Green low-carbon intelligent assessment method for welding influence of steel structure

Abstract

The invention discloses a green low-carbon intelligent assessment method for the welding influence of a steel structure, and relates to the technical field of green construction, and the method comprises the steps: collecting multi-source heterogeneous data in the welding process of the steel structure in real time, and calculating and visualizing a carbon emission dynamic value in the welding process; performing parameter extraction on the multi-source heterogeneous data, and inputting the multi-source heterogeneous data into a pre-trained defect identification model to obtain a weld defect identification result; and on the basis of the carbon emission dynamic value and the weld defect identification result, with minimization of carbon emission and residual stress as targets, iterative simulation is conducted in the twin welding model, and a low-carbon-low-stress process parameter window is obtained through optimization. The technical problems that a traditional steel structure welding related evaluation mode cannot give consideration to the low-carbon requirement and the mechanical property guarantee, the evaluation result is one-sided, and the parameter adaptability is insufficient are solved, and the technical effects that welding low-carbon and low-stress collaborative guarantee is achieved, and green low-carbon evaluation and process optimization of steel structure welding are more accurate and efficient are achieved.

Description

Technical Field

[0001] This invention relates to the field of green construction technology, and in particular to a green and low-carbon intelligent assessment method for the impact of welding on steel structures. Background Technology

[0002] Green and low-carbon development is a crucial trend in the steel structure manufacturing industry. Welding, as a core process, directly impacts structural safety and the achievement of low-carbon goals due to its carbon emissions, weld defects, and residual stress. Accurate assessment and process optimization are therefore essential. Current technologies for steel structure welding assessment often consider low-carbon requirements and mechanical properties separately, relying on experience or single data detection methods, lacking integrated intelligent assessment tools. These methods exhibit significant limitations in complex welding scenarios: they cannot quantify the entire process's carbon emissions in real time, the correlation between defect identification and process parameters is inaccurate, and process parameters are mostly statically set, making them difficult to adapt to dynamic welding processes. This results in one-sided assessments, failing to meet the precise control requirements for ensuring both green and low-carbon development and mechanical performance in steel structure welding. Summary of the Invention

[0003] This application solves the technical problem that traditional steel structure welding-related evaluation methods cannot simultaneously take into account low carbon requirements and mechanical performance assurance, and the evaluation results are one-sided and the parameters are not well adapted.

[0004] To address the aforementioned technical issues, this application proposes a green and low-carbon intelligent assessment method for the impact of welding on steel structures. The method includes: real-time acquisition of multi-source heterogeneous data from the steel structure welding process; driving the operation of a twin welding model; real-time calculation and visualization of dynamic carbon emissions during the welding process; parameter extraction of the multi-source heterogeneous data and inputting it into a pre-trained defect identification model; real-time identification and classification of weld defects; association of causal relationships between defects and process parameters; obtaining weld defect identification results; and based on the dynamic carbon emissions and weld defect identification results, iterative simulation is performed in the twin welding model with the goal of minimizing carbon emissions and residual stress to optimize and obtain a low-carbon, low-stress process parameter window.

[0005] This application proposes one or more technical solutions, which have at least the following technical effects: This application uses multi-source data collected in real time during the steel structure welding process to drive a twin model to calculate dynamic carbon emission values. After defect identification and causal correlation processing of process parameters, relevant results are obtained. Combined with dual objectives, parameters are iteratively simulated and optimized in the twin model. Then, through real-time monitoring and closed-loop continuous comparison and adjustment, the technical effect of achieving synergistic protection of low carbon and low stress in welding is realized, making the green and low carbon assessment and process optimization of steel structure welding more accurate and efficient. Attached Figure Description

[0006] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0007] Figure 1 This is a flowchart illustrating a green, low-carbon, and intelligent assessment method for the impact of welding on steel structures, provided in an embodiment of this application.

[0008] Figure 2 This is a flowchart illustrating the calculation of dynamic carbon emission values ​​in a green and low-carbon intelligent assessment method for the impact of welding on steel structures, provided in an embodiment of this application. Detailed Implementation

[0009] This application provides a green and low-carbon intelligent assessment method for the impact of welding on steel structures, which solves the technical problems that traditional assessment methods for welding of steel structures cannot take into account both low-carbon requirements and mechanical performance assurance, and the assessment results are one-sided and the parameters are not well adapted.

[0010] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0011] It should be noted that any variation of the terms "comprising" and "having" is intended to cover non-exclusive inclusion, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such processes, methods, products, or devices.

[0012] like Figure 1 As shown, a method for assessing the impact of welding on green, low-carbon, and intelligent systems for steel structures is provided, wherein the method includes: Multi-source heterogeneous data from the steel structure welding process are collected in real time to drive the twin welding model and calculate and visualize the dynamic value of carbon emissions during the welding process in real time.

[0013] In this embodiment, the twin welding model is a dedicated integrated simulation model of digital twin technology in welding scenarios. It is used to drive the dynamic calculation model to calculate carbon emissions, perform thermo-mechanical coupling simulation to predict residual stress, and carry out iterative optimization of multi-objective optimization algorithms.

[0014] Specifically, firstly, a dynamic calculation model covering both direct and indirect carbon emissions is established. The twin welding model is driven by the collected multi-source heterogeneous welding data. The dynamic calculation model is then called to complete the calculation of the two types of carbon emissions and form a green factor matrix. Finally, the green factor matrix is ​​weighted and quantitatively evaluated to generate and visualize the dynamic value of carbon emissions in the welding process.

