Pre-control of welding deformation and stress release construction method for high-altitude steel structure of chemical plant

Abstract

The application discloses a high-altitude steel structure welding deformation pre-control and stress release construction method, belongs to the technical field of welding construction, and aims to solve the problems of low control precision of high-altitude welding deformation and stress release lag. The method arranges sensors on the structure, calibrates the current dynamic welding twin model by using the trial welding data, divides the weld, plans welding according to the model, and instantly identifies the stress concentration area as the intervention target area, and then implements online stress reduction by the following ultrasonic impact device; then, the model is iteratively updated based on the data collected in the whole process to guide the subsequent segmentation; finally, the target processing is performed on the measured residual stress exceeding area. The method is mainly used for high-altitude steel structure welding construction, and realizes the precise active control of deformation and stress.

Description

Technical Field

[0001] This invention relates to the field of welding construction technology. More specifically, this invention relates to a construction method for pre-controlling deformation and releasing stress during welding of high-altitude steel structures in chemical plants. Background Technology

[0002] In major industrial sectors such as petroleum, chemical, and energy, large-scale installations typically include numerous high-altitude steel structures, such as pipe racks, tower platforms, and large frames. The installation of these high-altitude steel structures primarily relies on on-site welding operations. High-altitude welding operations are characterized by a complex environment, featuring high spatial location, large component dimensions, strong restraint conditions, and stringent construction safety requirements. Welding is a process of rapid localized heating followed by cooling, inevitably introducing uneven thermal stress into the structure. This leads to welding deformations such as shrinkage, angular deformation, and bending, resulting in significant residual stress concentrations in the weld and heat-affected zone. For high-altitude steel structures, excessive welding deformation directly affects the installation accuracy, geometric dimensions, and stress state, even preventing the connection and installation of subsequent equipment. High residual stress significantly reduces the structure's fatigue strength and, under the combined action of corrosive media, induces stress corrosion cracking, becoming a long-term safety hazard. Therefore, effectively controlling welding deformation and timely releasing welding stress are core technical challenges for ensuring the construction quality and long-term safe operation of high-altitude steel structures.

[0003] Currently, in welding deformation control, engineering practice still heavily relies on traditional methods such as experience-based pre-deformation correction, pre-set rigid restraints, and optimized welding sequences. While these methods have some effectiveness, their application is highly dependent on the experience level of the construction personnel and lacks versatility. For each new structure and environment, a new approach is required, making precise quantitative pre-control difficult. Especially under complex restraint conditions at high altitudes, the deformation mechanism is more intricate, and traditional experience-based methods suffer from low prediction accuracy and poor reliability, often leading to rework and increased costs. This not only delays the project but also poses significant safety risks due to the high-altitude rework itself. In recent years, numerical simulation techniques such as the finite element method have been introduced into welding process analysis, predicting deformation and stress through computer simulation. However, such simulations are typically based on idealized material parameters, standard boundary conditions, and simplified heat source models. Their calculation results differ significantly from the dynamic factors of the real, changing environment at high altitudes, such as wind speed, sunlight, and actual installation restraint, leading to simulation distortion and making it difficult to directly guide high-precision on-site construction.

[0004] Traditional methods for stress relief in welding mainly include post-weld overall heat treatment, local heat treatment, hammering, and vibration aging. For high-altitude steel structures, overall heat treatment requires massive on-site heating equipment and insulation facilities, making it extremely difficult to implement, energy-intensive, and posing a fire risk, thus rendering it almost impractical. While local heat treatment and hammering can be implemented locally, their timing is delayed, typically occurring after welding is complete, stress has accumulated sufficiently, and deformation has solidified, resulting in limited corrective effects and reliance on operator experience. The force, location, and frequency of hammering are difficult to control precisely, posing a risk of damaging the base material; vibration aging has stringent requirements for the fixing method and excitation parameters of large structures, making it difficult to apply in confined spaces at high altitudes. A common problem with these methods is that they are all passive, post-construction stress reduction measures, unable to actively intervene and control the stress during the welding process.

[0005] While existing research has attempted to integrate monitoring technology with the welding process, such as using limited sensors to monitor temperature or deformation and adjust welding parameters locally, these methods are mostly in the open-loop or simple feedback stage. They lack a digital brain capable of integrating multi-source information, accurately reflecting the current working conditions, and dynamically evolving for global decision-making. Existing digital models of the welding process are often static and offline, unable to utilize real-time data continuously generated during construction to calibrate and evolve the model itself in real time. Consequently, they cannot adapt to dynamic disturbances such as wind speed, temperature differences, and changes in restraint conditions in high-altitude environments. The fundamental flaw lies in the fact that the existing technological system has failed to construct a complete intelligent closed loop of perception-decision-execution-learning. This results in insufficient precision in welding deformation pre-control, poor timeliness and weak targeting of stress release, making it difficult to meet the increasingly stringent requirements of precision, reliability, and efficiency in welding quality for high-altitude steel structures in modern chemical installations. Summary of the Invention

[0006] One object of the present invention is to solve at least the above-mentioned problems and to provide at least the advantages that will be described later.

[0007] Another objective of this invention is to provide a construction method for pre-controlling deformation and releasing stress during welding of high-altitude steel structures in chemical plants. This method achieves precise deformation pre-control by dynamically calibrating the model using trial welding data; it effectively prevents stress accumulation by employing segmented, accompanying online stress reduction; and iteratively optimizes the model and process using data from the entire process to form an adaptive control closed loop, significantly improving the quality and efficiency of high-altitude welding.

[0008] To achieve these objectives and other advantages according to the present invention, a construction method for pre-controlling welding deformation and releasing stress in high-altitude steel structures of chemical plants is provided, comprising: S1. After the high-altitude steel structure to be welded is in place, temperature sensors and strain sensors are arranged at multiple monitoring points as multi-source sensing units. S2. Before formal welding, control the welding torch to perform a trial weld on the initial weld area, and synchronously collect the thermal-mechanical response data of the trial weld process through the multi-source sensing unit. Based on the thermal-mechanical response data of the trial weld process, reverse-calibrate the model parameters of the preset welding twin basic model to generate a dynamic welding twin model that matches the current working conditions. S3. Divide the total weld into several construction segments, and perform the following steps sequentially during the welding process of each segment: execute welding based on the prediction and planning of the current dynamic welding twin model; implement online stress reduction intervention on the intervention target area of ​​the completed weld in the current segment using a portable ultrasonic impact device that moves synchronously with the welding torch, based on the transient stress distribution indication of the current dynamic welding twin model; and iteratively update the current dynamic welding twin model based on the actual thermo-mechanical response data and impact force-displacement response data collected during the entire welding and stress reduction intervention process, and apply it to the welding planning of the next segment. S4. After all segmented welding and iterative updates are completed, targeted post-processing is performed on the areas where the measured residual stress obtained by the multi-source sensing unit exceeds the final prediction result of the current dynamic welding twin model after iterative updates. In step S3, the intervention target area refers to the stress concentration area calculated by the current dynamic welding twin model at the instant the current segmented welding operation is completed. The stress concentration area includes the area where the Mises equivalent stress exceeds 70-85% of the yield strength of the material at room temperature at the corresponding calculation point of the current dynamic welding twin model.

[0009] Preferably, the stress concentration region further includes: At geometric discontinuities, the rate of change of principal stress per unit length exceeds 2.0 to 3.5 times the average stress per unit length of the cross section at the corresponding calculation point in the current dynamic welding twin model; The region where the transverse tensile stress component perpendicular to the weld direction exceeds 30-40% of the room temperature yield strength of the material at the corresponding calculation point in the current dynamic welding twin model.

[0010] Preferably, the multi-source sensing unit, welding torch, and portable ultrasonic impact device are all communicatively connected to the central processing unit; the multi-source sensing unit also includes an ambient temperature sensor that monitors the ambient temperature of the welding operation area in real time; the central processing unit is pre-set with the welding process specification of the high-altitude steel structure to be welded, the three-dimensional structural model of the high-altitude steel structure to be welded, and the welding twin basic model, the welding twin basic model including at least the following physical parameters: boundary heat transfer coefficient, equivalent restraint stiffness of the fixture, elastic modulus of the material at high temperature, effective power coefficient of the heat source, high strain rate cyclic hardening parameters, and equivalent boundary stiffness coefficient.

[0011] Preferably, step S2 specifically includes: S21. The central processing unit controls the welding torch to perform a trial weld of length L on the initial weld area using preset basic process parameters. The length L ranges from 50 to 200 mm and is not less than 4 times the thickness d of the base material of the high-altitude steel structure to be welded. During the trial weld, the temperature sensor and strain sensor of the multi-source sensing unit synchronously collect the spatiotemporal distribution data of the temperature field in a strip area with a radius of not less than 2d on both sides of the weld centerline as the reference, as well as the real-time strain data of the trial weld area and adjacent monitoring points. The arrangement range of the adjacent monitoring points is no more than 1 / 2 of the distance from the edge of the trial weld area to the nearest structural discontinuity on the main force transmission component of the trial weld area. S22. The central processing unit uses the spatiotemporal distribution data of the temperature field collected by the multi-source sensing unit as the thermal input boundary condition, and loads it into the welding twin basic model for forward thermo-mechanical coupling calculation to obtain the calculated strain field data of the test welding area. The calculated strain field data is compared with the real-time strain data in the whole domain, and based on the comparison deviation, the following physical parameters in the welding twin basic model are dynamically adjusted through the reverse analysis algorithm: boundary heat transfer coefficient, equivalent restraint stiffness of the fixture, and elastic modulus of the material at high temperature, until the matching degree between the calculated and real-time strain data reaches the preset accuracy, thereby generating the dynamic welding twin model that matches the current working condition. S23. Based on the dynamic welding twin model generated in step S22 that matches the current working conditions, within the feasible domain of parameters such as welding current, arc voltage, and welding torch travel speed, an iterative simulation method is used to simulate the welding process of the initial segment of the formal welding, and a set of welding parameters that can make the simulated deformation less than the preset allowable threshold is found as the recommended welding parameters for the first segment of welding in step S3; the simulated deformation includes at least: the lateral shrinkage and angular deformation of the welding local area, and the maximum deflection of the entire component.

