag normalized cross-correlation coefficient:

[0062] (10)

[0063] in, This represents the mean of the physical sequence within window N. This represents the mean of the twin sequence within window N.

[0064] Further considering possible time misalignment, we define the maximum cross-correlation and its delay: , , Represents the physical side state sequence With twin-side state sequence The relative time offset between them This represents the manually set maximum latency search boundary. The timing synchronization consistency sub-index comprehensively considers correlation strength and time misalignment penalties:

[0065] (11)

[0066] in, The allowable time misalignment tolerance. When and When fully synchronized and highly correlated When there is time delay or waveform distortion Significant decrease.

[0067] S2, Deviation Source Decomposition: Calculate the total deviation of the spatiotemporally aligned data pairs described in S1. The total deviation is decomposed into sensor deviation components. Model structure deviation components Parameter drift deviation components and time delay deviation components There are four types of components, and the contribution of each deviation component is identified based on its energy proportion. This leads to explainable results for tracing the sources of bias.

[0068] Furthermore, the total deviation source decomposition model is as follows:

[0069] (12)

[0070] The identification method for each deviation component is as follows: sensor deviation component The deviation components of the model structure are obtained by extracting the high-frequency components of the deviation signal in the frequency domain. The parameter drift deviation component is obtained by analyzing the mean and systematic shift of the deviation signal over a long time window. Time delay deviation component obtained through low-frequency gradual change trend analysis under a sliding window. pass and The peak position shift of the cross-correlation function is estimated. This represents the residual unmodelable random noise term; the contribution formula for each bias component is as follows:

[0071] (13)

[0072] in, The contribution of the sensor bias component. The contribution of the model structure bias component. The contribution of the parameter drift deviation component. The contribution of the time delay deviation component.

[0073] S3. Adaptive hierarchical compensation: Based on the deviation components and their contribution rates identified in step S2. The system invokes a compensation sub-strategy that matches the deviation component to implement tiered compensation, and adaptively adjusts the fusion weight of each sub-strategy based on the welding condition characteristics and the change in the overall consistency index. The compensated digital twin output ;

[0074] Furthermore, the compensated digital twin output is represented as:

[0075] (14)

[0076] in, The compensation function corresponding to the i-th type of deviation component is configured as follows for different deviation components:

[0077] For sensor bias components , Kalman filtering is used to denoise and perform optimal state estimation on high-frequency measurement signals such as welding clamp current and electrode pressure.

[0078] For model structure deviation components , A residual neural network is used to learn and compensate for the structural deviation between the mechanism model and the actual welding thermo-mechanical coupling process;

[0079] For parameter drift bias components , The recursive least squares method is used to identify and update slowly varying parameters such as electrode wear coefficient and fixture stiffness coefficient online.

[0080] For time delay deviation components , A Long Short-Term Memory (LSTM) network time-series prediction algorithm is used to proactively compensate for state misalignment caused by network communication and computation delays;

[0081] The fusion weight Satisfying constraints and .

[0082] Furthermore, integrate weights The update rules employ an online optimization strategy aimed at maximizing the improvement in the overall consistency index.

[0083] (15)

[0084] in, For learning rate, This represents the increase in the overall consistency index resulting from this round of compensation. The overall consistency index before compensation. The overall consistency index is obtained after compensation; the fusion weight update further includes contribution. Perform initialization: This ensures that the fusion weights are initially aligned with the bias energy distribution.

[0085] S4. Closed-loop iterative convergence: Output the compensated digital twin. Feedback is sent to step S1 to recalculate the overall consistency index. ;like If the preset convergence criterion is met, the compensation process ends and the final twin state and compensation source tracing report are output; otherwise, the process ends. The new simulation output is returned to step S2 for the next iteration, until the convergence criterion is met or the maximum number of iterations is reached. .

[0086] Furthermore, the convergence criterion includes any one or a combination of the following conditions:

[0087] (a) Consistency compliance conditions: threshold The range of values ​​is determined by the physical process standards for welding islands;

[0088] (b) Incremental convergence condition: Convergence accuracy The value range is [0.001, 0.01];

[0089] (c) Iteration count limit condition: The current iteration count reaches the limit. , The value range is [5, 20];

[0090] When any condition is met, the method terminates the iteration and outputs the final compensation result and compensation traceability report. The traceability report includes the contribution of each deviation component, the fusion weight of each compensation sub-strategy, the consistency index change trajectory, and the corresponding welding process identifier.