[0015] After extracting parameters from the multi-source heterogeneous data, the data is input into a pre-trained defect recognition model to identify and classify weld defects in real time, and to associate the causal relationship between defects and process parameters to obtain weld defect recognition results.

[0016] Optionally, feature parameters containing spatial location and temporal correlations are first extracted from multi-source heterogeneous data to form a spatiotemporal matrix of welding process features. Submatrices related to weld morphology and thermal processes are then input into a pre-trained defect recognition model with a hybrid architecture of convolutional neural network and attention mechanism. This model simultaneously processes image-based temperature field distribution data and point cloud-based weld surface data, outputting the weld defect type and three-dimensional spatial coordinates. Finally, this output is temporally correlated and matched with the welding process parameters at the corresponding time. Causal inference analysis is used to determine the key parameter deviation events that cause defects, ultimately yielding the weld defect recognition result.

[0017] Based on the dynamic values ​​of carbon emissions and the results of weld defect identification, with the goal of minimizing carbon emissions and residual stress, iterative simulations are performed in the twin welding model to optimize the low-carbon, low-stress process parameter window.

[0018] In one embodiment of this application, the optimization process for the low-carbon-low-stress process parameter window is as follows: With the goal of minimizing carbon emissions and residual stress, welding current, voltage, speed, interpass temperature, and welding sequence are set as optimization variables in the twin welding model, and the allowable process range is defined as the search space. A multi-objective optimization algorithm is used, combined with weld defect identification results, to generate several sets of candidate process parameter combinations within the search space. For each candidate combination, the twin welding model is driven to perform rapid simulation, and the residual stress distribution is calculated through thermo-mechanical coupling. Simultaneously, a dynamic calculation model is invoked to predict carbon emission values, ultimately selecting a set of process parameters that meet both green and low-carbon requirements and mechanical performance constraints, thus forming the low-carbon-low-stress process parameter window.

[0019] Furthermore, the method provided in this application embodiment includes: The multi-source heterogeneous data includes welding process parameters, environmental parameters, weld morphology point cloud data, spatiotemporal distribution data of welding temperature field, and welding stress evolution data.

[0020] Specifically, for the acquisition of welding process parameters, current sensors, voltage sensors and welding speed encoders supporting the welding equipment are used. These sensors are respectively connected to the output terminal of the welding power source and the driving mechanism of the welding carriage. After the welding operation is started, the sensors capture the instantaneous values of the welding current and voltage and the moving speed of the welding carriage in real time, and transmit the data to the data processing terminal in real time through wired transmission. The acquisition frequency is set to 100 Hz to ensure accurate capture of the dynamic changes of the process parameters.

[0021] For the acquisition of environmental parameters, general-purpose digital temperature and humidity sensors and barometric pressure sensors are selected and fixed at positions within 1 meter around the welding station without affecting the welding operation. The sensors are pre-calibrated, and the acquisition frequency is set to 1 time per second. The environmental temperature, relative humidity and atmospheric pressure data of the welding area are detected in real time, and the digital signals collected are transmitted to the data processing terminal through a wireless Bluetooth module to achieve interference-free real-time acquisition of environmental parameters.

[0022] The acquisition of weld morphology point cloud data uses a portable three-dimensional laser scanner. The scanner is fixed on an adjustable bracket, and the height and angle of the bracket are adjusted so that the scanning range completely covers the weld and the areas on both sides with a width of 50 mm. The scanning accuracy of the scanner is set to 0.1 mm and the scanning frequency is set to 10 Hz. The scanning is started synchronously during the welding process. The scanner obtains the three-dimensional coordinate information of the weld surface by emitting laser beams, directly generates a point cloud data file, and transmits it to the data storage unit in real time through a USB interface to ensure the temporal synchronization of the morphology data with the welding process.

[0023] The acquisition of the spatio-temporal distribution data of the welding temperature field is carried out with an infrared thermal imager. The thermal imager is installed on a stable bracket on the side of the welding station, and the lens is aimed at the center area of the weld. The focal length is adjusted to make the weld area clearly imaged. The temperature measurement range of the thermal imager is set to 0 - 1500 °C and the frame rate is set to 30 frames per second. After the welding starts, the thermal images of the weld and the surrounding area are continuously captured. The temperature value corresponding to each pixel point is automatically extracted from each frame of image through the built-in software, and combined with the shooting timestamp to form a spatio-temporal distribution data set of the temperature field containing spatial position, temperature value and time information, which is uploaded to the data processing terminal in real time.

[0024] The acquisition of welding stress evolution data uses acoustic emission sensors. The sensors are pasted on the non-welded area of the steel structure near the weld through a high-temperature coupling agent to ensure that the sensors are closely fitted to the workpiece surface. The number of sensors is set to 2 - 4 according to the weld length and they are evenly distributed on both sides of the weld. The sensors are connected to an acoustic emission detector, and the detection frequency is set to 1 MHz. During the welding process, the sensors capture the elastic wave signals generated by the steel structure due to stress changes. The detector converts the original signals into stress change curve data and transmits it to the data processing terminal in real time through a wired network to achieve continuous monitoring of the stress evolution process.

