Heat exchanger welding process based on intelligent nitrogen protection

Abstract

The invention relates to the technical field of welding, and particularly discloses a heat exchanger welding process based on intelligent nitrogen protection, which comprises an intelligent nitrogen protection subsystem, a multi-sensor fusion monitoring subsystem and a central intelligent control subsystem, and adopts the process flow of path pre-scanning, dynamic environment establishment, closed-loop adaptive control and data archiving. Through a welding quality prediction model based on a deep neural network and an online optimization algorithm, the system can sense the welding state in real time and dynamically optimize process parameters such as power, speed and nitrogen proportion. Meanwhile, the system integrates an acoustic emission defect early warning and model self-learning updating mechanism. According to the method, the welding process is converted from fixed parameter dependence to real-time perception and dynamic decision making, disturbance such as assembly errors and thermal deformation can be actively adapted, the controllability of the welding process, the welding seam quality stability and the robustness to abnormal conditions are effectively improved, and finally the continuous process optimization effect of data driving is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of welding, in particular to a heat exchanger welding process based on intelligent nitrogen protection. BACKGROUND

[0002] As a key equipment in energy and chemical industries, the welding quality of the core components of the heat exchanger directly determines the sealing performance, pressure-bearing capacity and long-term operation reliability of the equipment. These components are often made of materials such as stainless steel, aluminum alloy and titanium alloy, which are extremely sensitive to heat input, oxidation and element loss during the welding process.

[0003] Currently, inert gas protection welding combined with high proportion of nitrogen is mainly used for protection, but the existing technology has significant problems: first, the protection gas control is rough, mostly in a static mode of fixed flow and proportion, which cannot adapt to the dynamic heat demand of complex welds, and is easy to form a protection blind area in the local area, resulting in oxidation or nitrogen element loss.

[0004] Secondly, the process monitoring and quality control are disconnected, and the monitoring is mostly limited to basic parameters such as current and voltage, lacking real-time and accurate perception of multi-dimensional physical fields such as molten pool morphology, temperature field and local atmosphere, and the process adjustment relies on manual experience, which cannot realize closed-loop optimization based on real-time state, and quality problems are often found after welding.

[0005] Thirdly, the nitrogen retention strategy for high-nitrogen steel is passive, lacking intelligent ability to dynamically and accurately adjust the nitrogen partial pressure of the protection atmosphere according to the real-time thermodynamic state of the molten pool, making it difficult to ensure the uniformity and stability of the nitrogen content in the weld.

[0006] Finally, the process data is isolated, and there is a lack of effective correlation and in-depth analysis between welding process parameters, sensing data and final quality results, making it difficult to form a data-driven process optimization closed loop, which restricts the quality traceability and standardization level.

[0007] Therefore, there is an urgent need for a welding process and system that can realize dynamic and accurate regulation of protection gas, real-time perception of multi-source information, and online closed-loop decision based on intelligent algorithms, to systematically solve the above problems and improve the stability, consistency and controllability of the welding quality of the heat exchanger. SUMMARY

[0008] The present application provides a heat exchanger welding process based on intelligent nitrogen protection to solve the technical problems of unstable welding quality, nitrogen element loss and difficulty in realizing closed-loop optimization and quality traceability caused by rough protection gas control, disconnection between process monitoring and quality control, lack of intelligent nitrogen retention strategy and isolated process data in the prior art.

[0009] The application adopts the following technical scheme: a heat exchanger welding process based on intelligent nitrogen protection, an integrated welding system is adopted, the system comprises an intelligent nitrogen protection subsystem, a multi-sensor fusion monitoring subsystem and a central intelligent control subsystem; the process is automatically executed by the central intelligent control subsystem, and comprises the following steps: S1, path pre-scanning and modeling: a three-dimensional scanning module in the multi-sensor fusion monitoring subsystem is controlled to perform full-line scanning on a to-be-welded weld of a heat exchanger workpiece, and three-dimensional topographic data of the weld is acquired; the central intelligent control subsystem generates an initial welding path based on the data and estimates welding heat input parameters; S2, establishing a dynamic protection environment: the intelligent nitrogen protection subsystem is started, an initial nitrogen mixture ratio and total flow are set according to a welding material type and the heat input parameters; a local environment cavity and a multi-path jet execution module are started, a welding area is isolated from the atmosphere, and oxygen content in the isolated area is confirmed to be reduced to below a set threshold value through an atmosphere monitoring unit; S3, closed-loop adaptive control: a welding power source is started, a welding torch is moved along a planned path and welding is started; in the welding process, the following closed-loop control process is cyclically executed: S31, the multi-sensor fusion monitoring subsystem acquires and uploads topographic, atmosphere, thermal field and position data of a welding area in real time; S32, the central intelligent control subsystem inputs the real-time data into a preset welding quality prediction model, and calculates a predicted quality index under current parameters; S33, taking optimal weld forming and performance as a target, an optimal process parameter combination in a next control cycle is solved through an optimization algorithm, the parameters at least comprising welding power, welding speed and nitrogen mixture ratio; S34, the optimized parameter instructions are synchronously issued to the welding power source and the intelligent nitrogen protection subsystem, and the welding process is adjusted; S4, post-welding evaluation and data archiving: after welding is completed, a final topographic scanning is performed on the weld, forming parameters of the weld are acquired; sensing data, control instructions and result data of the whole welding process are stored in association, and a welding quality digital archive is generated.