[0091] Furthermore, after each iteration, the deviation source decomposition results, compensation sub-strategy fusion weights, and consistency index changes of this round are written into the historical experience base as samples. In the next iteration or when a new workpiece of the same type of welding island is started, the fusion weights are initialized and the compensation sub-strategy is recommended based on the historical experience base to accelerate the iteration convergence process and support knowledge accumulation across workpieces and shifts.

[0092] Furthermore, when the method is applied to an automotive island-type welding scenario, it further includes: using the deviation of critical RPS points on the body-in-white, the weld point position deviation, and the welding current-pressure deviation as... and The core inputs are the single-station cycle time deviation and the cycle time deviation between upper and lower parts. The core input; using the heterogeneous timestamp deviation of the robot controller, PLC, and vision inspection system as... The core input; and the compensated twin output is written back to the MES and production line monitoring system in real time for solder joint quality early warning and cycle time optimization decision-making.

[0093] This invention also provides a system for tracing and hierarchical compensation of the consistency of welding data in automotive island-type structures, specifically including:

[0094] Data acquisition and spatiotemporal alignment module: used to acquire measured data of the physical entity of the automotive welding island and output data of the digital twin simulation, and to perform timestamp synchronization and spatial coordinate system registration to obtain spatiotemporal aligned data pairs;

[0095] Multi-dimensional quantity module: connected to the data acquisition and spatiotemporal alignment module, used to construct consistency sub-indicators based on the spatiotemporal alignment data pair in the dimensions of geometric pose, physical process, behavioral beat and temporal synchronization, and obtain a comprehensive consistency index through scene adaptive weighted fusion.

[0096] Deviation source decomposition module: connected to the multi-dimensional quantity module, used to calculate the total deviation of the spatiotemporal aligned data pair, decompose the total deviation into four components: sensor deviation, model structure deviation, parameter drift deviation, and time delay deviation, identify the contribution of each component, and output interpretable source decomposition results.

[0097] Adaptive hierarchical compensation module: connected to the deviation source decomposition module, used to call the corresponding compensation sub-strategy for each deviation component and adaptively adjust the fusion weight, and output the compensated digital twin state;

[0098] Closed-loop iterative convergence module: It is connected to both the adaptive hierarchical compensation module and the multi-dimensional quantity module. It is used to evaluate the compensation effect and trigger the iterative process until the preset convergence criterion is met, and generate a compensation source tracing report that includes the contribution of the deviation component, the fusion weight of each compensation sub-strategy, and the trajectory of the consistency index change.

[0099] Furthermore, taking the welding island on the side of the body-in-white as an example: the welding island consists of 6 six-axis welding robots, 6 servo welding guns, 1 set of RPS positioning fixtures, and 1 set of loading and unloading mechanisms, with 48 welding points per station and a nominal cycle time of 58 seconds. On the physical side, the robot controller outputs TCP pose at 250Hz, the servo welding guns output electrode pressure and welding current at 1kHz, the laser tracker and 3D vision output the positions of key RPS points on the body-in-white at 100Hz, and the PLC outputs the fixture clamping status and cycle time stamp at 50Hz; on the twin side, the welding thermo-mechanical coupling model, the robot dynamics model, and the fixture stiffness model jointly output the corresponding simulation state variables.

[0100] The timestamp synchronization adopts the IEEE1588PTP protocol, and the master clock is deployed at the edge gateway of the welding island, with an alignment accuracy better than 1ms. Spatial coordinate system registration is carried out according to formula (1): 11 sets of corresponding points are selected, including 4 calibration blocks of the fixture reference surface, robot flange reference point and 6 RPS points of the body-in-white. R and T are solved by the least squares method to make the digital twin coordinate system coincide with the physical island absolute coordinate system.

[0101] The total deviation is decomposed according to equation (3). High-frequency noise components of the sensor are identified above 200Hz in the frequency domain. Extracting mean shift to identify model structural bias within a 60s (one beat) long window. Extracting low-frequency, slowly varying trend identification parameters and drift bias within a 4-hour (half-shift) sliding window. Time delay deviation is identified by the peak offset of the cross-correlation function between physical and twin signals. .

[0102] Compensation is applied to the four types of components according to formula (5): Extended Kalman filtering is used to perform optimal state estimation of welding current and electrode pressure signals; A 6-layer residual neural network with 64 neurons in each layer is used to learn the residuals of the thermo-mechanical coupling mechanism model; Using a forgetting factor ( The recursive least squares method (=0.98) is used to identify the electrode wear coefficient and the fixture stiffness coefficient online. A single-layer LSTM (32 hidden units) is used to perform two-step lead prediction compensation for a 50ms communication delay. (Fusing weights) Based on contribution Initialize and update online according to formula (6). =0.05).