[0025] Each of the above types of data acquisition devices is communicatively connected to the same data processing terminal, and the terminal synchronizes and aligns the received welding process parameters, environmental parameters, weld appearance point cloud data, spatio-temporal distribution data of the welding temperature field, and welding stress evolution data, and integrates them to form a multi-source heterogeneous data set based on time stamps.

[0026] Further, as Figure 2 shown, the method provided by the embodiment of the present application includes: Establish a dynamic calculation model including direct carbon emissions and indirect carbon emissions; drive the twin welding model to run with the multi-source heterogeneous data, call the dynamic calculation model to calculate direct carbon emissions and indirect carbon emissions, and form a green factor matrix; perform weighted quantitative evaluation based on the green factor matrix to generate the dynamic carbon emission value.

[0027] Optionally, first construct a dynamic calculation model, and first clarify the accounting boundary and calculation logic of direct carbon emissions and indirect carbon emissions. For direct carbon emissions, use the existing power measurement method, collect the instantaneous power data during the welding process in real time through the power meter supporting the welding equipment, calculate the real-time power consumption by combining time integration, and at the same time obtain the current regional power grid carbon emission factor from the public channels of the regional ecological environment department or power grid enterprise, and multiply the power consumption by this factor to obtain the direct carbon emission accounting formula, which is used as the core of the direct carbon emission calculation module of the dynamic calculation model.

[0028] For indirect carbon emissions, for the consumption of welding materials, use the weighing method to pre-weigh the mass of welding materials before and after welding, calculate the consumption rate of welding materials per unit time, and look up the unit mass carbon emission coefficient of this type of welding material in the industry standard, and multiply the two to obtain the indirect carbon emissions of welding materials; for the shielding gas, collect the flow data in real time through the gas flow meter, and calculate the indirect carbon emissions of the shielding gas by combining the unit volume carbon emission coefficient corresponding to the gas type; for the post-weld treatment materials, use the volumetric measurement method to record the material usage, and convert the indirect carbon emissions of the post-weld treatment materials according to the unit usage carbon emission data provided by the material manufacturer. Integrate the accounting formulas of the three types of indirect carbon emissions into the indirect carbon emission calculation module of the dynamic calculation model, and finally integrate the direct and indirect carbon emission calculation modules to form a complete dynamic calculation model, and reserve data input and output interfaces for convenient docking with other models.

[0029] Next, the construction of the twin welding model was carried out. First, based on the design drawings of the steel structure welded workpiece, a 1:1 geometric model of the actual workpiece was built using CAD software such as SolidWorks to ensure that the weld position and dimensions were consistent with the actual workpiece. Then, ANSYS software was used to mesh the geometric model, with a fine mesh used in the weld area and a conventional mesh used in other areas to improve simulation accuracy and computational efficiency. Afterwards, thermal and mechanical field simulation modules were integrated. The thermal field module is based on the Fourier heat conduction equation, and the mechanical field module is based on elastoplastic mechanics theory, both using mature solvers from the simulation software.

[0030] For the functional interface design of the twin welding model, a data input interface is designed for the dynamic calculation model, using the TCP / IP transmission protocol to support the reception of data such as power consumption, welding material consumption rate, and shielding gas flow rate. A carbon emission data receiving interface is also designed to obtain the direct and indirect carbon emission results output by the dynamic calculation model. For the defect identification model, a defect information input interface is reserved to support the reception of results such as defect type, 3D spatial coordinates, and time-series correlation data. The interface data format is uniformly JSON to ensure data compatibility. Furthermore, a data synchronization module is added to the twin welding model to achieve real-time alignment of carbon emission data from the dynamic calculation model, defect identification results from the defect identification model, and welding process simulation data based on timestamps. This lays the foundation for subsequent collaborative simulation and optimization, ultimately forming a twin welding model with data reception, simulation calculation, and interface integration functions.

[0031] After completing the construction of the two models mentioned above, the twin welding model is driven by multi-source heterogeneous data. The multi-source heterogeneous data, such as welding process parameters and environmental parameters collected in the previous steps, are transmitted in real-time to the data source interface of the twin welding model via a data bus. Upon model startup, the data synchronization module is automatically triggered to classify and organize the multi-source data according to timestamps. Subsequently, the twin welding model activates the integrated dynamic calculation model interface through preset calling commands, pushing the organized energy consumption data, welding material consumption rate, shielding gas flow rate, and other relevant data to the dynamic calculation model in real-time. After receiving the data, the dynamic calculation model performs synchronous calculations through the direct and indirect carbon emission calculation modules to obtain the direct carbon emission value and various indirect carbon emission values ​​at each time point. These carbon emission data are then organized according to time series to construct a green factor matrix. The row vectors of the matrix correspond to different time points, and the column vectors correspond to direct carbon emissions, welding material indirect carbon emissions, shielding gas indirect carbon emissions, and post-treatment material indirect carbon emissions, respectively, achieving structured storage of carbon emission data.

[0032] Finally, based on the constructed green factor matrix, the weights of each carbon emission type were determined using the analytic hierarchy process (AHP). Three to five experts in welding engineering and low-carbon accounting were invited to score the impact of different carbon emission types on the overall carbon emissions of welding. After consistency testing, the weight values ​​corresponding to each column vector were obtained. Each element in the green factor matrix was multiplied by its corresponding weight, and the weighted values ​​at each time point were summed to obtain the dynamic carbon emission value for that time point. Simultaneously, the real-time plotting function of professional plotting software was used to plot the dynamic carbon emission values ​​for each time point as a change curve in chronological order. This curve was then displayed in real-time through the visualization module of the twin welding model, allowing operators to intuitively understand the changes in carbon emissions during the welding process.