[0010] In some embodiments, the intelligent nitrogen protection subsystem comprises a nitrogen source, a dynamic mixing module, a multi-path jet execution module and a local environment cavity; An air inlet of the dynamic mixing module is connected with the nitrogen source and at least one auxiliary inert gas source, an air outlet thereof is connected with the multi-path jet execution module through a pipeline, and the dynamic mixing module internally comprises a mass flow controller for accurately controlling a mixing ratio and an online analyzer for feeding back a gas composition; The multi-path jet execution module comprises a plurality of micro electromagnetic valve-controlled nozzles arranged around a welding torch head and on a back surface of a weld; The local environment cavity is a flexible closed cover that can move with the welding gun; The subsystem has at least two controlled working states: The first protection state: the dynamic mixing module outputs gas according to the preset basic proportion, and all the nozzles of the multi-path gas jet execution module jet out at the basic flow rate, which is used to establish and maintain the initial protection environment before welding; The second protection state: according to the instruction of step S33, the dynamic mixing module adjusts the mixing proportion in real time, and the selected nozzles in the multi-path gas jet execution module adjust their opening degree and flow rate according to the molten pool position and thermal field distribution, which is used for dynamic local reinforcement protection during the welding process.

[0011] In some embodiments, the multi-sensor fusion monitoring subsystem comprises: The welding appearance visual module is a ring-shaped laser three-dimensional scanner, which is used to obtain high-precision three-dimensional point cloud data of the width, misalignment, fusion width and reinforcement of the weld before, during and after welding; The local atmosphere analysis module is integrated in the local environment cavity, which is used to monitor the oxygen content, nitrogen content and humidity at key positions in the cavity in real time; The molten pool thermal field monitoring module comprises an infrared thermal imager, which is used to collect the temperature field distribution data of the welding pool and its heat-affected zone; The weld tracking module is used to identify the offset between the welding gun and the center of the weld in real time; The central intelligent control subsystem receives the multi-source heterogeneous data of the above-mentioned modules and performs time and space synchronization and fusion processing.

[0012] In some embodiments, the welding quality prediction model in step S32 is a multi-physical field coupling model based on a deep neural network; the "multi-physical field coupling model" referred to in the present application refers to a mathematical model that can simultaneously process and correlate multiple physical field information (including temperature field, flow field, stress field, chemical composition field, etc.), and establish a mapping relationship between these physical field parameters and welding quality through a deep neural network.

[0013] The input layer feature vector of the model at least includes: real-time weld width and misalignment, protective atmosphere oxygen content, molten pool maximum temperature and temperature gradient, and current welding power and speed; The model is trained by historical process data to establish a nonlinear mapping relationship between the input features and the welding quality evaluation indexes; the welding quality evaluation indexes are the predicted penetration, porosity rate and nitrogen element loss and gain rate; wherein, the "nitrogen element loss and gain rate" is defined as the percentage change of nitrogen content in the weld metal relative to the nitrogen content of the base material, and the calculation formula is: N = (C_weld- C_base) / C_base × 100%, wherein C_weld is the nitrogen content of the weld, C_base is the nitrogen content of the base material, and a negative value indicates nitrogen loss and a positive value indicates nitrogen gain; The central intelligent control subsystem makes an optimized decision through a model predictive control algorithm based on the output of the model.

[0014] In some embodiments, the optimal combination of process parameters is solved by an optimization algorithm in step S33, specifically an online iterative implementation of gradient descent method. The optimization objective function is: min J = α·(Dt - Dp) + β·Pp + γ·|N| 2 Wherein, Dt is the target penetration, Dp is the model predicted penetration, Pp is the model predicted porosity, N is the model calculated nitrogen element loss and gain rate, and α, β, γ are weight coefficients set according to product requirements. By automatically adjusting the welding power, speed, nitrogen proportion and other process parameters, the value of the above weighted comprehensive loss function reaches the possible minimum value. In each control cycle, the algorithm takes the current process parameters as the starting point, calculates the partial derivative of the objective function with respect to each controllable parameter (welding power, welding speed, nitrogen proportion), and updates the parameters in the gradient descent direction until the parameter combination that minimizes the objective function is found as the optimal instruction output.