[0012] Preferably, step S3, which involves performing welding based on the predicted planning of the current dynamic welding twin model, specifically includes: For the current construction segment, the central processing unit retrieves the current dynamic welding twin model updated from the previous segment for use in the current segment. The central processing unit obtains the bevel geometry of the current segment from the pre-stored welding procedure specifications, the welding spatial position and weld orientation of the current segment from the structural 3D model, and the current ambient temperature read in real-time from the multi-source sensing unit, using these as input conditions to execute a welding process simulation. The simulation output includes the initial combination of recommended welding current, arc voltage, and welding torch travel speed for the current segment, as well as multi-layer, multi-pass welding parameters designed to balance heat input and deformation. The sequence of weld beads / layers and the start and end positions of each weld bead; during welding, the welding torch operates based on the above initial value combination, while the multi-source sensing unit monitors the actual arc power and molten pool morphology in real time. If the detected penetration state exceeds the acceptable range in the welding process specification, or the arc power exceeds its recommended value by ±10%, or the welding torch travel speed exceeds its recommended value by ±10%, the central processing unit will fine-tune the welding speed within a preset correction range. The preset correction range does not exceed ±20% of the planned welding speed to ensure that the actual heat input is consistent with the simulation planning target.

[0013] Preferably, in step S3, based on the transient stress distribution indication of the current dynamic welding twin model, online stress reduction intervention is implemented on the intervention target area of ​​the currently completed weld segment using a portable ultrasonic impact device that synchronously follows the movement of the welding torch. This specifically includes: At the moment the current segment welding operation is completed, the central processing unit immediately drives the current dynamic welding twin model, using the transient temperature distribution data actually collected at the moment the current segment welding operation is completed as the heat input, to perform a transient thermo-mechanical coupling simulation calculation, and obtain the transient calculated strain field data at the moment the current segment welding operation is completed. The central processing unit runs a stress analysis algorithm to perform a grid scan on the obtained transient calculated stress field data, automatically identify all stress concentration points, and cluster the stress concentration points that are spatially close into stress concentration regions as intervention target areas. Based on the spatial distribution of the intervention target area, the central processing unit plans the optimal movement path and operation sequence of the portable ultrasonic shock device with the principle of minimizing the total movement distance. At the same time, based on the highest stress level in each intervention target area, it matches and specifies the corresponding shock process parameters of the portable ultrasonic shock device from a preset parameter database. The shock process parameters include at least the number of single-point shocks and the amplitude of the shock head. Finally, a digital intervention map integrating the coordinates of the intervention target area, the shock path, the operation sequence, and the shock process parameters is generated. The central processing unit sends the digital intervention map to the portable ultrasonic impact device in real time. Based on the digital intervention map, the portable ultrasonic impact device sequentially completes the impact operation on each intervention target area after welding is completed and before the interlayer temperature drops to 100~110℃. During the impact process, the force sensor and accelerometer built into the portable ultrasonic impact device synchronously collect the impact force-displacement response spectrum and vibration spectrum, and transmit them back to the central processing unit for subsequent verification and iterative updates of the current dynamic welding twin model.

[0014] Preferably, step S3, which iteratively updates the current dynamic welding twin model based on the actual thermo-mechanical response data and impact-displacement response data collected throughout the welding and stress reduction intervention processes, and applies this update to the next segmented welding plan, specifically includes: S31. The central processing unit uses the actual temperature field distribution data and measured welding deformation data collected during the current segmented welding process as a benchmark, compares them with the prediction results of the current dynamic welding twin model for the current segmented welding process, calculates the prediction deviation in heat input, temperature field, and transient deformation, and optimizes and adjusts the effective power coefficient of the heat source and the equivalent boundary stiffness coefficient in the dynamic welding twin model based on the above prediction deviation to form a transitional dynamic welding twin model. S32. Using the transitional dynamic welding twin model obtained in step S31 as a benchmark, the impact force-displacement response data collected throughout the stress reduction intervention process are used to perform inversion iterative calculations, and the high strain rate cyclic hardening parameters of the transitional dynamic welding twin model are adjusted to form a new dynamic welding twin model: (1) Set initial values ​​and variation ranges for the high strain rate cyclic hardening parameters; (2) Use the impact force-displacement response data collected during the stress reduction intervention process as the target curve, simulate the impact process in the transition dynamic welding twin model, and calculate a set of simulated impact response spectra; (3) Calculate the characteristic error between the simulated response spectrum and the target curve, wherein the characteristic error includes at least the error of peak force and the error of the area under the load-displacement curve; (4) Use an optimization algorithm to automatically adjust the value of the high strain rate cyclic hardening parameters, and repeat steps (2) to (3) until the sum of the characteristic errors is minimized and the convergence condition is met, and update the high strain rate cyclic hardening parameters obtained by the final optimization to the transition dynamic welding twin model to form a new dynamic welding twin model; S33. Using the new dynamic welding twin model obtained in step S32, re-simulate the welding process of the current segment and verify the consistency between the predicted data and the measured data. When the consistency reaches the preset standard, the iterative update is completed, and the current dynamic welding twin model for the next segment welding planning is generated.

[0015] Preferably, after all segmented welding and iterative updates are completed in step S4, targeted post-processing is performed on the region where the measured residual stress obtained by the multi-source sensing unit exceeds the final prediction result of the current dynamic welding twin model after the iterative update. The specific operations include: S41. After all welding, segmented online stress reduction intervention and current dynamic welding twin model iteration update are completed, a portable residual stress detector is used to measure the residual stress of the high-altitude steel structure and obtain the measured residual stress distribution data. The portable residual stress detector is connected to the central processing unit. S42. The central processing unit performs spatial registration and comparison between the measured residual stress distribution data and the residual stress prediction data finally output by the current dynamic welding twin model after iterative update. It identifies all areas where the measured residual stress values ​​exceed the predicted values ​​at the corresponding positions of the current dynamic welding twin model after iterative update, and the excess is greater than 15% of the material's room temperature yield strength. These areas are then identified as targeted post-processing areas. S43. The central processing unit plans post-processing parameters for each targeted post-processing area based on the measured residual stress magnitude, direction, and distribution characteristics within the targeted post-processing area, and generates targeted post-processing operation instructions: when the processing method is local heat treatment, the operation instructions include heating temperature, heating rate, holding time, and heating range; when the processing method is supplementary ultrasonic impact, the instructions include impact energy, coverage area, and impact path. S44. Process the corresponding post-processing area according to the post-processing operation instruction. After the processing is completed, re-measure the residual stress in the post-processing area. If the re-measurement result is lower than the preset safety threshold or consistent with the model prediction value, the processing is completed; otherwise, return to step S42, adjust the post-processing parameters based on the new re-measurement data and execute again until the requirements are met.

[0016] Preferably, the welded twin basic model is specifically constructed as a transient thermo-mechanical coupled finite element model based on the geometry of the three-dimensional structural model, wherein: The mesh density of the weld and the base material on both sides, where the peak temperature during the welding thermal cycle is between the Ac1 phase transformation point and the melting point of the material, is higher than that of the base material region of the high-altitude steel structure. The material property database in the model pre-stores functions of density, specific heat capacity, thermal conductivity, elastic modulus, and yield strength as a function of temperature. The heat source in the model is mathematically described using a moving double ellipsoid or Gaussian model, and its effective power is defined as a variable that is adjusted by the effective power coefficient of the heat source. The structural boundary constraints of the model are simulated by applying stiffness values ​​at the mounting connection points using spring elements defined by the equivalent restraint stiffness of the clamp and the equivalent boundary stiffness coefficient. The model has a preset ambient temperature loading point, which is used to receive real-time ambient temperature data from the multi-source sensing unit to dynamically update the heat transfer boundary conditions. The boundary heat transfer coefficient, the equivalent restraint stiffness of the clamp, the elastic modulus of the material at high temperature, the effective power coefficient of the heat source, the equivalent boundary stiffness coefficient, and the high strain rate cyclic hardening parameters are all defined in the model as independent input variables connected to the reverse analysis algorithm and optimization algorithm in the central processing unit, so as to support parameter calibration and iterative updates in steps S2 and S3.