[0033] Furthermore, the method provided in this application embodiment includes: The direct carbon emissions are calculated based on real-time electricity consumption data and the regional power grid carbon emission factor; the indirect carbon emissions are calculated based on the welding material consumption rate, shielding gas flow rate and the amount of post-weld treatment materials used.

[0034] Specifically, the first step is to prepare for direct carbon emission calculations. A digital power meter is used, connected in series between the welding power source and the power supply line to ensure a stable electrical connection between the power meter and the welding equipment. After the welding operation starts, the power meter captures instantaneous power data in real time at a sampling frequency of 10Hz, automatically recording the instantaneous power value every second. The data is transmitted to the data processing terminal in real time via an RS485 interface. The terminal uses a time integration method to multiply the instantaneous power value in each time interval by the time interval, and sums them to obtain the real-time energy consumption for the corresponding time period. At the same time, the current regional power grid carbon emission factor is obtained through existing channels such as the official website of the regional power grid company and the public database of the ecological and environmental departments. This factor is updated monthly to ensure data timeliness. The terminal multiplies the real-time energy consumption by the current regional power grid carbon emission factor to obtain the direct carbon emission value for the corresponding time period.

[0035] Subsequently, indirect carbon emission conversion was carried out. For welding material consumption, an electronic weighing method was used. Before welding, the total mass of welding material was weighed using an electronic scale with an accuracy of 0.1 grams and recorded. After welding, the remaining welding material was weighed again, and the difference between the two was the total welding material consumption. Combined with the total welding time, the welding material consumption rate per unit time was calculated. Existing industry standards such as the "Guideline for Carbon Emission Accounting of Welding Materials" were consulted to obtain the unit mass carbon emission coefficient corresponding to the welding material used. The welding material consumption rate was multiplied by the unit mass carbon emission coefficient to obtain the corresponding indirect carbon emissions of the welding material. For shielding gas, a digital gas flow meter was installed on the gas delivery pipeline. The flow meter was wirelessly connected to a data processing terminal to collect the instantaneous flow data of the shielding gas in real time. The average flow rate was calculated by minute, and combined with the welding time, the total shielding gas consumption was obtained. Based on the shielding gas type, the unit volume carbon emission data was obtained from the product technical specifications provided by the gas manufacturer. The indirect carbon emissions of the shielding gas were multiplied by the shielding gas consumption data to obtain the indirect carbon emissions of the shielding gas. For post-weld treatment materials, such as polishing agents and rust-preventive paints, the actual usage is recorded using volumetric measurement or electronic weighing methods. For example, the volume of liquid materials is measured using a graduated cylinder, and the mass of solid materials is weighed using an electronic scale. Based on the carbon emission parameters per unit usage provided by the material manufacturers, the indirect carbon emissions of post-weld treatment materials are calculated through multiplication.

[0036] Finally, the direct carbon emission values ​​and the indirect carbon emission values ​​corresponding to welding materials, shielding gases, and post-weld treatment materials are organized by timestamp to ensure that the various types of carbon emission data correspond one-to-one for each time period, providing structured and accurate basic data for the subsequent construction of the green factor matrix and the generation of dynamic carbon emission values.

[0037] Furthermore, the method provided in this application embodiment includes: Feature parameters containing spatial location and temporal correlation are extracted from the multi-source heterogeneous data to form a feature spatiotemporal matrix of the welding process. The sub-matrices related to weld morphology and thermal process in the feature spatiotemporal matrix are input into the pre-trained defect recognition model, wherein the defect recognition model is a hybrid architecture of convolutional neural network and attention mechanism, used to simultaneously process image-based temperature field distribution data and point cloud-based weld surface data. The defect type and three-dimensional spatial coordinates of the weld region are output through the defect recognition model. The defect type and three-dimensional spatial coordinates are temporally correlated and matched with the welding process parameters at the corresponding time. The key parameter deviation events that cause defects are determined through causal inference analysis to obtain the weld defect recognition result.

[0038] Specifically, the first step is feature parameter extraction. For the multi-source heterogeneous data collected in the preceding steps, spatial location and temporal correlation features are extracted separately. For weld morphology point cloud data, PCA principal component analysis is used to remove noise points from the point cloud, and then principal component features are extracted by calculating the covariance matrix, resulting in a parameter set containing geometric features such as weld width, height, and surface roughness. For the spatiotemporal distribution data of the welding temperature field, the gray-level co-occurrence matrix algorithm is used to extract texture and regional features such as temperature gradient and heat-affected zone range. For time-series data such as welding process parameters and stress evolution data, time windows are divided at 10-millisecond intervals, and temporal features such as the mean, peak value, and rate of change of parameters within each window are extracted. All extracted spatial and temporal features are aligned by timestamps to construct a uniformly dimensional spatiotemporal matrix of welding process features. The rows of the matrix correspond to time nodes, and the columns correspond to various feature parameters, ensuring that the matrix can completely represent the spatiotemporal variation law of the welding process.