[0015] In some embodiments, during the welding process in step S3, a defect early warning sub-step based on acoustic emission signals is further included: Acoustic emission signals during the welding process are collected by acoustic emission sensors. Wavelet packet transform analysis is performed on the acoustic emission signals to extract signal energy features in a specific frequency band (100 kHz-300 kHz); the selection of this frequency band is based on the following basis: signals below 100 kHz are mainly derived from mechanical vibration and environmental noise, signals above 300 kHz are severely attenuated and greatly affected by electromagnetic interference, while the 100 kHz-300 kHz frequency band can effectively represent acoustic emission events related to welding defects such as unstable molten pool, bubble formation and micro-crack initiation. When the energy feature continuously exceeds a predetermined threshold, it is determined that the welding process has a spatter or micro-crack tendency. The central intelligent control subsystem immediately triggers a first-level alarm and calls a preset emergency parameter set (including reducing power and adjusting gas flow) for intervention. The "first-level alarm" in this application refers to the highest priority alarm that the system responds to automatically, and after triggering, the system immediately executes the preset emergency intervention measures without human confirmation.

[0016] ​In some embodiments, the auxiliary inert gas is helium or argon; when the welding material is high-nitrogen stainless steel, the initial nitrogen gas proportion is set to 95%-99.5%; the selection basis of this range is that: when lower than 95%, the nitrogen partial pressure is insufficient to effectively inhibit the nitrogen loss of the weld, and when higher than 99.5%, the ionization characteristics and heat conduction performance of the mixed gas decrease, affecting the arc stability and the protection effect of the molten pool. During the welding process, if the local atmosphere analysis module monitors that the nitrogen content continues to decrease, or the welding quality prediction model outputs a higher nitrogen loss rate prediction value, the central intelligent control subsystem dynamically instructs the dynamic mixing module to increase the nitrogen gas proportion by 1%-3%; the selection basis of this adjustment range is that: an adjustment range less than 1% has limited effect on the nitrogen partial pressure, and it is difficult to effectively compensate for the nitrogen loss, and an adjustment range greater than 3% may cause the mixed gas proportion to fluctuate too much, affecting the arc and molten pool stability.

[0017] In some embodiments, after step S4, step S5 of calibration verification and model updating is further included. Specifically: periodically use standard test pieces for verification welding, perform nondestructive testing and mechanical property testing on the welded test pieces, and obtain actual welding quality data; Compare the actual data with the data predicted by the model during the welding process, and calculate the prediction error; If the error exceeds the allowed range, use the complete data chain (sensing data, process parameters, and actual results) generated by this verification as a new sample, and fine-tune the internal weight parameters of the welding quality prediction model through the back propagation algorithm, to realize self-learning and precision evolution of the model.

[0018] The application also discloses the following technical solutions: A heat exchanger welding system based on intelligent nitrogen protection, comprising: an intelligent nitrogen protection subsystem, a multi-sensing fusion monitoring subsystem, a central intelligent control subsystem, and a welding execution unit; The central intelligent control subsystem is electrically connected with all the other subsystems and units; The central intelligent control subsystem stores an executable program, and the program is used for automatically executing the process flow of steps S1-S4 or executing the whole process flow including step S5 when the program is run.

[0019] In some embodiments, the system further comprises a remote interaction interface; Through the interface, the system can receive a welding task file, material parameter library update, or model algorithm update package issued by an upper computer or a cloud platform; At the same time, the system can encrypt and upload the welding process data and the generated welding quality digital archives for remote monitoring, big data analysis, or supply chain quality traceability.

[0020] The application realizes the transformation of the welding process from relying on fixed parameters and experience judgment to real-time state perception and dynamic optimization decision, thereby actively adapting to dynamic disturbances such as workpiece assembly error and thermal deformation, and improving the controllability of the welding process and the stability of the weld quality (such as weld penetration consistency and forming quality).

[0021] Secondly, the application realizes quality prediction and parameter optimization of the welding process by setting up a welding quality prediction model based on a deep neural network and an online optimization algorithm taking subsequent process parameters as optimization variables and quantitative quality indicators as objective functions, converts the process relationship that is difficult to accurately model into a data-driven intelligent decision problem, automatically solves the optimal combination of power, speed and nitrogen ratio parameters in each control cycle, and thereby actively maintains the precision of weld penetration, low porosity and uniform and stable nitrogen content in environmental fluctuations.

[0022] Furthermore, the application realizes early detection and rapid inhibition of the micro-defect tendency of the welding process and continuous calibration and evolution of the system prediction performance by setting up a parallel independent acoustic emission defect early warning thread and a model self-learning update mechanism based on standard specimen verification, optimizes the control to increase the dynamic safety line, and responds to factors such as material batch fluctuation and equipment state drift, thereby improving the process level and product quality consistency of the heat exchanger welding. BRIEF DESCRIPTION OF DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings of the embodiments will be briefly introduced below. Obviously, the drawings in the following description only relate to some embodiments of the present application, but not limit the present application.