[0017] The present invention has at least the following beneficial effects: Firstly, this invention achieves a fundamental leap from relying on static experience to dynamic intelligent control in high-altitude welding construction. Through the constructed closed-loop process of trial welding calibration model - segmented welding and intervention - data iteration update, the traditional discrete and lagging quality control steps are integrated into a continuous intelligent process with a dynamic digital twin model as the decision core. This makes welding deformation pre-control and stress release no longer a passive response after the fact, but an active guidance and synchronous resolution based on real-time data and high-confidence model prediction. As a result, the scientificity, accuracy and reliability of welding construction under complex high-altitude conditions are greatly improved. Secondly, this invention significantly improves the targeting and adaptability of welding process planning and execution. By reverse-calibrating the basic model through trial welding test data, a dynamic model that accurately matches the current specific environment, materials, and restraint conditions is generated. Based on this model, customized welding parameters are planned for each segment. At the same time, fine-tuning is performed during execution based on real-time monitoring of key indicators such as penetration and heat input. This ensures that the welding process can dynamically adapt to the ever-changing environmental interference and structural state changes at high altitudes, effectively overcoming the problem of inaccurate prediction of traditional fixed process parameters and ensuring deformation control from the source. Thirdly, this invention utilizes a dynamic model to calculate and indicate high-stress areas in real time, guiding a portable ultrasonic impact device to immediately intervene after welding, reducing most of the harmful stress on-site before it accumulates and solidifies. At the same time, the real data generated throughout the welding and impact process is fed back to the model, driving its key parameters (heat source efficiency, material dynamic properties, and overall constraints) to continuously iterate and update, forming an enhanced intelligent closed loop of execution generating data, data optimizing the model, and model improving decision-making. This enables the control system to have online self-learning and adaptive capabilities, and the construction accuracy increases with the progress. Fourth, this invention constructs a closed-loop final quality assurance system that covers the entire process and is based on actual measurement and verification. It also provides a specific implementation architecture for a high-fidelity model. Finally, residual stress is measured through independent testing, and the residual deviations predicted by the model are targeted and eliminated based on the measured results, thus forming the ultimate quality control from process control to result verification.

[0018] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the construction method described in one of the technical solutions of the present invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0021] It should be understood that terms such as “having,” “comprising,” and “including” as used herein do not exclude the presence or addition of one or more other elements or combinations thereof.

[0022] like Figure 1 As shown, this invention provides a construction method for pre-controlling welding deformation and releasing stress in high-altitude steel structures of chemical plants, including: S1. After the high-altitude steel structure to be welded is in place, temperature sensors and strain sensors are arranged at multiple monitoring points as multi-source sensing units. S2. Before formal welding, control the welding torch to perform a trial weld on the initial weld area, and synchronously collect the thermal-mechanical response data of the trial weld process through the multi-source sensing unit. Based on the thermal-mechanical response data of the trial weld process, reverse-calibrate the model parameters of the preset welding twin basic model to generate a dynamic welding twin model that matches the current working conditions. S3. Divide the total weld into several construction segments, and perform the following steps sequentially during the welding process of each segment: execute welding based on the prediction and planning of the current dynamic welding twin model; implement online stress reduction intervention on the intervention target area of ​​the completed weld in the current segment using a portable ultrasonic impact device that moves synchronously with the welding torch, based on the transient stress distribution indication of the current dynamic welding twin model; and iteratively update the current dynamic welding twin model based on the actual thermo-mechanical response data and impact force-displacement response data collected during the entire welding and stress reduction intervention process, and apply it to the welding planning of the next segment. S4. After all segmented welding and iterative updates are completed, targeted post-processing is performed on the areas where the measured residual stress obtained by the multi-source sensing unit exceeds the final prediction result of the current dynamic welding twin model after iterative updates. In step S3, the intervention target area refers to the stress concentration area calculated by the current dynamic welding twin model at the instant the current segmented welding operation is completed. The stress concentration area includes the area where the Mises equivalent stress exceeds 70-85% of the yield strength of the material at room temperature at the corresponding calculation point of the current dynamic welding twin model.

[0023] The construction method provided by the above technical solution is achieved through a closed-loop control system integrating real-time sensing, digital simulation, and physical intervention. After the high-altitude steel structure to be welded is in place, commercially available temperature sensors (such as K-type thermocouples or infrared temperature probes) and strain sensors (such as resistance strain gauges or fiber optic grating sensors) are installed at multiple key points on its surface. These sensors together form a multi-source sensing unit to collect thermodynamic data. The multi-source sensing unit is communicatively connected to the system's central processing unit, overcoming the shortcomings of traditional methods that rely on manual experience and lack real-time quantitative data support. Before formal welding, a controllable test weld of length L (usually between 50 and 200 mm) is performed at the starting weld using a general-purpose automated welding torch that is communicatively connected to the central processing unit, simultaneously collecting temperature and strain field data of the test weld area. The collected real data is input into a pre-set welding twin basic model based on the finite element method. The parameters affecting the accuracy of the model (such as material thermal properties) are reverse-calibrated by parameter inversion algorithm (such as optimization algorithm). This generates a current dynamic welding twin model that is highly consistent with the current wind speed, ambient temperature and actual constraint conditions. This fundamentally solves the core problem of traditional numerical simulation, which leads to model distortion and the inability to directly know the prediction results of complex high-altitude on-site construction due to the use of idealized standard parameters and boundary conditions.

[0024] The long weld is divided into multiple construction segments (each segment's length can be set according to the situation, for example, 300~800mm). Within the construction cycle of each segment, the system first calls the current segment's dynamic model to predict the optimal welding parameters (current, voltage, speed) and controls the welding torch to execute. As soon as welding is completed, the model immediately uses the newly collected transient temperature data to simulate the stress field of that weld segment, and automatically identifies high-stress intervention target areas based on the quantification threshold that the Mises equivalent stress exceeds 70%~85% of the material's yield strength. Then, the portable ultrasonic impact device (a stress relief device already available on the market) is precisely moved to these intervention target areas for immediate impact treatment, realizing stress relief as it is generated. This effectively solves the dilemma of continuous accumulation of welding stress in traditional processes, where post-treatment can only be carried out after all welding is completed, and the effect is limited.

[0025] Simultaneously, new data generated throughout the entire welding and impact treatment process is collected and used to iteratively update the current dynamic welding twin model, continuously optimizing it and providing more accurate predictions for the next welding segment. This endows the system with adaptive and self-learning capabilities to cope with dynamic changes in the high-altitude environment. After all welding is completed, a portable residual stress detector (such as a blind hole stress meter) is used for final measurement. The measured results are compared with the model's final predictions. Only local areas where the measured values ​​are significantly higher than the predicted values ​​are subject to final, targeted post-processing (such as local heating). This abandons the traditional mode of blindly and inefficiently post-processing the entire weld or a large area, greatly improving the accuracy and efficiency of the final defect elimination operation and reducing the safety risks of secondary high-altitude operations.

[0026] Based on the above technical solution, a specific workflow of the construction method for pre-controlling welding deformation and releasing stress in high-altitude steel structures of chemical plants provided by the present invention is as follows: A sensor network (temperature and strain sensors) is installed at selected monitoring locations on the high-altitude steel structure (such as both sides of the weld centerline and near the restraint points), and system debugging is completed. The first step is trial welding and model generation: a trial weld of approximately 100mm (not less than 4 times the thickness d of the base material of the high-altitude steel structure to be welded) is performed at the starting end. The central processing unit simultaneously records all sensor data and completes the calibration of the welding twin basic model within a few minutes, outputting a current dynamic welding twin model adapted to the site environment. The second step is cyclic segmented construction: taking a 10m long weld as an example, it is divided into 20 segments. Starting from the first segment, the central processing unit reads the optimal welding parameters for that segment from the current dynamic welding twin model and drives the welding torch operation. After the torch is extinguished, the system calculates the stress cloud map of the weld segment within seconds, identifies local points where the Mises stress reaches 80% of the yield strength (i.e., the intervention target area), and then plans a path to guide the adjacent ultrasonic impact device to immediately impact these points. After the impact is completed, all temperature, deformation, and impact force data for this segment are packaged for rapid optimization of model parameters. This welding-calculation-impact-learning cycle is repeated in the second, third, and final segments, and the current dynamic welding twin model becomes increasingly accurate as construction progresses. The final step is acceptance and targeted reinforcement: After the entire weld seam is welded and segmented impact is completed, a portable stress meter (connected to the central processing unit) is used to perform scanning measurements on the weld seam. The central processing unit overlays and compares the measured stress spectrum with the final stress spectrum predicted by the model, automatically identifying a few areas where the residual stress exceeds the standard (for example, points where the measured value is 15% higher than the model prediction value of the yield strength). Then, only the points exceeding the standard are subjected to local heat treatment until the retest is qualified.

[0027] The aforementioned technical solution constructs a real-time intelligent closed loop of perception-simulation-intervention-learning, fundamentally changing the passive mode of high-altitude welding deformation and stress control. By calibrating the current dynamic welding twin model in real time using on-site trial welding data, a high degree of consistency between the prediction model and complex working conditions is ensured. Through the instantaneous linkage between welding and stress reduction during segmented construction, stress is eliminated as it is produced, effectively preventing cumulative deformation and residual stress concentration. By continuously iterating and updating the current dynamic welding twin model using new data from each segment, the control accuracy of the entire system possesses self-learning and adaptive capabilities, becoming increasingly accurate with each weld. Ultimately, the construction method provided by this invention significantly improves the deformation control accuracy, structural safety, and first-pass yield of high-altitude steel structure welding without requiring large post-processing facilities, while reducing the safety risks and overall costs of high-altitude straightening operations.