[0039] Next, the defect recognition model was constructed and trained. This model employs a hybrid architecture of convolutional neural networks and an attention mechanism. The base convolutional neural network is ResNet50, which has strong feature extraction capabilities. A CBAM attention mechanism module is embedded between the convolutional and fully connected layers of the ResNet50. Through the synergistic effect of channel attention and spatial attention, the weights of weld defect-related features are strengthened, and interference from irrelevant background features is suppressed. The model input is set as a sub-matrix related to weld morphology and thermal processes in the feature spatiotemporal matrix. The morphology-related sub-matrix contains point cloud geometric feature parameters, and the thermal process-related sub-matrix contains temperature field texture and region feature parameters. Both types of sub-matrixes are converted to a 32×32 two-dimensional matrix format as model input. The model output is set as the weld defect type, such as porosity, crack, incomplete penetration, etc., and the corresponding three-dimensional spatial coordinates, i.e., x, y, and z axis values.

[0040] The training process for the defect identification model is as follows: First, sample data of common defects in steel structure welding are collected, covering different defect types, sizes, and welding conditions. A total of 10,000 valid samples are collected, of which 8,000 are used as the training set and 2,000 as the validation set. Data augmentation processing is performed on the sample data, including random rotation, scaling, and mirroring, to improve the model's generalization ability. During training, the cross-entropy loss function is used to measure the difference between the predicted results and the true labels. The Adam optimizer is used to adjust the model parameters. The initial learning rate is set to 0.001, and the learning rate decays to 0.9 times the original rate every 10 iterations. After 50 iterations of training, when the accuracy of the validation set stabilizes above 95%, training is stopped, the model parameters are saved, and the pre-training of the defect identification model is completed.

[0041] Subsequently, the sub-matrices related to weld morphology and thermal processes from the feature spatiotemporal matrix are input into the pre-trained defect recognition model according to the model's required format. The model extracts deep features from the sub-matrices through convolutional layers, strengthens key features through an attention mechanism module, and then feeds them into a fully connected layer for classification and coordinate regression calculations, outputting the defect type and corresponding three-dimensional spatial coordinates of the weld area in real time. The output results are stored in structured data format, with each result including the defect name, x-axis coordinate value, y-axis coordinate value, z-axis coordinate value, and recognition confidence score.

[0042] Finally, temporal correlation matching and causal inference analysis are performed. Specifically, welding process parameters, including welding current, voltage, welding speed, and interpass temperature, are retrieved from the data processing terminal that perfectly correspond to the timestamps of the defect identification results. A Bayesian network causal inference method is used, with defect type and three-dimensional spatial coordinates as result variables and the corresponding welding process parameters as cause variables, to construct a causal relationship network. By calculating the conditional probability between each process parameter and the defect type, process parameters with probability values ​​exceeding a preset threshold, such as 0.8, are selected to identify key parameter deviation events that cause the defect, such as current fluctuations exceeding ±10A or voltage deviations from the set value of ±2V. Finally, the defect information and key parameter deviation events are integrated to form a complete weld defect identification result.

[0043] Furthermore, the method provided in this application embodiment includes: In the twin welding model, welding current, voltage, speed, interpass temperature, and welding sequence are set as optimization variables, and the allowable process range is set as the search space. A multi-objective optimization algorithm is used to generate several sets of candidate process parameter combinations within the search space based on the weld defect identification results. For each set of candidate process parameter combinations, the twin welding model is driven to perform rapid simulation, and the residual stress distribution is calculated by thermo-mechanical coupling. The dynamic calculation model is called simultaneously to predict carbon emission values, and a set of process parameters that meet the dual constraints of green low-carbon and mechanical performance is selected as the low-carbon-low-stress process parameter window.

[0044] Specifically, the optimization variables and search space are first defined in the twin welding model. Those skilled in the art need to refer to existing steel structure welding industry standards and the rated operating parameters of the welding equipment to set the allowable ranges for welding current, welding voltage, and welding speed. The interpass temperature is determined based on the steel material; for example, 80-150℃ for low-carbon steel and 100-200℃ for high-strength steel. The welding sequence is listed as 3-5 common options, such as sequential welding, reverse welding, and segmented welding. These allowable ranges are integrated into the search space and preset in the parameter configuration module of the twin welding model to ensure that the search process does not exceed the feasible boundaries of the actual welding process.

[0045] Next, the NSGA-II multi-objective optimization algorithm is used to transform the defect types and key parameter deviation events in the weld defect identification results into constraints. For example, if a porosity defect is identified and causal inference determines that it is related to welding current fluctuations, then current stability is included as a constraint. Subsequently, the algorithm initializes the population size to 100 and sets the maximum number of iterations to 50. Through roulette wheel selection, single-point crossover, and random mutation operations, 100 sets of candidate process parameter combinations are randomly generated within the preset search space. Each combination includes specific values ​​or methods for welding current, voltage, speed, interpass temperature, and welding sequence.