[0024] Figure 1 is the main process flowchart of the present application; Figure 2 is the sub-flowchart of the closed-loop adaptive control in the present application; Figure 3 is the sub-flowchart of the defect early warning in the present application; Figure 4 is the sub-flowchart of the process data packet in the present application. DETAILED DESCRIPTION

[0025] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application. Example 1

[0026] This application discloses a heat exchanger welding process based on intelligent nitrogen protection, referring to... Figure 1 The main process flow diagram shown includes the following steps: Step S1: First, the heat exchanger workpiece to be welded, such as the corrugated plate of a stainless steel plate heat exchanger, is fixed on the welding worktable. The 3D scanning module (laser 3D scanner) in the multi-sensor fusion monitoring subsystem is controlled to perform a full-line scan along the theoretical trajectory of the weld seam on the workpiece. This scanning process acquires high-density 3D point cloud data of the weld seam and its adjacent area by emitting a line laser and receiving its deformed contour. After receiving this point cloud data, the central intelligent control subsystem generates an actual 3D topographic model of the weld seam area through point cloud processing algorithms (triangular meshing and surface reconstruction).

[0027] Based on this model, the system extracts the actual centerline, cross-sectional width, and potential assembly misalignment of the weld, thereby generating an initial welding path adapted to the actual workpiece condition. Simultaneously, the system preliminarily estimates the required range of thermal input parameters based on the weld's cross-sectional dimensions, the thermophysical properties of the material (retrieved from a pre-defined material process database), and the preset welding speed range.

[0028] In this step, a ring laser 3D scanner is used to pre-scan the weld seam and construct a 3D model, enabling precise digital perception of the actual workpiece condition. This allows welding path planning to be based on the actual shape of the workpiece, reducing the risk of welding defects caused by assembly errors.

[0029] Step S2: Subsequently, the intelligent nitrogen protection subsystem is activated. The central intelligent control subsystem retrieves the corresponding initial process parameter set from the process database based on the welding base material type (such as SUS304 austenitic stainless steel) identified in step S1 and the estimated heat input parameters.

[0030] The parameter set contains the initial nitrogen mixture ratio (for example, for high-nitrogen stainless steel, set to 98% N2 and 2% Ar) and the initial total gas flow rate for the current material. The system instructs the dynamic mixing module to mix the gases according to the ratio and flow rate, and ensures accuracy through the mass flow controller. At the same time, the instruction multi-path gas injection execution module is started, causing the multiple micro-nozzles arranged around the head of the welding torch and on the back of the workpiece to start spraying protective gas at the base flow rate.

[0031] At the same time, the local environment cavity made of flexible high-temperature-resistant material is lowered under the action of the driving mechanism, and the sealing strip at its edge is attached to the surface of the workpiece, isolating the laser action point, the molten pool and its surrounding area, forming a local controllable atmosphere environment. The atmosphere monitoring unit (such as a laser oxygen analyzer) integrated in the cavity starts to work, and the oxygen content in the cavity is monitored in real time. The central intelligent control subsystem continuously reads the oxygen content data, and when it is confirmed that it is stably decreasing and lower than the preset threshold (for example, 50 ppm), it is determined that the dynamic protection environment has been successfully established and meets the standards, and is ready to enter the welding stage.

[0032] In this step, by first establishing a local protection environment and then starting welding, an initial, clean, low-oxygen, high-nitrogen atmosphere is provided for the welding pool, effectively preventing metal oxidation at the beginning of welding.

[0033] Step S3, refer to Figure 2 After confirming that the protection environment meets the standards, the welding power supply is started, and the laser welding head is driven to start moving along the initial welding path generated in step S1, the laser beam acts on the workpiece to form a molten pool, and the welding process officially starts, with a fixed cycle (for example, 100 milliseconds) for the welding process.

[0034] In each control cycle, first execute sub-step S31: the multi-sensor fusion monitoring subsystem synchronously collects multi-physical field data of the welding area. The welding appearance vision module captures the image of the molten pool and the weld, and extracts the characteristics such as the width and length of the molten pool; the local atmosphere analysis module continuously reads the nitrogen and oxygen concentrations above the molten pool in the cavity; the molten pool thermal field monitoring module (infrared thermal imager) outputs the temperature field distribution map of the molten pool area, calculates the highest temperature and temperature gradient; the weld tracking module calculates the deviation between the welding torch spot and the actual center line of the weld in real time. All these real-time data are synchronously uploaded to the central intelligent control subsystem.

[0035] Then, in sub-step S32, the central intelligent control subsystem combines the above real-time data, i.e., the molten pool width Wp, the oxygen content [O2], and the molten pool temperature Tmax, into a feature vector, and inputs the feature vector into a preset welding quality prediction model. The welding quality prediction model is a deep neural network trained by a large amount of historical welding data. After receiving the real-time feature vector, the model calculates the predicted values of key quality indicators that can be formed in the weld under the current welding power, welding speed, and environmental state. The predicted values mainly include a predicted penetration depth Dp, a predicted porosity Pp, and a predicted nitrogen element loss and gain rate N.

[0036] Subsequently, in sub-step S33, the system starts an optimization algorithm with the optimal weld formation and performance as the decision target. Specifically, the algorithm constructs an objective function based on the predicted values (Dp, Pp, N) obtained in step S32, in combination with a preset target penetration depth Dt and an expected nitrogen retention level: min J = a · (Dt - Dp) 2 + β · Pp + γ · |N|; In the function, the penetration depth accuracy, low porosity, and nitrogen stability are quantified into a scalar value. The gradient descent method is used to perform a rapid search near the current process parameter point, and to calculate how to adjust the three key controllable parameters, i.e., the welding power, the welding speed, and the nitrogen gas mixing ratio, so that the objective function value is minimized, i.e., the optimal parameter combination (Pn, Vn, Rn) that comprehensively optimizes the quality is found.