[0028] In one of the technical solutions, the stress concentration region further includes: At geometric discontinuities, the rate of change of principal stress per unit length exceeds 2.0 to 3.5 times the average stress per unit length of the cross section at the corresponding calculation point in the current dynamic welding twin model; The region where the transverse tensile stress component perpendicular to the weld direction exceeds 30-40% of the room temperature yield strength of the material at the corresponding calculation point in the current dynamic welding twin model.

[0029] The aforementioned technical solution further supplements the quantitative judgment criteria for stress gradient and transverse tensile stress, expanding the identification of stress concentration areas from a single strength index to multi-dimensional mechanical characteristics. This significantly improves the comprehensiveness and accuracy of high-stress point identification, effectively preventing the risk of crack initiation or increased deformation due to missed detection at geometric abrupt changes or tensile stress-dominant areas. For example, at the weld toe of a high-altitude steel structure beam welded to a stiffening rib, the dynamic welding twin model calculates that the average stress of the cross-section in this area is 100 MPa. Within a very small range at the weld toe root (e.g., within a distance of 1 mm), the principal stress increases sharply from 100 MPa to 350 MPa. The stress gradient is 250 MPa / mm. This gradient value is 2.5 times the average stress per unit length (100 MPa / mm), exceeding the judgment threshold of 2.0 times per unit length. Therefore, the weld toe root is automatically identified by the system as a high stress gradient concentration area and marked as an intervention target area.

[0030] In one technical solution, the multi-source sensing unit, welding torch, and portable ultrasonic impact device are all communicatively connected to the central processing unit; the multi-source sensing unit also includes an ambient temperature sensor that monitors the ambient temperature of the welding operation area in real time; the central processing unit is pre-set with the welding process specification of the high-altitude steel structure to be welded, the three-dimensional structural model of the high-altitude steel structure to be welded, and the welding twin basic model, the welding twin basic model including at least the following physical parameters: boundary heat transfer coefficient, equivalent restraint stiffness of the fixture, elastic modulus of the material at high temperature, effective power coefficient of the heat source, high strain rate cyclic hardening parameters, and equivalent boundary stiffness coefficient.

[0031] In the above technical solutions, the boundary heat transfer coefficient is a parameter characterizing the heat exchange capacity between the welded structure surface and the surrounding environment (air). It integrates convection and radiation effects and directly affects the cooling rate of the welded area in the simulation. The equivalent restraint stiffness of the fixture is used to quantify the mechanical parameters of the actual installation fixture's constraint on the thermal deformation of the welded component, reflecting the fixture's ability to resist component deformation. The elastic modulus of the material at high temperature is a physical quantity describing the ease with which the material undergoes elastic deformation under high-temperature welding conditions. Its value decreases significantly with increasing temperature and is crucial for calculating welding thermal stress. The effective power coefficient of the heat source is used to correct the coefficients of the ideal heat source model (such as a double ellipsoid), reflecting the proportion of actual welding arc converted into effective heat energy acting on the workpiece, and determining the accuracy of the input heat in the simulation. The high strain rate cyclic hardening parameter describes the characteristics of the material's yield strength improvement (hardening) behavior under high-rate, cyclic plastic deformation conditions induced by ultrasonic impact treatment, affecting the prediction of impact stress relaxation effects. The equivalent boundary stiffness coefficient is used in the numerical model to comprehensively characterize the changes in the constraint state of the entire welded structure (not just the fixture) caused by cumulative deformation and redistribution of internal forces, reflecting the evolution of the structure's own stiffness. The above technical solution constructs an integrated intelligent control system with a central processing unit at its core, interconnecting sensing, decision-making, and execution devices. It also pre-configures a digital foundation model containing key physical parameters, enabling the welding twin foundation model and the updated dynamic welding twin model to systematically integrate welding process specifications, three-dimensional geometry, and dynamic environmental information. This provides a unified digital foundation for subsequent accurate simulation, parametric calibration, and iterative optimization, thereby ensuring the stable, collaborative, and efficient operation of the entire intelligent construction method at the system level.

[0032] In one of the technical solutions, step S2 specifically includes: S21. The central processing unit controls the welding torch to perform a trial weld of length L on the initial weld area using preset basic process parameters. The length L ranges from 50 to 200 mm and is not less than 4 times the thickness d of the base material of the high-altitude steel structure to be welded. During the trial weld, the temperature sensor and strain sensor of the multi-source sensing unit synchronously collect the spatiotemporal distribution data of the temperature field in a strip area with a radius of not less than 2d on both sides of the weld centerline as the reference, as well as the real-time strain data of the trial weld area and adjacent monitoring points. The arrangement range of the adjacent monitoring points is no more than 1 / 2 of the distance from the edge of the trial weld area to the nearest structural discontinuity on the main force transmission component of the trial weld area. S22. The central processing unit uses the spatiotemporal distribution data of the temperature field collected by the multi-source sensing unit as the thermal input boundary condition, and loads it into the welding twin basic model for forward thermo-mechanical coupling calculation to obtain the calculated strain field data of the test welding area. The calculated strain field data is compared with the real-time strain data in the whole domain, and based on the comparison deviation, the following physical parameters in the welding twin basic model are dynamically adjusted through the reverse analysis algorithm: boundary heat transfer coefficient, equivalent restraint stiffness of the fixture, and elastic modulus of the material at high temperature, until the matching degree between the calculated and real-time strain data reaches the preset accuracy, thereby generating the dynamic welding twin model that matches the current working condition. S23. Based on the dynamic welding twin model generated in step S22 that matches the current working conditions, within the feasible domain of parameters such as welding current, arc voltage, and welding torch travel speed, an iterative simulation method is used to simulate the welding process of the initial segment of the formal welding, and a set of welding parameters that can make the simulated deformation less than the preset allowable threshold is found as the recommended welding parameters for the first segment of welding in step S3; the simulated deformation includes at least: the lateral shrinkage and angular deformation of the welding local area, and the maximum deflection of the entire component.

[0033] This invention further optimizes the trial welding and model generation process, solving the problem in traditional methods where welding simulation relies on standard material libraries and ideal boundary conditions. These parameters are severely out of sync with variable wind speeds, sunlight, and actual installation constraints at high altitudes, leading to model distortion and making accurate predictions impossible. This invention obtains realistic environmental and structural response data through short weld trial welding, thereby transforming a pre-set welding twin basic model into a frozen welding twin model highly matched to the field conditions. Specifically, the above technical solution executes a trial welding segment of 50-200mm in length and at least four times the plate thickness, simultaneously collecting complete temperature field and strain data of a strip-shaped area near the weld, providing a realistic data anchor point to solve the problem of inaccurate initial setups in the pre-set welding twin basic model. The system uses the measured temperature field as input to drive the welding twin basic model for forward calculation, and compares the calculation results with the measured strain. It uses a reverse analysis algorithm to dynamically adjust the boundary heat transfer coefficient, the equivalent restraint stiffness of the fixture, and the high-temperature elastic modulus of the material until the model output matches the response of the physical world, thereby generating a dynamic welding twin model rooted in the current working condition.

[0034] The establishment of a high-confidence current dynamic welding twin model directly serves the process formulation for the first stage of formal welding. Traditional process planning relies on experience or general procedures, which cannot adapt to the subtle differences in specific structures. This scheme utilizes the calibrated current dynamic welding twin model to conduct rapid iterative simulations within the feasible domain of parameters such as welding current, voltage, and speed, simulating the entire welding and cooling process under different parameter combinations. The simulation objective is clear: to find a set of parameter combinations that ensures that the local lateral shrinkage, angular deformation, and overall component deflection are all less than preset allowable thresholds (referencing GB50661-2011 and GB55006-2021). Through automatic optimization, the optimal welding parameters are finally selected as precise guidance for the first stage of construction, thereby overcoming the risk of accumulated overall deviations due to improper processes in the first stage of welding.

[0035] The main steps for dynamically adjusting the boundary heat transfer coefficient, equivalent restraint stiffness of the fixture, and high-temperature elastic modulus of the material using the inverse analysis algorithm are as follows: First, the system establishes optimization objectives and initial parameters. The algorithm sets the real-time strain data collected during the trial welding process as the target true value to be matched, and sets the boundary heat transfer coefficient, equivalent restraint stiffness of the fixture, and high-temperature elastic modulus of the material to be calibrated in the welding twin basic model as freely adjustable optimization variables. At the same time, it assigns reasonable initial values ​​and physically possible value ranges to these variables based on engineering experience or material handbooks.

[0036] Next, in each iteration, the algorithm executes a sub-loop of forward calculation, comparison and evaluation, and parameter adjustment. Specifically, it first uses the collected spatiotemporal distribution data of the temperature field as a defined thermal input, combined with the values ​​of a set of undetermined parameters in the current iteration, to drive the welding twin basic model to perform a complete forward thermo-mechanical coupling simulation, outputting the calculated strain field data under that set of parameters. Then, the algorithm quantitatively compares the strain field obtained from this calculation with the measured strain field as a benchmark across the entire domain, calculating the overall deviation between the two (often using quantitative indicators such as root mean square error). This deviation value directly measures the degree of agreement between the simulation results and physical reality under the current model parameters.