[0046] Next, rapid simulation and data calculation of candidate parameter combinations were performed. First, a Kriging surrogate model was trained. Based on high-fidelity simulation data samples covering simulation results of heat source loading and heat transfer processes under different parameter combinations, the Kriging surrogate model was constructed using the DACE toolbox in MATLAB software. Model parameters were adjusted through cross-validation to ensure the model prediction error was controlled within 5%. For each candidate process parameter combination, it was input into a twin welding model. The model called the trained Kriging surrogate model to replace repetitive heat source loading and heat transfer calculations. Then, combined with the thermo-mechanical coupling solver of ANSYS software, the residual stress distribution in the weld and surrounding area was calculated based on thermo-elastic-plastic theory, outputting the maximum residual stress value and stress concentration area information. Simultaneously, the twin welding model called a dynamic calculation model through a preset interface, inputting data such as the estimated power consumption and welding material consumption rate corresponding to the candidate parameter combinations into the dynamic calculation model, which then quickly predicted the carbon emission value under that parameter set.

[0047] Finally, parameter combinations were screened. The green and low-carbon constraint was set as follows: the predicted carbon emission value should not exceed 85% of the industry average, a value determined with reference to publicly available industry statistics on carbon emissions from steel structure welding. The mechanical performance constraint was that the maximum residual stress value should not exceed 80% of the yield strength of the welded steel, set based on the mechanical property parameters of the steel material. The predicted carbon emission value and residual stress value of each candidate process parameter combination were double-verified. All parameter combinations that simultaneously met both constraints were selected. These combinations were then organized into a structured parameter set, clarifying the value range or optional method for each parameter, ultimately forming a low-carbon, low-stress process parameter window.

[0048] Furthermore, the method provided in this application embodiment includes: When driving the twin welding model for rapid simulation, a surrogate model is introduced. For repetitive heat source loading and heat transfer calculations, the Kriging surrogate model is pre-trained using high-fidelity simulation samples. The repetitive physical simulation process is then executed before driving the twin welding model.

[0049] In one embodiment, high-fidelity simulation samples are first prepared to provide a data foundation for training the Kriging surrogate model. Using ANSYS simulation software, based on the steel structure geometry model and mesh generation results already constructed from the twin welding model, the range of sample parameters is determined. Referring to the allowable range of welding processes, for example, the welding current is set to 80-350A, the voltage to 20-45V, the welding speed to 5-30mm / s, and the interpass temperature to 80-200℃. Three typical welding sequences are selected: forward welding, reverse welding, and segmented welding. Using the Latin hypercube sampling method, 200 sets of sample points are uniformly drawn within the above parameter range, with each set of sample points corresponding to a complete combination of process parameters. For each set of sample points, a high-fidelity thermo-mechanical coupling simulation is started in ANSYS. The heat source model is set strictly according to the welding physical process. A double ellipsoidal heat source model can be used. Material thermophysical parameters, such as thermal conductivity and specific heat capacity as they change with temperature, as well as boundary conditions, are calculated. After simulation, the temperature field evolution data, heat transfer efficiency data, and corresponding preliminary stress calculation results of the heat source loading area are output. The input parameters and output results of all samples are organized into a structured dataset as the original data for model training.

[0050] Subsequently, dataset preprocessing was performed to ensure the validity of the training data. Z-Score normalization was used to normalize input parameters such as current and voltage, and output results such as temperature field data and heat transfer efficiency, eliminating the impact of differences in parameter magnitudes on model training. Outliers in the dataset were identified using box plots, and outliers exceeding 1.5 times the interquartile range were removed, retaining 180 valid samples. These valid samples were then randomly divided in an 8:2 ratio: 144 samples were used as the training set for model parameter fitting, and 36 samples were used as the validation set for model accuracy evaluation. The partitioning process used a random function to ensure the randomness and representativeness of the partitioning results.

[0051] Next, the Kriging surrogate model is constructed and trained using the DACE toolbox in MATLAB software. This toolbox is mature and easy to use for implementing Kriging models. First, the variogram type of the model is defined, and the Gaussian variogram, widely used in industrial parameter prediction, is chosen. Its expression conforms to the smooth change characteristics of heat source loading and heat transfer processes, and can better fit the data patterns. Using the input parameters of the training set as independent variables and the output results as dependent variables, the DACE toolbox is used. Maximum likelihood estimation is employed to solve for the parameters of the Gaussian variogram, such as amplitude and correlation length. Iterative calculations minimize the model's fitting error to the training set data. During training, the model accuracy is evaluated every 10 iterations using validation set data. The average relative error between the predicted values ​​and the high-fidelity simulation values ​​is calculated. When the average relative error stabilizes within 5% after 50 iterations, training is stopped, the model's parameter configuration file is saved, and the construction of the Kriging surrogate model is complete.

[0052] The trained Kriging proxy model was then integrated with the twin welding model. A dedicated interface was designed in the simulation module of the twin welding model, using the TCP / IP communication protocol to achieve bidirectional data transmission between the two models. The interface design clearly defined the data format requirements: the input data transmitted from the twin welding model to the proxy model consisted of standardized combinations of process parameters, while the output data returned by the proxy model to the twin welding model included key characteristic values ​​of temperature field evolution, heat transfer efficiency, and preliminary stress prediction results. Data transmission latency was controlled within 10 milliseconds. Simultaneously, a trigger mechanism was set in the twin welding model. When repeated heat source loading and heat transfer calculations were required for multiple sets of candidate process parameter combinations, the proxy model call command was automatically activated without manual intervention, ensuring smooth collaborative operation of the integrated models.