[0037] Finally, in sub-step S34, the central intelligent control subsystem converts the optimal parameter combination obtained in sub-step S33 into specific control instructions, and synchronously issues the control instructions to the welding power source and the intelligent nitrogen protection subsystem. The welding power source adjusts the laser output power according to the control instructions, the motion controller adjusts the welding speed, and the intelligent nitrogen protection subsystem instructs the dynamic mixing module to adjust the gas mixing ratio. One control cycle ends, and the system immediately enters the next cycle, repeating S31 to S34, so as to realize continuous, dynamic, and closed-loop optimization control of the entire welding process.

[0038] In this step, by setting a closed-loop control process including real-time sensing (S31), intelligent prediction (S32), multi-objective optimization (S33), and precise execution (S34), and by using a welding quality prediction model with a deep neural network as the core, the status of the welding process is sensed and dynamically optimized, and the current situation of presetting process parameters by experience in the traditional welding process is changed. Furthermore, by re-solving the optimal parameters based on the real-time state in each control cycle, the welding process can adaptively cope with dynamic disturbances such as micro-unevenness of the workpiece itself and deformation caused by heat accumulation, so as to actively maintain the stability of the welding quality in the welding fluctuation, thereby ensuring that the weld has consistent penetration depth, compactness, and uniformity of chemical composition.

[0039] Step S4, when the welding torch completes the entire path scanning, the welding process is completed. The three-dimensional scanning module controls the final topography scanning of the cooled weld to obtain the final forming parameters such as the weld reinforcement, the weld width, and the surface depression. The central intelligent control subsystem aligns, correlates, and encapsulates the time-stamped sequence of sensing data streams recorded during the entire welding process (all data collected in step S31), the control instruction stream issued (all instructions output in step S34), the intermediate result stream predicted by the model (output in step S32), and the final topography scanning result.

[0040] The above data is packaged to generate a digital file of the welding quality of the workpiece to record the final welding result of the workpiece, thereby providing a complete data basis for quality traceability, process analysis, and continuous optimization of finished products. Embodiment 2

[0041] Based on embodiment one, this embodiment further details the specific composition and interaction of the intelligent nitrogen protection subsystem and the multi-sensor fusion monitoring subsystem, as well as the specific implementation method of the welding quality prediction model and optimization decision, specifically: Referring to the system principle block diagram, the intelligent nitrogen protection subsystem is used to realize dynamic regulation and control of gas composition and spatial distribution. The gas inlet of this module is connected to a pure nitrogen gas source and an auxiliary inert gas source through high-pressure pipelines. A mass flow controller is installed on each gas inlet branch, and its control signal is received from the central intelligent control subsystem. The two gases are mixed in the instructed proportion, flow through the online laser gas analyzer to detect the actual composition of the mixed gas in real time, and feedback the data to the central controller, forming a closed-loop control circuit for the gas mixing ratio, thereby ensuring that the composition of the output gas in the welding area is consistent with the instructed value.

[0042] The mixed gas is delivered to the multi-way gas jet execution module. This module forms a main protective gas curtain covering the front and side of the molten pool to cooperate with the local environmental cavity to physically isolate the welding area. The local environmental cavity can be in the form of a bell-shaped structure made of high-temperature resistant ceramic fiber cloth or flexible metal bellows, which is not specifically limited in the present application.

[0043] This system has two controlled working states: Static preparation: Before the welding starts, the dynamic mixing module outputs gas in the initial proportion (such as 98% N2-2% Ar for high-nitrogen stainless steel), and the gas nozzle is opened at a uniform and relatively low "base flow". The air in the cavity is replaced with protective gas, and a stable flow field is established to reduce the oxygen content below the threshold value.

[0044] Dynamic welding: after the welding is started, the system switches to this state. At this time, the gas control is directly driven by the optimization algorithm output instruction of step S33. When the model predicts that the current segment heat input is high and the risk of nitrogen escape increases, the algorithm instruction dynamic mixing module will fine-tune the nitrogen proportion from 98% to 99%. At the same time, according to the infrared thermal image, the temperature of the tail of the molten pool is too high, and the algorithm instruction increases the flow of an auxiliary nozzle pointing to the region to enhance cooling and suppress grain coarsening.

[0045] The multi-sensor fusion monitoring subsystem is a multi-source heterogeneous data acquisition and preprocessing platform. The welding morphology vision module uses a ring laser three-dimensional scanner. The emitted ring laser is projected on the workpiece surface, and the image of the deformed ring is captured by a CCD camera. The two-dimensional profile of the position of the laser stripe is calculated in real time by the triangulation principle, and the three-dimensional point cloud of the weld area is reconstructed by combining the scanning motion. Thus, the image and point cloud data are directly output, and the weld width, molten pool boundary, and other geometric parameters are calculated in real time by the processing unit in the central intelligent control subsystem.