[0103] Convergence Criterion , , The typical iterative trajectory in this embodiment is as follows: ,in, It is the initial overall consistency index. It is the overall consistency index obtained from the first round of iteration. It is the overall consistency index obtained from the second round of iteration. It is the comprehensive consistency index obtained from the third iteration. The third iteration meets the criteria and terminates, outputting the final twin state and compensation traceability report.

[0104] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0105] Although the invention has been described with respect to a limited number of embodiments, those skilled in the art will understand from the foregoing description that other embodiments are conceivable within the scope of the invention described herein. Furthermore, it should be noted that the language used in this specification has been chosen primarily for readability and instructional purposes, and not for the purpose of explaining or limiting the subject matter of the invention.

Claims

1. A method for traceability and hierarchical compensation of consistency in automotive island-type welding assembly, characterized in that, Includes the following steps: S1. Multi-dimensional measurement: Synchronously collect measured state data of the physical entity of the automotive island welding assembly. Simulation output data of the corresponding digital twin The measured state data includes at least the welding robot tool center point pose, welding clamp electrode pressure and current, key measurement point positions of the body-in-white, fixture positioning pin status, and process cycle time stamps. The measured state data and simulation output data are synchronized with time stamps and registered in a spatial coordinate system to obtain spatiotemporally aligned data pairs. , The simulation output data is synchronized with timestamps and registered with spatial coordinate systems; Based on this spatiotemporally aligned data pair, consistency sub-indices are constructed in the dimensions of geometric pose, physical process, behavioral beat, and temporal synchronization. , , and A comprehensive consistency index is obtained through scenario-adaptive weighted fusion. ; S2, Deviation Source Decomposition: Calculate the total deviation of the spatiotemporally aligned data pairs described in S1. The total deviation is decomposed into sensor deviation components. Model structure deviation components Parameter drift deviation components and time delay deviation components There are four types of components, and the contribution of each deviation component is identified based on its energy proportion. This leads to explainable results for tracing the sources of bias. S3. Adaptive hierarchical compensation: Based on the deviation components and their contribution rates identified in step S2. The system invokes a compensation sub-strategy that matches the deviation component to implement tiered compensation, and adaptively adjusts the fusion weight of each sub-strategy based on the welding condition characteristics and the change in the overall consistency index. The compensated digital twin output ; S4. Closed-loop iterative convergence: Output the compensated digital twin. Feedback is sent to step S1 to recalculate the overall consistency index. ;like If the preset convergence criterion is met, the compensation process ends and the final twin state and compensation source tracing report are output; otherwise, the process ends. The new simulation output is returned to step S2 for the next iteration, until the convergence criterion is met or the maximum number of iterations is reached. .

2. The method according to claim 1, characterized in that, The timestamp synchronization in step S1 is achieved using the Precise Time Protocol (PTP), with an alignment accuracy better than 1ms; the spatial coordinate system registration is achieved through the following coordinate transformation relationship: （1） in, Let be a rotation matrix. As translation vectors, both are deployed within the welding island at calibration feature points on the fixture reference surface, robot flange, and body-in-white reference point system, and are solved using the least squares method.

3. The method according to claim 1, characterized in that, The fusion formula for the comprehensive consistency index mentioned in step S1 is: （2） in, The geometric pose consistency sub-index is calculated based on the normalized reciprocal of the Euclidean distance between the robot tool center point pose and the vehicle body measurement point. It is a sub-index of physical process consistency, calculated based on the reciprocal of the normalized root mean square error of welding current, electrode pressure, electrode temperature, and weld nugget stress. The behavior rhythm consistency sub-index is calculated based on the dynamic time warping distance between the welding action sequence and the loading and unloading action sequence; The temporal synchronization consistency sub-index is calculated based on the cross-correlation coefficient of the state time series; weights The weights are dynamically assigned based on the welding conditions and process standards, where j is the weight index.

4. The method according to claim 1, characterized in that, The total deviation source decomposition model in step S2 is as follows: （3） The identification method for each deviation component is as follows: sensor deviation component The deviation components of the model structure are obtained by extracting the high-frequency components of the deviation signal in the frequency domain. The parameter drift deviation component is obtained by analyzing the mean and systematic shift of the deviation signal over a long time window. Time delay deviation component obtained through low-frequency gradual change trend analysis under a sliding window. pass and The peak position shift of the cross-correlation function is estimated. This represents the residual, unmodelable random noise term; The formula for calculating the contribution of each deviation component is as follows: （4） in, The contribution of the sensor bias component. The contribution of the model structure bias component. The contribution of the parameter drift deviation component. The contribution of the time delay deviation component.