[0053] During rapid simulation using the twin welding model, for each combination of candidate process parameters, the twin welding model first standardizes the parameters and then sends them to the Kriging surrogate model via an interface. The surrogate model, based on the trained parameter configuration, quickly calculates and returns the temperature field and heat transfer-related prediction results without initiating a complex physical simulation process. After receiving the prediction results, the twin welding model directly connects to the subsequent thermo-mechanical coupled stress calculation module, combining the prediction data to quickly solve for the residual stress distribution, and simultaneously calls the dynamic calculation model to predict carbon emissions. To ensure overall accuracy, a periodic calibration mechanism is implemented. After every 20 sets of candidate parameter simulations are completed, one set of parameters is randomly selected for high-fidelity physical simulation, and its results are compared with the surrogate model's prediction results. If the error exceeds 5%, the new high-fidelity sample is added to the training set for incremental training of the surrogate model, ensuring that the model maintains high accuracy throughout the iterative optimization process.

[0054] By employing simulation software, sampling methods, toolkits, and communication protocols, the construction, training, and integration of the Kriging surrogate model were fully realized, replacing the repetitive physical simulation process and significantly improving the iterative simulation efficiency of the twin welding model. This provides efficient technical support for the rapid optimization of low-carbon and low-stress process parameter windows.

[0055] Furthermore, the method provided in this application embodiment includes: The low-carbon, low-stress process parameter window and the corresponding predicted performance index are pushed to the welding control terminal. In the welding control terminal, a real-time monitoring closed loop is established to continuously collect actual welding data, calculate the actual carbon emission intensity and the stress signal monitored by acoustic emission, and compare them with the predicted performance index in real time. When the deviation exceeds the preset tolerance, the local iterative optimization is restarted according to the direction and magnitude of the deviation.

[0056] In this embodiment, the welding control terminal is an industrial-grade PLC or a touch screen controller.

[0057] Optionally, the low-carbon, low-stress process parameter window and predicted performance indicators are pushed first. The pushed content clearly includes the specific value range of each process parameter, the upper limit of predicted carbon emission intensity, the target value of residual stress control, and the estimated threshold for defect incidence. These data all come from the structured results of iterative optimization in the preceding steps. The MQTT communication protocol is used; this protocol is highly stable and has low latency in industrial equipment data transmission. A communication connection is established between the output interface of the twin welding model and the welding control terminal, with a communication baud rate configured at 9600bps to ensure real-time and reliable data transmission. After the push command is initiated, the twin welding model packages and sends the structured data according to a preset format. The welding control terminal receives the data, automatically parses it, and displays it in sections on the interface for easy viewing and access by operators.

[0058] Next, a real-time monitoring closed-loop system is established. When continuously collecting actual welding data, the sensor equipment deployed in the previous steps is used: power meters, electronic scales, and gas flow meters collect actual power consumption, welding material consumption rate, and shielding gas flow data, respectively; acoustic emission sensors collect stress elastic wave signals during the welding process. The acquisition frequency of all devices remains consistent with the previous steps to ensure data synchronization. The welding control terminal has a built-in data processing module that uses the same calculation method as in the previous steps to calculate the actual carbon emission intensity in real time; acoustic emission signal analysis software converts the elastic wave signals collected by the sensors into actual residual stress values, forming a complete dataset of actual performance indicators. The preset tolerance standards refer to industry welding quality control specifications, setting the deviation tolerance between actual carbon emission intensity and predicted values ​​to ±5%, and the deviation tolerance between actual residual stress and predicted values ​​to ±8%. The welding control terminal compares the actual performance indicators with the predicted performance indicators in real time and generates a comparison result log.

[0059] When the comparison results show that the deviation exceeds the preset tolerance, a local iterative optimization process is initiated. The welding control terminal automatically extracts the process parameters related to the deviation. For example, if the actual carbon emissions are high and the deviation exceeds the tolerance, the focus is on parameters affecting energy consumption, such as welding current, voltage, and welding speed. If the residual stress deviation is too large, the focus is on interpass temperature and welding sequence parameters. Without restarting the full iteration, these related parameters are used as local optimization variables, with the search space limited to ±10% of the original process parameter window, narrowing the optimization range to improve efficiency. The welding control terminal sends local optimization instructions to the twin welding model via the communication interface. This model calls the pre-trained Kriging surrogate model and dynamic calculation model to quickly perform 3-5 rounds of small-scale iterative simulation, recalculating the parameter combinations that satisfy the dual constraints, forming an updated local process parameter window.

[0060] Finally, the updated local process parameters are pushed to the welding control terminal to replace the original parameter configuration. At the same time, actual welding data under the new parameters is continuously collected, and the comparison process is repeated. If the deviation returns to the preset tolerance range, the current parameters are maintained; if the deviation still exists, local fine-tuning continues until the carbon emission intensity and stress state of the actual welding are stable within the tolerance range of the predicted performance indicators, forming a continuous adaptive monitoring closed loop.

[0061] Through the above-mentioned sequential steps, a complete closed loop was constructed, from parameter push and real-time monitoring to deviation correction. This enabled dynamic adaptation of welding process parameters, continuously ensuring the low-carbon and low-stress goals of the steel structure welding process, and improving the stability and control accuracy of welding quality.