[0046] The local atmosphere analysis module directly inserts the sampling probe of a miniaturized laser gas analyzer or mass spectrometer into the local environment cavity to capture the most representative atmosphere sample. Its output is a real-time gas concentration digital signal (O2, N2, H2O, etc.).

[0047] The molten pool thermal field monitoring module uses a high-speed infrared thermal imager to capture the molten pool area. After radiance correction, the temperature field matrix data can extract the molten pool center temperature, molten pool size, solid-liquid interface isotherm, and its moving speed.

[0048] The weld tracking module usually uses a coaxial or paraxial high-speed CMOS camera combined with structured light assistance to identify the relative position of the weld bevel edge and the laser spot in real time, and output the deviation of the welding torch in the horizontal and height directions.

[0049] The central intelligent control subsystem ensures that the data collected by different sensors are aligned in time. After all the data are transmitted, they are sent to a data fusion processing unit for timestamp alignment to unify the data of different physical quantities (length, temperature, concentration, pixel deviation) to the same space-time coordinate system.

[0050] For example, the position of the highest temperature point of the tail of the molten pool in the infrared thermal image at a certain time t is spatially registered with the geometric profile of the molten pool in the three-dimensional point cloud at the same time t, so that the actual temperature of a certain specific geometric position can be analyzed.

[0051] The welding quality prediction model in step S32 is constructed and run as follows: In the offline stage, a large amount of welding test data covering different materials, process parameters, and working conditions is collected. Each data point is a complete data chain, including input features during the welding process and the actual quality results obtained after welding through destructive testing (such as metallographic analysis to measure weld penetration, nitrogen content analysis, and X-ray flaw detection to measure porosity). The above data is used to supervise the training of a deep neural network (preferably an LSTM network for processing time-series data, whose typical structure includes: an input layer receiving 8-12 dimensional feature vectors, 2-3 LSTM hidden layers each containing 64-128 neurons, and a fully connected output layer outputting a 3-dimensional prediction vector). After training, a complex nonlinear mapping relationship from the input feature space to the output quality space is formed within the network.

[0052] During online prediction, the real-time feature vector processed in step S31, for example: [weld width = 3.2mm, misalignment = 0.1mm, O2 = 20ppm, N2 = 99.1%, maximum molten pool temperature = 2100K, temperature gradient = 150K / mm, current power = 3kW, current velocity = 1.2m / min], is input into the network. After forward propagation calculation, the network outputs the corresponding predicted value vector: [predicted weld depth = 4.1mm, predicted porosity = 0.3%, predicted nitrogen loss rate = -0.05%]. Here, a negative "nitrogen loss rate" indicates that the predicted weld nitrogen content is 0.05% lower than that of the base metal.

[0053] The optimization decision process in step S33 is an online solver. Its core is the optimization objective function as described above. For example, for a high-nitrogen stainless steel weld, the required penetration depth is 4.0 mm, with nitrogen loss as minimal as possible. Weighting coefficients are set as α = 1.0, β = 10.0 (highly emphasizing porosity), and γ = 5.0 (highly emphasizing nitrogen retention). Assuming the current model predicts Dp = 4.1 mm, Pp = 0.3%, N = -0.05%, and the target penetration depth Dt = 4.0 mm, then the current objective function value is J = 1.0 × (4.0 - 4.1). 2 + 10.0×0.003 + 5.0×0.0005 = 0.0425.

[0054] An optimization algorithm using gradient descent is employed to find a new set of parameters (P, V, R) such that substituting the new parameters and the predicted new state into the function yields a value smaller than 0.0425. The algorithm starts with the current parameters (Pc, Vc, Rc) and uses numerical methods to estimate the partial derivatives of the objective function J with respect to each parameter. The gradient of the distribution is characterized by calculating its partial derivatives, and the gradient indicates the direction in which the function value decreases the fastest.

[0055] If calculations reveal A positive and large value indicates that increasing the power at the current point will increase the function value (i.e., the overall quality loss), therefore, the power should be reduced. The algorithm updates the parameters in the opposite direction of the gradient, with a small "learning rate" as the step size, obtaining a set of candidate new parameters. The changes in sensor data caused by these new parameters are input into the welding quality prediction model to obtain the predicted quality value under the new parameters, and then the new function value is calculated. Through several iterations, a local minimum point of the function value is found, and the corresponding parameter combination (Pn, Vn, Rn) is identified as the optimal solution for the current cycle. Example 3

[0056] Based on Examples 1 and 2, this example further elaborates on the defect early warning mechanism that is executed in parallel during the welding process, as follows: While the main closed-loop control process in step S3 is running, the defect early warning and monitoring thread is started and runs continuously in the background of the system. This thread is used to process the acoustic emission signals generated during the welding process.

[0057] The execution logic of this thread refers to Figure 3 First, a high-sensitivity acoustic emission sensor is coupled to the workpiece via a waveguide rod. The waveguide rod is installed on the worktable or in the non-welding area of ​​the workpiece to collect elastic stress wave signals excited by events such as molten pool fluctuations, bubble formation and escape, and microcrack generation during the welding process. The original acoustic emission signal is broadband and contains a large amount of noise.