5. The method according to claim 4, characterized in that, The compensated digital twin output described in step S3 is represented as follows: （5） in, The compensation function corresponding to the i-th type of deviation component is configured as follows for different deviation components: For sensor bias components , Kalman filtering is used to denoise and perform state-optimal estimation on high-frequency measurement signals; For model structure deviation components , A residual neural network is used to learn and compensate for the structural deviation between the mechanism model and the actual welding thermo-mechanical coupling process; For parameter drift bias components , The recursive least squares method is used for online identification and updating of slowly varying parameters; For time delay deviation components , A long short-term memory network timing prediction algorithm is used to proactively compensate for state misalignment caused by network communication and computation delays. The fusion weight Satisfying constraints and .

6. The method according to claim 5, characterized in that, The fusion weight The update rules employ an online optimization strategy aimed at maximizing the improvement in the overall consistency index. （6） in, For learning rate, This represents the increase in the overall consistency index resulting from this round of compensation. The overall consistency index before compensation. The overall consistency index is obtained after compensation; the fusion weight update further includes contribution. Perform initialization: .

7. The method according to claim 6, characterized in that, The convergence criterion in step S4 includes any one or a combination of the following conditions: (a) Consistency compliance conditions: threshold The range of values ​​is determined by the physical process standards for welding islands; (b) Incremental convergence condition: Convergence accuracy The value range is [0.001, 0.01]; (c) Iteration count limit condition: The current iteration count reaches the limit. , The value range is [5, 20]; When any condition is met, the method terminates the iteration and outputs the final compensation result and compensation traceability report. The compensation traceability report includes the contribution of each deviation component, the fusion weight of each compensation sub-strategy, the consistency index change trajectory, and the corresponding welding process identifier.

8. The method according to claim 1, characterized in that, After each iteration, the method writes the deviation source decomposition results, compensation sub-strategy fusion weights, and consistency index changes of the current iteration into the historical experience database as samples. In the next iteration or when a new workpiece of the same type of welding island is started, the fusion weights are initialized and compensation sub-strategies are recommended based on the historical experience database.

9. The method according to claim 3, characterized in that, When applied to automotive island welding scenarios, the method further includes: using the deviation of the critical reference point system of the body-in-white, the weld point position deviation, and the welding current-pressure deviation as... and The core inputs are the single-station cycle time deviation and the cycle time deviation between upper and lower parts. The core input; using heterogeneous timestamp deviations from robot controllers, programmable logic controllers, and vision inspection systems as... The core input; and the compensated twin output is written back to the manufacturing execution system and production line monitoring system in real time for solder joint quality early warning and cycle time optimization decision-making.

10. A system for traceability and hierarchical compensation of the consistency of welding data in an automotive island assembly, characterized in that, The system is used to implement the method according to any one of claims 1 to 9, and the system comprises five core modules: The data acquisition and spatiotemporal alignment module is used to acquire measured data of the physical entity of the automotive welding island and simulation output data of the digital twin, and to perform timestamp synchronization and spatial coordinate system registration to obtain spatiotemporal aligned data pairs. The multi-dimensional quantity module is connected to the data acquisition and spatiotemporal alignment module. It is used to construct consistency sub-indicators based on the spatiotemporal alignment data pair in the dimensions of geometric pose, physical process, behavioral beat and temporal synchronization, and obtain a comprehensive consistency index through scene adaptive weighted fusion. The deviation source decomposition module, connected to the multi-dimensional quantity module, is used to calculate the total deviation of the spatiotemporal aligned data pair, decompose the total deviation source into four components: sensor deviation, model structure deviation, parameter drift deviation, and time delay deviation, identify the contribution of each component, and output interpretable source decomposition results. An adaptive hierarchical compensation module, connected to the deviation source decomposition module, is used to call the corresponding compensation sub-strategy for each deviation component and adaptively adjust the fusion weight, and output the compensated digital twin state. The closed-loop iterative convergence module is connected to both the adaptive hierarchical compensation module and the multi-dimensional quantity module. It is used to evaluate the compensation effect and trigger the iterative process until the preset convergence criterion is met, and to generate a compensation source tracing report that includes the contribution of the deviation component, the fusion weight of each compensation sub-strategy, and the trajectory of the consistency index change.

Images