[0037] Finally, the reverse analysis algorithm (such as gradient descent-based optimization algorithms or genetic algorithms) intelligently generates a set of correction values ​​for the three parameters to be adjusted based on the calculated deviation values ​​and their changing trends, thereby updating the values ​​of these parameters. The system then determines whether the updated parameters have achieved a preset accuracy requirement for the matching degree between the calculated strain and the measured strain. If not, it automatically enters the next iteration, repeating the above forward calculation and comparative evaluation process; if the requirement has been met, the iteration terminates, and the model parameters (boundary heat transfer coefficient, equivalent restraint stiffness of the fixture, and high-temperature elastic modulus of the material) are determined as the optimal solution matching the current working condition, thus generating the dynamic welding twin model.

[0038] The above technical solution utilizes a low-cost, short-cycle on-site micro-test to efficiently complete the one-time accurate calibration of a complex high-altitude welding system. It transforms a general digital model into a customized prediction engine specifically for the current task and provides reliable and optimal initial process parameters for the construction of the entire weld seam. This minimizes the risk of cumulative deformation caused by inaccurate models and improper initial processes, and realizes a key transformation in welding construction from experience-based trial and error to data-driven approaches.

[0039] In one of the technical solutions, step S3, which involves performing welding based on the predictive planning of the current dynamic welding twin model, specifically includes: For the current construction segment, the central processing unit retrieves the current dynamic welding twin model updated from the previous segment for use in the current segment. The central processing unit obtains the bevel geometry of the current segment from the pre-stored welding procedure specifications, the welding spatial position and weld orientation of the current segment from the structural 3D model, and the current ambient temperature read in real-time from the multi-source sensing unit, using these as input conditions to execute a welding process simulation. The simulation output includes the initial combination of recommended welding current, arc voltage, and welding torch travel speed for the current segment, as well as multi-layer, multi-pass welding parameters designed to balance heat input and deformation. The sequence of weld beads / layers and the start and end positions of each weld bead; during welding, the welding torch operates based on the above initial value combination, while the multi-source sensing unit monitors the actual arc power and molten pool morphology in real time. If the detected penetration state exceeds the acceptable range in the welding process specification, or the arc power exceeds its recommended value by ±10%, or the welding torch travel speed exceeds its recommended value by ±10%, the central processing unit will fine-tune the welding speed within a preset correction range. The preset correction range does not exceed ±20% of the planned welding speed to ensure that the actual heat input is consistent with the simulation planning target.

[0040] In traditional methods, once process parameters are set, they are difficult to dynamically adjust during construction and cannot respond to real-time fluctuations. This invention further addresses the problem of inaccurate process parameters in high-altitude segmented welding caused by environmental interference, component posture changes, or the cumulative effects of previous welding processes by proposing a closed-loop control process of predictive planning, real-time monitoring, and dynamic fine-tuning. For the current construction segment, the central processing unit retrieves a dynamically updated welding twin model that incorporates real data from the previous segment and more accurately reflects the current state of the structure. The system then obtains the bevel dimensions and spatial posture data of the current segment from pre-stored procedures and models, and reads the ambient temperature in real time. Using these specific conditions as input, the system drives the current dynamically welded twin model to perform a rapid welding process simulation, thereby customizing a set of recommended initial values ​​for welding current, voltage, and travel speed, as well as a detailed weld sequence plan for the current segment.

[0041] During the welding execution phase, this solution further addresses the core issue of actual heat input deviating from planned values ​​due to arc instability, subtle differences in operator technique, or external disturbances. While the welding torch begins operation based on the aforementioned customized initial value combination, a multi-source sensing unit monitors in real time, continuously collecting key process signals such as actual arc power and weld pool morphology. The system continuously compares these real-time signals with the preset acceptable range in the welding process specification and the ideal values ​​recommended by the simulation. If an anomaly is detected in the signal representing the penetration state, or if the actual values ​​of the two key parameters—arc power and welding torch travel speed—deviate from their recommended values ​​by more than ±10%, the central processing unit determines that a deviation requiring correction has occurred. Within a preset correction range not exceeding ±20% of the planned speed, the welding speed is dynamically fine-tuned. For example, if the actual power is too low, the welding speed is appropriately reduced to increase heat input; conversely, the welding speed is increased. Through rapid feedback adjustment, the welding process is forcibly pulled back to the ideal trajectory of the simulation plan, ensuring that the actual heat input acting on the workpiece remains highly consistent with the model's simulation planning target.

[0042] The above technical solution describes the sequence of weld beads / layers for multi-layer, multi-pass welding, designed to balance heat input and deformation, as well as the start and end positions of each weld bead. The central processing unit executes an automated digital planning and optimization process based on the current dynamic welding twin model. The specific steps are as follows: Step 1: Defining Optimization Objectives and Extracting Planning Constraints: The central processing unit first clarifies the core planning objective, which is to minimize the instantaneous heat input distribution during welding by arranging the weld sequence and position, ultimately minimizing welding deformation (lateral shrinkage, angular deformation, and deflection). Subsequently, the system extracts the precise geometric dimensions (such as width, depth, and angle) of the current bevel layer from pre-stored welding process specifications, extracts the specific spatial location and weld orientation (horizontal, horizontal, vertical, and overhead) of this segment from the 3D structural model, and obtains the latest material thermophysical parameters and boundary constraints from the current dynamic welding twin model. This information collectively constitutes the input conditions and physical constraints for planning.

[0043] Step 2: Generating candidate weld bead schemes in the current dynamic welding twin model: Based on the bevel geometry, the system discretizes the bevel section to be filled into a mesh in the thickness and width directions within the dynamic welding twin model. Following the basic principles of multi-layer, multi-pass welding (such as welding the middle first, then the sides, lower weld beads providing support for upper layers, and staggering adjacent weld beads), the algorithm automatically generates multiple initial weld bead filling sequences and spatial placement schemes that conform to metallurgical and operational feasibility. Each scheme clearly defines the precise coordinates of the start and end positions of each weld bead in the 3D model.

[0044] Step 3: Iterative Simulation and Optimization Selection: The central processing unit drives the current dynamic welding twin model to perform rapid welding process simulation for each candidate weld sequence and position scheme. The simulation sequentially simulates the welding of each weld seam, calculating the accumulated thermal cycle, transient temperature field, and final cooling deformation. The model outputs the predicted deformation (lateral shrinkage, angular deformation, deflection) and heat input concentration index for each scheme. The system compares these results with preset allowable thresholds and evaluates and ranks all candidate schemes using optimization algorithms (such as those based on minimizing deformation) until the optimal weld sequence and start / end position scheme that ensures all predicted deformations are less than the threshold and the heat input distribution is relatively balanced is selected.

[0045] Step 4: Outputting Executable Welding Operation Instructions: Finally, this optimal weld bead planning scheme is transformed into specific, executable operation instructions. These instructions integrate the number of each weld bead, the precise three-dimensional coordinates of its start and end points, the welding direction (e.g., from the middle to both ends), and the interval between each weld bead and the previous one (used to control interpass temperature). These digital instructions, along with welding current, voltage, and speed parameters, constitute the complete welding process package for the current segment, and are sent to the welding torch execution system to guide it in completing high-precision, low-deformation automated welding operations.

[0046] The above technical solution constructs a real-time sensing and fast-response adaptive control closed loop at the welding execution level of each segment. By deeply integrating customized model prediction with online process monitoring and dynamic parameter fine-tuning, it ensures that the actual welding heat input in complex high-altitude environments can highly accurately match the simulation planning target. Thus, from the fundamental link of process execution, it provides a stable and reliable foundation for the entire welding deformation pre-control system.

[0047] In one technical solution, step S3, based on the transient stress distribution indication of the current dynamic welding twin model, uses a portable ultrasonic impact device that synchronously follows the movement of the welding torch to implement online stress reduction intervention on the intervention target area of ​​the currently completed weld segment. Specifically, this includes: At the moment the current segment welding operation is completed, the central processing unit immediately drives the current dynamic welding twin model, and uses the transient temperature distribution data actually collected at the moment the current segment welding operation is completed as the heat input to perform a transient thermo-mechanical coupling simulation calculation to obtain the transient calculated strain field data at the moment the current segment welding operation is completed. The central processing unit runs a stress analysis algorithm to perform a grid scan on the obtained transient calculated stress field data, automatically identify all stress concentration points, and cluster the stress concentration points that are spatially close into stress concentration regions as intervention target areas. Based on the spatial distribution of the intervention target area, the central processing unit plans the optimal movement path and operation sequence of the portable ultrasonic shock device with the principle of minimizing the total movement distance. At the same time, based on the highest stress level in each intervention target area, it matches and specifies the corresponding shock process parameters of the portable ultrasonic shock device from a preset parameter database. The shock process parameters include at least the number of single-point shocks and the amplitude of the shock head. Finally, a digital intervention map integrating the coordinates of the intervention target area, the shock path, the operation sequence, and the shock process parameters is generated. The central processing unit sends the digital intervention map to the portable ultrasonic impact device in real time. Based on the digital intervention map, the portable ultrasonic impact device sequentially completes the impact operation on each intervention target area after welding is completed and before the interlayer temperature drops to 100~110℃. During the impact process, the force sensor and accelerometer built into the portable ultrasonic impact device synchronously collect the impact force-displacement response spectrum and vibration spectrum, and transmit them back to the central processing unit for subsequent verification and iterative updates of the current dynamic welding twin model.