[0058] The acquired raw signal is sent to the signal processing unit. First, bandpass filtering is performed to remove significant low-frequency mechanical vibration noise and high-frequency electrical noise. Then, the initially filtered bands are subjected to wavelet packet transform analysis. Wavelet packet transform can decompose the signal into a finer frequency band, providing higher frequency resolution. The system decomposes the filtered signal into a series of preset frequency band sub-bands.

[0059] The signal processing unit extracts the signal energy characteristics within these specific frequency bands and calculates the average energy or root mean square value of the signal in the 100-300kHz frequency band within each control cycle (e.g., 10 milliseconds, shorter than the main control cycle). This energy value is transmitted to the central intelligent control subsystem in real time.

[0060] The central intelligent control subsystem has a preset calibrated energy threshold. This threshold was obtained through extensive preliminary process experiments, statistically analyzing the acoustic emission energy distribution during normal welding and welding with known defects. The early warning thread continuously compares the real-time energy characteristics with this threshold. To improve the reliability of the early warning, the system is configured with continuous triggering conditions, such as requiring the energy characteristic value to exceed the threshold within three consecutive early warning cycles (i.e., 30 milliseconds), to avoid false alarms caused by accidental electrical noise or transient interference.

[0061] Once the aforementioned continuous over-threshold conditions are met, the system immediately determines that the welding process has an abnormal tendency, such as an unstable molten pool leading to increased spatter, or thermal stress concentration potentially causing microcrack initiation. At this time, the central intelligent control subsystem immediately triggers a level one alarm. Simultaneously, it calls up a preset emergency parameter set, instantly reducing the laser power by 10%–20%, increasing the welding speed by 5%–10%, and instructing the intelligent nitrogen protection subsystem to increase the protective gas flow rate to stabilize the molten pool. At the same time, it interrupts the parameters output by the current main control cycle and directly issues them for execution, thereby strongly intervening in the welding process.

[0062] After intervention, the early warning thread continues to monitor acoustic emission energy. Typically, effective intervention weakens the abnormal signal. Once the energy characteristic value falls below the threshold and stabilizes for a period of time, the system automatically deactivates the alarm and gradually returns control to the main control loop based on the optimization algorithm, or alerts the operator to intervene and investigate.

[0063] Step S4, which completes data archiving, is not the end of the process data. This system also includes a step S5 that enables the system to continuously evolve: calibration verification and model update.

[0064] Specifically, the system is required to perform a verification process after welding a certain number of production products (e.g., every 50 workpieces), or periodically (e.g., weekly). The operator mounts a standard verification specimen onto the workbench. This specimen is made of the same material and has the same thickness as the production product, with a manually machined standard bevel or a groove of known dimensions. The system performs the complete S1 to S4 welding process on this specimen, using all parameters and models within the current system, just as it would with a production product.

[0065] After welding, the verification specimen is removed and sent to the testing laboratory for comprehensive offline testing. This typically includes: wire cutting to obtain metallographic samples of the weld cross-section; precise measurement of the actual weld penetration, weld width, and heat-affected zone width under an optical microscope; measurement of hardness distribution using a microhardness tester; measurement of the actual nitrogen content at the weld center, possibly using electron probe microanalysis (EPMA) or spectral analysis; and detection of the number and size of internal pores using real-time X-ray imaging (RT) or computed tomography (CT). These test results provide accurate and reliable quality data for the weld.

[0066] Subsequently, the system performs data comparison and error analysis. The central intelligent control subsystem retrieves the entire process digital archive generated during the welding of the verification specimen, especially the continuous predictions of quality indicators (penetration depth, porosity, and nitrogen loss rate) at various locations on the specimen by the model during welding. These "predicted value curves" are compared with the "actual value points" (actual data from several key sections on the specimen) obtained from offline detection at the same coordinates. The system calculates the prediction errors for key indicators, such as the mean absolute penetration depth error, the false negative / false positive rate of porosity prediction, and the deviation of nitrogen content prediction.

[0067] If all error indicators are within the preset allowable range (e.g., melting depth error < ±5%, nitrogen content error < ±0.02%), then the current system model and parameter performance are considered reliable, the verification is passed, and the system can continue to be used in production.

[0068] If one or more error indicators exceed the allowable range, it indicates that the current welding quality prediction model has deviated in its predictive ability under the current process boundaries or material properties. At this point, the system initiates a model update procedure. The complete data chain generated from this welding verification—from multi-sensor fusion data to process parameters and precise offline detection results—constitutes a "new sample." This sample is added to the historical training database. The system uses the backpropagation algorithm to fine-tune the existing welding quality prediction neural network model with this new sample (or combined with a recent batch of production data). The fine-tuning process iteratively optimizes the model's weights with a small learning rate, bringing the model's predictive output closer to the new real data, thereby correcting its mapping relationship. After verification, the updated model will replace the old model and be put into online use. Example 4

[0069] Based on Examples 1 to 3, this example provides a specific implementation method of the present invention in a real industrial production scenario, as follows: I. Application Scenarios and Equipment Configuration The company manufactures plate heat exchangers for the chlor-alkali industry. The heat exchange plates are made of 2205 duplex stainless steel with a thickness of 2.0 mm and a total weld length of approximately 1200 mm per plate.