[0062] In summary, the green, low-carbon, and intelligent assessment method for the welding impact on steel structures provided in this application has the following technical effects: This application achieves the technical effect of synergistic protection of low carbon and low stress in welding by real-time acquisition of multi-source heterogeneous data of steel structure welding, driving the twin model to calculate the dynamic value of carbon emissions, identifying defects and causally linking process parameters, iteratively optimizing process parameters with dual objectives, and monitoring, comparing and adjusting in a closed loop. This makes the green and low carbon assessment and process optimization of steel structure welding more accurate and efficient.

[0063] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0064] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for assessing the impact of welding on green, low-carbon, and intelligent systems for steel structures, characterized in that, include: Real-time acquisition of multi-source heterogeneous data from the steel structure welding process drives the twin welding model to calculate and visualize the dynamic value of carbon emissions during the welding process. After extracting parameters from the multi-source heterogeneous data, the data is input into a pre-trained defect identification model to identify and classify weld defects in real time, and to associate the causal relationship between defects and process parameters to obtain weld defect identification results. Based on the dynamic values ​​of carbon emissions and the results of weld defect identification, with the goal of minimizing carbon emissions and residual stress, iterative simulations are performed in the twin welding model to optimize the low-carbon, low-stress process parameter window.

2. The green, low-carbon, and intelligent assessment method for the impact of welding on steel structures as described in claim 1, characterized in that, The multi-source heterogeneous data includes welding process parameters, environmental parameters, weld morphology point cloud data, spatiotemporal distribution data of welding temperature field, and welding stress evolution data.

3. The green, low-carbon, and intelligent assessment method for the impact of welding on steel structures as described in claim 1, characterized in that, Real-time acquisition of multi-source heterogeneous data from the steel structure welding process drives the operation of a twin welding model, enabling real-time calculation and visualization of dynamic carbon emissions during the welding process, including: Establish a dynamic calculation model that includes both direct and indirect carbon emissions; The twin welding model is driven by the multi-source heterogeneous data, and the dynamic calculation model is called to calculate direct and indirect carbon emissions, forming a green factor matrix. The dynamic value of carbon emissions is generated by performing a weighted quantitative assessment based on the green factor matrix.

4. The green, low-carbon, and intelligent assessment method for the impact of welding on steel structures as described in claim 3, characterized in that, The direct carbon emissions are calculated based on real-time electricity consumption data and the regional power grid carbon emission factor; the indirect carbon emissions are calculated based on the welding material consumption rate, shielding gas flow rate and the amount of post-weld treatment materials used.

5. The green, low-carbon, and intelligent assessment method for the impact of welding on steel structures as described in claim 1, characterized in that, After extracting parameters from the multi-source heterogeneous data, the data is input into a pre-trained defect recognition model to identify and classify weld defects in real time, and to correlate the causal relationship between defects and process parameters, thereby obtaining weld defect recognition results, including: Feature parameters containing spatial location and temporal correlation are extracted from the multi-source heterogeneous data to form a feature spatiotemporal matrix of the welding process; The submatrices related to weld morphology and thermal process in the feature spatiotemporal matrix are input into the pre-trained defect recognition model, wherein the defect recognition model is a hybrid architecture of convolutional neural network and attention mechanism, used to simultaneously process image-based temperature field distribution data and point cloud-based weld surface data. The defect identification model outputs the defect type and three-dimensional spatial coordinates of the weld area. The defect type and three-dimensional spatial coordinates are matched with the welding process parameters at the corresponding time. The key parameter deviation events that cause the defect are determined by causal inference analysis, and the weld defect identification result is obtained.

6. The green, low-carbon, and intelligent assessment method for the impact of welding on steel structures as described in claim 1, characterized in that, Based on the aforementioned dynamic values ​​of carbon emissions and the results of weld defect identification, with the goal of minimizing carbon emissions and residual stress, iterative simulations are performed in the twin welding model to optimize and obtain a low-carbon, low-stress process parameter window, including: In the twin welding model, welding current, voltage, speed, interpass temperature and welding sequence are set as optimization variables, and the process allowable range is set as the search space; A multi-objective optimization algorithm is used to generate several sets of candidate process parameter combinations in the search space based on the weld defect identification results; For each set of candidate process parameter combinations, the twin welding model is driven to perform rapid simulation, and the residual stress distribution is calculated by thermo-mechanical coupling. At the same time, the dynamic calculation model is called to predict the carbon emission value, and the set of process parameters that meet the dual constraints of green and low carbon and mechanical performance is selected as the low carbon-low stress process parameter window.

7. The green, low-carbon, and intelligent assessment method for the impact of welding on steel structures as described in claim 6, characterized in that, When driving the twin welding model for rapid simulation, a surrogate model is introduced. For repetitive heat source loading and heat transfer calculations, the Kriging surrogate model is pre-trained using high-fidelity simulation samples. The repetitive physical simulation process is then executed before driving the twin welding model.

8. The green, low-carbon, and intelligent assessment method for the impact of welding on steel structures as described in claim 1, characterized in that, After obtaining the optimal low-carbon, low-stress process parameter window, the following steps are also included: The low-carbon, low-stress process parameter window and the corresponding predicted performance indicators are pushed to the welding control terminal. At the welding control terminal, a real-time monitoring closed loop is established to continuously collect actual welding data, calculate the actual carbon emission intensity and the stress signal monitored by acoustic emission, and compare them with the predicted performance indicators in real time. When the deviation exceeds the preset tolerance, the local iterative optimization is restarted according to the direction and magnitude of the deviation.

Images