[0048] The aforementioned technical solution constructs a highly intelligent, precise, and closed-loop online stress reduction intervention system, fundamentally changing the traditional mode of stress release that is lagging, vague, and reliant on human experience. This solution, by immediately using the latest acquired temperature data to drive a dynamic model for millisecond-level transient stress simulation upon completion of the current segment welding, achieves real-time visualization of the stress field and automatically and accurately identifies multi-dimensional hazardous areas, including high stress, high gradient, and high tensile stress, as intervention target areas, thus eliminating the risk of inaccurate or missed judgments by humans. The system then automatically plans the optimal operating path and matching impact parameters for the impact device, generating a directly executable digital intervention map, transforming complex mechanical diagnosis into precise navigation commands for automated equipment, ensuring the timeliness (completed within the 100~110℃ interpass temperature window) and scientific rigor of the intervention. Crucially, the dynamic response data collected synchronously during the impact process is fed back for model verification and iteration, making the stress reduction process itself a driving force for model evolution and optimization of subsequent decisions. This forms an enhanced closed loop of real-time diagnosis, precise execution, and feedback learning, thereby significantly improving the immediacy, accuracy, and overall adaptive optimization capability of high-altitude welding stress control.

[0049] In one technical solution, step S3 involves iteratively updating the current dynamic welding twin model based on the actual thermo-mechanical response data and impact-displacement response data collected throughout the welding and stress reduction intervention processes, and then applying this update to the next segmented welding plan. Specifically, this includes: S31. The central processing unit uses the actual temperature field distribution data and measured welding deformation data collected during the current segmented welding process as a benchmark, compares them with the prediction results of the current dynamic welding twin model for the current segmented welding process, calculates the prediction deviation in heat input, temperature field, and transient deformation, and optimizes and adjusts the effective power coefficient of the heat source and the equivalent boundary stiffness coefficient in the dynamic welding twin model based on the above prediction deviation to form a transitional dynamic welding twin model. S32. Using the transitional dynamic welding twin model obtained in step S31 as a benchmark, the impact force-displacement response data collected throughout the stress reduction intervention process are used to perform inversion iterative calculations, and the high strain rate cyclic hardening parameters of the transitional dynamic welding twin model are adjusted to form a new dynamic welding twin model: (1) Set initial values ​​and variation ranges for the high strain rate cyclic hardening parameters; (2) Use the impact force-displacement response data collected during the stress reduction intervention process as the target curve, simulate the impact process in the transition dynamic welding twin model, and calculate a set of simulated impact response spectra; (3) Calculate the characteristic error between the simulated response spectrum and the target curve, wherein the characteristic error includes at least the error of peak force and the error of the area under the load-displacement curve; (4) Use an optimization algorithm to automatically adjust the value of the high strain rate cyclic hardening parameters, and repeat steps (2) to (3) until the sum of the characteristic errors is minimized and the convergence condition is met, and update the high strain rate cyclic hardening parameters obtained by the final optimization to the transition dynamic welding twin model to form a new dynamic welding twin model; S33. Using the new dynamic welding twin model obtained in step S32, re-simulate the welding process of the current segment and verify the consistency between the predicted data and the measured data. When the consistency reaches the preset standard, the iterative update is completed, and the current dynamic welding twin model for the next segment welding planning is generated.

[0050] The aforementioned technical solution completely solves the fundamental problem of model drift and prediction inaccuracies that inevitably occur in static simulation models during long-process, variable-condition high-altitude welding due to environmental cumulative effects and changes in material behavior. This solution deeply integrates the real thermodynamic data generated during the current segmented welding process with the dynamic response spectrum unique to stress reduction intervention, forming a complete welding excitation-impact response dataset. This dataset is then used as a standard for dual calibration of the current dynamic welding twin model. First, the welding data is used to optimize and adjust the effective power coefficient of the heat source, which directly affects the accuracy of the thermal process simulation, and the equivalent boundary stiffness coefficient, which reflects the evolution of the overall structural constraint state, enabling the model to track macroscopic state changes during construction. Then, it creatively utilizes impact force-displacement response data and a dedicated inversion iterative algorithm to calibrate the material high-strain rate cyclic hardening parameters, which are rarely involved in traditional models but are crucial. This gives the dynamic welding twin model the ability to characterize the special dynamic mechanical process of ultrasonic impact. The phased, focused parameter updates provided by this invention allow the current dynamic welding twin model to absorb the latest field knowledge after each segment, completing an iterative evolution. Ultimately, the validated and more realistic new model was immediately applied to the planning and decision-making of the next segment, thus forming a self-reinforcing intelligent closed loop in which execution generates data, data optimizes the model, and a better model guides the next round of more precise execution. This enables the prediction and decision-making capabilities of the entire welding deformation and stress control system to continuously improve as construction progresses, achieving a fundamental leap from static prediction to dynamic evolution.

[0051] In one technical solution, after all segmented welding and iterative updates are completed, step S4 involves targeted post-processing of the region where the measured residual stress obtained by the multi-source sensing unit exceeds the final prediction result of the current dynamic welding twin model after the iterative update. The specific operations include: S41. After all welding, segmented online stress reduction intervention and current dynamic welding twin model iteration update are completed, a portable residual stress detector is used to measure the residual stress of the high-altitude steel structure and obtain the measured residual stress distribution data. The portable residual stress detector is connected to the central processing unit. S42. The central processing unit performs spatial registration and comparison between the measured residual stress distribution data and the residual stress prediction data finally output by the current dynamic welding twin model after iterative update. It identifies all areas where the measured residual stress values ​​exceed the predicted values ​​at the corresponding positions of the current dynamic welding twin model after iterative update, and the excess is greater than 15% of the material's room temperature yield strength. These areas are then identified as targeted post-processing areas. S43. The central processing unit plans post-processing parameters for each targeted post-processing area based on the measured residual stress magnitude, direction, and distribution characteristics within the targeted post-processing area, and generates targeted post-processing operation instructions: when the processing method is local heat treatment, the operation instructions include heating temperature, heating rate, holding time, and heating range; when the processing method is supplementary ultrasonic impact, the instructions include impact energy, coverage area, and impact path. S44. Process the corresponding post-processing area according to the post-processing operation instruction. After the processing is completed, re-measure the residual stress in the post-processing area. If the re-measurement result is lower than the preset safety threshold or consistent with the model prediction value, the processing is completed; otherwise, return to step S42, adjust the post-processing parameters based on the new re-measurement data and execute again until the requirements are met.

[0052] The aforementioned technical solution constructs a targeted closed-loop defect elimination mechanism based on final measured data as the authoritative basis and high-confidence model prediction as the scientific benchmark, achieving ultimate quality assurance for welding stress control. This solution, by conducting independent residual stress measurements after the completion of each process and intelligently comparing the measured values ​​with the model's final predicted values, precisely identifies only areas where the measured values ​​significantly exceed the model's predictions, thus completely abandoning the traditional, blind, and inefficient post-processing method for the entire weld. The system automatically plans and executes targeted treatment based on the stress characteristics of the exceeding areas, and immediately re-tests and verifies the treatment, forming a closed loop of treatment-verification-re-treatment until the stress meets the standard. This not only greatly improves the accuracy and efficiency of the final defect elimination operation and significantly reduces the workload and safety risks of secondary high-altitude operations, but also ensures that the residual stress state of the structure can be reliably controlled within a safe or expected range, providing a solid and reliable quality endpoint for the entire intelligent welding construction method.

[0053] In one of the technical solutions, the welded twin basic model is specifically constructed as a transient thermo-mechanical coupled finite element model based on the geometric foundation of the three-dimensional model of the structure, wherein: The mesh density of the weld and the base material on both sides, where the peak temperature during the welding thermal cycle is between the Ac1 phase transformation point and the melting point of the material, is higher than that of the base material region of the high-altitude steel structure. The material property database in the model pre-stores functions of density, specific heat capacity, thermal conductivity, elastic modulus, and yield strength as a function of temperature. The heat source in the model is mathematically described using a moving double ellipsoid or Gaussian model, and its effective power is defined as a variable that is adjusted by the effective power coefficient of the heat source. The structural boundary constraints of the model are simulated by applying stiffness values ​​at the mounting connection points using spring elements defined by the equivalent restraint stiffness of the clamp and the equivalent boundary stiffness coefficient. The model has a preset ambient temperature loading point, which is used to receive real-time ambient temperature data from the multi-source sensing unit to dynamically update the heat transfer boundary conditions. The boundary heat transfer coefficient, the equivalent restraint stiffness of the clamp, the elastic modulus of the material at high temperature, the effective power coefficient of the heat source, the equivalent boundary stiffness coefficient, and the high strain rate cyclic hardening parameters are all defined in the model as independent input variables connected to the reverse analysis algorithm and optimization algorithm in the central processing unit, so as to support parameter calibration and iterative updates in steps S2 and S3.

[0054] The aforementioned technical solution explicitly constructs a finite element model with transient thermo-mechanical coupling simulation capabilities, based on a three-dimensional structural model. Through a series of specific technical means, such as mesh refinement of the weld heat-affected zone, pre-setting a temperature-related material property database, defining a parameterized moving heat source model, and simulating adjustable boundary constraints using spring elements, the abstract concept of a digital twin is grounded in a dedicated computational tool that can be implemented in engineering. This concrete construction makes all key physical parameters in the model (such as the effective power coefficient of the heat source and the equivalent boundary stiffness coefficient) adjustable variables connected to the optimization algorithm. This fundamentally guarantees the accurate and efficient execution of reverse calibration and online iterative updates of the current dynamic welding twin model, providing a unique and reliable digital simulation foundation for the entire closed-loop process from perception and prediction to control and optimization.

[0055] This invention uses the butt welding construction of a steel structure crossbeam in a high-altitude pipe gallery at a chemical plant as an example to further illustrate the construction method of this invention. In this chemical plant renovation and expansion project, a symmetrical high-altitude steel pipe gallery needs to be installed on-site. One span of the main beam is a welded box-section made of Q345B steel, with cross-sectional dimensions of 800mm × 400mm × 20mm (height × width × plate thickness). The weld is a full-penetration groove butt weld, welded horizontally (in a horizontal position), with a total weld length of 12m and a height of approximately 15m above the ground. This structure represents a typical high-altitude, large-size, and highly constrained welding scenario, with strict requirements for deformation control and residual stress levels.

[0056] Preliminary preparation and trial welding calibration: 24 K-type thermocouples and 16 fiber optic strain sensors were arranged in the beam joint area to form a multi-source sensing network. A trial weld of 120mm in length (6 times the plate thickness of 20mm) was performed at the starting end using CO2 gas shielded welding. Basic parameters: current 280A, voltage 30V, welding speed 0.4m / min. Trial welding data acquisition: highest temperature: 1580℃ (near the weld); temperature rise at 40mm from the weld center: 620℃; maximum instantaneous tensile strain recorded by the strain sensors: 0.18%. The central processing unit adjusted the boundary heat transfer coefficient in the welding twin basic model through reverse calibration (from the standard value of 25 W / (m²)). 2·K) corrected to field value 18 W / (m 2 ·K)), equivalent restraint stiffness of the clamp (from the theoretical value of 5×10 5 The N / mm value has been corrected to the measured equivalent value of 3.2 × 10⁻⁶. 5 (N / mm), generating the current dynamic welding twin model.

[0057] Segmented welding and online stress reduction: The 12m weld was divided into 24 segments, each 500mm long; the welding parameters for the first segment were optimized and recommended by the model: current 260A, voltage 28V, welding speed 0.45m / min; within 10 seconds after welding, the model identified 3 intervention target areas based on real-time temperature field simulation, with a Mises equivalent stress of 245MPa (approximately 80% of the room temperature yield strength of Q345B); a portable ultrasonic impact device impacted the target area according to the planned path, with impact parameters: amplitude 25μm, 15 single-point impacts; after impact, the measured surface stress of the target area decreased by approximately 35%.

[0058] Model Iterative Updates: After each welding and impact segment is completed, the system collects actual heat input, deformation data, and impact response spectrum to update the model parameters; by the 6th segment, the average error between the model's predicted lateral shrinkage and the measured value has decreased from the initial 18% to less than 5%.

[0059] Post-weld targeted treatment: After all welding was completed, a portable X-ray stress analyzer was used to scan the residual stress in the whole field. The actual measurement found that the residual stress in two areas exceeded the model prediction value, with the excess reaching 20% ​​of the yield strength. The system planned local heat treatment: heating to 300℃ and holding for 20 minutes. After treatment, the stress was measured again and reduced to within the safe range (below 50% of the yield strength).

[0060] The same location of the symmetrical high-altitude steel pipe gallery was welded using the traditional welding method (experience-based adjustment + post-weld treatment). The results of this embodiment compared with the traditional welding method are shown in Table 1.

[0061] Table 1 Welding index results As shown in Table 1, by using a real-time calibration model during trial welding and combining it with segmented dynamic optimization parameters, both angular deformation and lateral shrinkage were reduced by more than 60%, achieving high-precision installation of large-size steel structures at high altitudes and significantly improving the accuracy of welding deformation control. Online ultrasonic impact intervention intervenes at the initial stage of stress generation, reducing the peak residual stress by more than 30% and the stress concentration area by more than 70%, significantly improving the structural fatigue life and resistance to stress corrosion, and drastically reducing the residual stress level. This invention also avoids multiple corrections and rework in traditional methods, shortening the total construction period by 30%, reducing the number of high-altitude operations, lowering safety risks and overall costs, and significantly improving construction efficiency. From trial welding calibration and segmented control to post-weld targeted treatment, this invention forms a data-driven intelligent closed loop throughout the entire process, achieving a first-pass yield rate of over 98% and significantly enhancing quality controllability. Furthermore, the model of this invention is continuously iterated and updated during construction, with the prediction error gradually decreasing from the initial 18% to within 5%, demonstrating the system's good environmental adaptability and process self-optimization capabilities.

[0062] The number of devices and processing scale described herein are for simplification of the invention. Applications, modifications, and variations of the present invention's method for pre-controlling welding deformation and stress release in high-altitude steel structures of chemical plants will be readily apparent to those skilled in the art.

[0063] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A construction method for pre-controlling welding deformation and releasing stress in high-altitude steel structures of chemical plants, characterized in that, include: S1. After the high-altitude steel structure to be welded is in place, temperature sensors and strain sensors are arranged at multiple monitoring points as multi-source sensing units. S2. Before formal welding, control the welding torch to perform a trial weld on the initial weld area, and synchronously collect the thermal-mechanical response data of the trial weld process through the multi-source sensing unit. Based on the thermal-mechanical response data of the trial weld process, reverse-calibrate the model parameters of the preset welding twin basic model to generate a dynamic welding twin model that matches the current working conditions. S3. Divide the total weld into several construction segments, and perform the following steps sequentially during the welding process of each segment: execute welding based on the prediction and planning of the current dynamic welding twin model; implement online stress reduction intervention on the intervention target area of ​​the completed weld in the current segment using a portable ultrasonic impact device that moves synchronously with the welding torch, based on the transient stress distribution indication of the current dynamic welding twin model; and iteratively update the current dynamic welding twin model based on the actual thermo-mechanical response data and impact force-displacement response data collected during the entire welding and stress reduction intervention process, and apply it to the welding planning of the next segment. S4. After all segmented welding and iterative updates are completed, targeted post-processing is performed on the areas where the measured residual stress obtained by the multi-source sensing unit exceeds the final prediction result of the current dynamic welding twin model after iterative updates. In step S3, the intervention target area refers to the stress concentration area calculated by the current dynamic welding twin model at the moment the current segmented welding operation is completed. The stress concentration area includes the area where the Mises equivalent stress exceeds 70-85% of the yield strength of the material at room temperature at the corresponding calculation point of the current dynamic welding twin model. The multi-source sensing unit, welding torch, and portable ultrasonic impact device are all communicatively connected to the central processing unit. The multi-source sensing unit also includes an ambient temperature sensor that monitors the ambient temperature of the welding operation area in real time. The central processing unit is pre-loaded with the welding process specification of the high-altitude steel structure to be welded, the three-dimensional structural model of the high-altitude steel structure to be welded, and the welding twin basic model. The welding twin basic model includes at least the following physical parameters: boundary heat transfer coefficient, equivalent restraint stiffness of the fixture, elastic modulus of the material at high temperature, effective power coefficient of the heat source, high strain rate cyclic hardening parameters, and equivalent boundary stiffness coefficient. Step S2 specifically includes: S21. The central processing unit controls the welding torch to perform a trial weld of length L on the initial weld area using preset basic process parameters. The length L ranges from 50 to 200 mm and is not less than 4 times the thickness d of the base material of the high-altitude steel structure to be welded. During the trial weld, the temperature sensor and strain sensor of the multi-source sensing unit synchronously collect the spatiotemporal distribution data of the temperature field in a strip area with a radius of not less than 2d on both sides of the weld centerline as the reference, as well as the real-time strain data of the trial weld area and adjacent monitoring points. The arrangement range of the adjacent monitoring points is no more than 1 / 2 of the distance from the edge of the trial weld area to the nearest structural discontinuity on the main force transmission component of the trial weld area. S22. The central processing unit uses the spatiotemporal distribution data of the temperature field collected by the multi-source sensing unit as the thermal input boundary condition, and loads it into the welding twin basic model for forward thermo-mechanical coupling calculation to obtain the calculated strain field data of the test welding area. The calculated strain field data is compared with the real-time strain data in the whole domain, and based on the comparison deviation, the following physical parameters in the welding twin basic model are dynamically adjusted through the reverse analysis algorithm: boundary heat transfer coefficient, equivalent restraint stiffness of the fixture, and elastic modulus of the material at high temperature, until the matching degree between the calculated and real-time strain data reaches the preset accuracy, thereby generating the dynamic welding twin model that matches the current working condition. S23. Based on the dynamic welding twin model generated in step S22 that matches the current working conditions, within the feasible domain of parameters such as welding current, arc voltage, and welding torch travel speed, an iterative simulation method is used to simulate the welding process of the initial segment of the formal welding, and a set of welding parameters that can make the simulated deformation less than the preset allowable threshold is found as the recommended welding parameters for the first segment of welding in step S3; the simulated deformation includes at least: the lateral shrinkage and angular deformation of the welding local area, and the maximum deflection of the entire component.

2. The construction method for pre-controlling welding deformation and releasing stress in high-altitude steel structures of chemical plants as described in claim 1, characterized in that, The stress concentration region also includes: At geometric discontinuities, the rate of change of principal stress per unit length exceeds 2.0 to 3.5 times the average stress per unit length of the cross section at the corresponding calculation point in the current dynamic welding twin model; The region where the transverse tensile stress component perpendicular to the weld direction exceeds 30-40% of the room temperature yield strength of the material at the corresponding calculation point in the current dynamic welding twin model.

3. The construction method for pre-controlling welding deformation and releasing stress in high-altitude steel structures of chemical plants as described in claim 2, characterized in that, Step S3, which involves predicting and planning welding based on the current dynamic welding twin model, specifically includes: For the current construction segment, the central processing unit retrieves the current dynamic welding twin model updated from the previous segment for use in the current segment. The central processing unit obtains the bevel geometry of the current segment from the pre-stored welding procedure specifications, the welding spatial position and weld orientation of the current segment from the structural 3D model, and the current ambient temperature read in real-time from the multi-source sensing unit, using these as input conditions to execute a welding process simulation. The simulation output includes the initial combination of recommended welding current, arc voltage, and welding torch travel speed for the current segment, as well as multi-layer, multi-pass welding parameters designed to balance heat input and deformation. The sequence of weld beads / layers and the start and end positions of each weld bead; during welding, the welding torch operates based on the above initial value combination, while the multi-source sensing unit monitors the actual arc power and molten pool morphology in real time. If the detected penetration state exceeds the acceptable range in the welding process specification, or the arc power exceeds its recommended value by ±10%, or the welding torch travel speed exceeds its recommended value by ±10%, the central processing unit will fine-tune the welding speed within a preset correction range. The preset correction range does not exceed ±20% of the planned welding speed to ensure that the actual heat input is consistent with the simulation planning target.

4. The construction method for pre-controlling welding deformation and releasing stress in high-altitude steel structures of chemical plants as described in claim 3, characterized in that, Step S3, based on the transient stress distribution indication of the current dynamic welding twin model, uses a portable ultrasonic impact device that synchronously follows the welding torch to implement online stress reduction intervention in the intervention target area of ​​the currently completed weld segment. Specifically, this includes: At the moment the current segment welding operation is completed, the central processing unit immediately drives the current dynamic welding twin model, and uses the transient temperature distribution data actually collected at the moment the current segment welding operation is completed as the heat input to perform a transient thermo-mechanical coupling simulation calculation to obtain the transient calculated strain field data at the moment the current segment welding operation is completed. The central processing unit runs a stress analysis algorithm to perform a grid scan on the obtained transient calculated stress field data, automatically identify all stress concentration points, and cluster the stress concentration points that are spatially close into stress concentration regions as intervention target areas. Based on the spatial distribution of the intervention target area, the central processing unit plans the optimal movement path and operation sequence of the portable ultrasonic shock device with the principle of minimizing the total movement distance. At the same time, based on the highest stress level in each intervention target area, it matches and specifies the corresponding shock process parameters of the portable ultrasonic shock device from a preset parameter database. The shock process parameters include at least the number of single-point shocks and the amplitude of the shock head. Finally, a digital intervention map integrating the coordinates of the intervention target area, the shock path, the operation sequence, and the shock process parameters is generated. The central processing unit sends the digital intervention map to the portable ultrasonic impact device in real time. Based on the digital intervention map, the portable ultrasonic impact device sequentially completes the impact operation on each intervention target area after welding is completed and before the interlayer temperature drops to 100~110℃. During the impact process, the force sensor and accelerometer built into the portable ultrasonic impact device synchronously collect the impact force-displacement response spectrum and vibration spectrum, and transmit them back to the central processing unit for subsequent verification and iterative updates of the current dynamic welding twin model.

5. The construction method for pre-controlling welding deformation and releasing stress in high-altitude steel structures of chemical plants as described in claim 4, characterized in that, Step S3 involves iteratively updating the current dynamic welding twin model based on the actual thermo-mechanical response data and impact-displacement response data collected throughout the welding and stress reduction intervention processes, and then applying this update to the next segmented welding plan. Specifically, this includes: S31. The central processing unit uses the actual temperature field distribution data and measured welding deformation data collected during the current segmented welding process as a benchmark, compares them with the prediction results of the current dynamic welding twin model for the current segmented welding process, calculates the prediction deviation in heat input, temperature field, and transient deformation, and optimizes and adjusts the effective power coefficient of the heat source and the equivalent boundary stiffness coefficient in the current dynamic welding twin model based on the above prediction deviation to form a transitional dynamic welding twin model. S32. Using the transitional dynamic welding twin model obtained in step S31 as a benchmark, the impact force-displacement response data collected throughout the stress reduction intervention process are used to perform inversion iterative calculations, and the high strain rate cyclic hardening parameters of the transitional dynamic welding twin model are adjusted to form a new dynamic welding twin model: (1) Set initial values ​​and variation ranges for the high strain rate cyclic hardening parameters; (2) Use the impact force-displacement response data collected during the stress reduction intervention process as the target curve, simulate the impact process in the transition dynamic welding twin model, and calculate a set of simulated impact response spectra; (3) Calculate the characteristic error between the simulated response spectrum and the target curve, wherein the characteristic error includes at least the error of peak force and the error of the area under the load-displacement curve; (4) Use an optimization algorithm to automatically adjust the value of the high strain rate cyclic hardening parameters, and repeat steps (2) to (3) until the sum of the characteristic errors is minimized and the convergence condition is met, and update the high strain rate cyclic hardening parameters obtained by the final optimization to the transition dynamic welding twin model to form a new dynamic welding twin model; S33. Using the new dynamic welding twin model obtained in step S32, re-simulate the welding process of the current segment and verify the consistency between the predicted data and the measured data. When the consistency reaches the preset standard, the iterative update is completed, and the current dynamic welding twin model for the next segment welding planning is generated.

6. The construction method for pre-controlling welding deformation and releasing stress in high-altitude steel structures of chemical plants as described in claim 5, characterized in that, Step S4 involves performing targeted post-processing on regions where the measured residual stress obtained by the multi-source sensing unit exceeds the final prediction result of the current dynamic welding twin model after the iterative update. The specific operations include: S41. After all welding, segmented online stress reduction intervention and current dynamic welding twin model iteration update are completed, a portable residual stress detector is used to measure the residual stress of the high-altitude steel structure and obtain the measured residual stress distribution data. The portable residual stress detector is connected to the central processing unit. S42. The central processing unit performs spatial registration and comparison between the measured residual stress distribution data and the residual stress prediction data finally output by the current dynamic welding twin model after iterative update. It identifies all areas where the measured residual stress values ​​exceed the predicted values ​​at the corresponding positions of the current dynamic welding twin model after iterative update, and the excess is greater than 15% of the material's room temperature yield strength. These areas are then identified as targeted post-processing areas. S43. The central processing unit plans post-processing parameters for each targeted post-processing area based on the measured residual stress magnitude, direction, and distribution characteristics within the targeted post-processing area, and generates targeted post-processing operation instructions: when the processing method is local heat treatment, the targeted post-processing operation instructions include heating temperature, heating rate, holding time, and heating range; when the processing method is supplementary ultrasonic impact, the targeted post-processing operation instructions include impact energy, coverage area, and impact path. S44. Process the corresponding post-processing area according to the post-processing operation instruction. After the processing is completed, re-measure the residual stress in the post-processing area. If the re-measurement result is lower than the preset safety threshold or consistent with the model prediction value, the processing is completed; otherwise, return to step S42, adjust the post-processing parameters based on the new re-measurement data and execute again until the re-measurement results are all lower than the preset safety threshold or consistent with the model prediction value.

7. The construction method for pre-controlling welding deformation and releasing stress in high-altitude steel structures of chemical plants as described in claim 6, characterized in that, The welded twin basic model is specifically constructed as a transient thermo-mechanical coupled finite element model based on the geometry of the three-dimensional structural model, wherein: The mesh density of the weld and the base material on both sides, where the peak temperature during the welding thermal cycle is between the Ac1 phase transformation point and the melting point of the material, is higher than that of the base material region of the high-altitude steel structure. The material property database in the model pre-stores functions of density, specific heat capacity, thermal conductivity, elastic modulus, and yield strength as a function of temperature. The heat source in the model is mathematically described using a moving double ellipsoid or Gaussian model, and its effective power is defined as a variable that is adjusted by the effective power coefficient of the heat source. The structural boundary constraints of the model are simulated by applying stiffness values ​​at the mounting connection points using spring elements defined by the equivalent restraint stiffness of the clamp and the equivalent boundary stiffness coefficient. The model has a preset ambient temperature loading point, which is used to receive real-time ambient temperature data from the multi-source sensing unit to dynamically update the heat transfer boundary conditions. The boundary heat transfer coefficient, the equivalent restraint stiffness of the clamp, the elastic modulus of the material at high temperature, the effective power coefficient of the heat source, the equivalent boundary stiffness coefficient, and the high strain rate cyclic hardening parameters are all defined in the model as independent input variables connected to the reverse analysis algorithm and optimization algorithm in the central processing unit, so as to support parameter calibration and iterative updates in steps S2 and S3.

Images