[0070] The specific hardware configuration of the system is as follows: the welding power supply adopts a fiber laser with a rated power of 4kW; the intelligent nitrogen protection subsystem is equipped with a 50L liquid nitrogen storage tank as a nitrogen source and a 10L high-purity argon cylinder as an auxiliary inert gas source; the dynamic mixing module adopts a two-way mass flow controller with a range of 0-50L / min (nitrogen) and 0-5L / min (argon) respectively, and an accuracy of ±1% of the full scale; the multi-channel jet execution module includes 6 miniature electromagnetic valve-controlled nozzles arranged around the welding torch head and 2 back protection nozzles located on the back of the workpiece; the local environment cavity is made of ceramic fiber cloth with a temperature resistance of 800℃, and the internal volume is about 200mL.

[0071] The multi-sensor fusion monitoring subsystem is configured as follows: the welding morphology vision module uses a ring laser 3D scanner with a resolution of 0.05mm; the local atmosphere analysis module uses a laser oxygen analyzer with a response time of less than 1 second and a range of 0-1000ppm; the molten pool thermal field monitoring module uses an infrared thermal imager with a frame rate of 100Hz and a temperature resolution of ±2℃; and the weld seam tracking module uses a high-speed CMOS camera with a frame rate of 200fps.

[0072] The central intelligent control subsystem uses an industrial computer with an Intel Core i7 processor, 32GB of memory, and runs a Linux-based real-time operating system with a control cycle set to 50ms.

[0073] II. Specific welding process: The operator fixes the 2205 duplex stainless steel heat exchange plate to be welded onto a special fixture on the welding workbench. Through the human-machine interface, the operator selects the material type as "2205 duplex stainless steel". The system automatically retrieves the corresponding initial process parameters from the process database: initial laser power 2.5kW, initial welding speed 1.0m / min, initial nitrogen ratio 97%, initial total gas flow rate 30L / min, and target penetration depth 1.8mm.

[0074] During step S1, the 3D scanning module performs a full-line scan along the theoretical weld trajectory at a speed of 200 mm / s, taking approximately 6 seconds and acquiring a total of about 24,000 3D point cloud data points of the weld. After processing the point cloud data, the central intelligent control subsystem identifies a maximum lateral offset of 0.15 mm between the actual weld centerline and the theoretical trajectory, as well as a local misalignment of 0.08 mm, and generates a corrected actual welding path accordingly.

[0075] During step S2, the system activates the intelligent nitrogen protection subsystem. The dynamic mixing module outputs a mixed gas at a ratio of 97% N2 - 3% Ar, with a total flow rate of 30 L / min. The local environmental chamber descends and comes into contact with the workpiece surface. The six welding torch nozzles and two back protection nozzles simultaneously open at the base flow rate. The atmosphere monitoring unit continuously monitors the oxygen content within the chamber. After approximately 15 seconds, the oxygen content decreases from the initial approximately 21% (atmospheric concentration) and stabilizes at 35 ppm, below the preset threshold of 50 ppm. The system then determines that the protective environment has been successfully established.

[0076] When step S3 is executed, the laser is activated, and the welding torch begins to move along the planned path. During the welding process, the system performs closed-loop control in a 50ms cycle. Taking a control cycle at the 600mm mark (middle of the weld) as an example: In sub-step S31, the real-time data collected by the multi-sensor fusion monitoring subsystem are: weld width 2.1 mm, misalignment 0.06 mm, oxygen content in the cavity 38 ppm, nitrogen content 96.8%, maximum temperature of the molten pool 1680 ℃, temperature gradient 125 ℃ / mm, and welding torch lateral offset 0.03 mm.

[0077] In sub-step S32, the above data and the current process parameters (power 2.5kW, speed 1.0m / min, nitrogen ratio 97%) form a feature vector input to the welding quality prediction model. The model outputs the following predicted values: predicted penetration depth 1.75mm, predicted porosity 0.15%, and predicted nitrogen loss rate -0.02%.

[0078] In sub-step S33, the system constructs an objective function with a target melt depth Dt = 1.8 mm and weighting coefficients α = 1.0, β = 8.0, and γ = 6.0. The current objective function value is J = 0.0157. After three iterations, the optimization algorithm finds the optimal parameter combination that reduces the objective function value to 0.0098: power 2.6 kW, speed 0.98 m / min, and nitrogen ratio 97.5%.

[0079] In sub-step S34, the system issues the optimized parameter commands: the laser power is adjusted from 2.5kW to 2.6kW, the motion controller adjusts the speed from 1.0m / min to 0.98m / min, and the dynamic mixing module adjusts the nitrogen ratio from 97% to 97.5%.

[0080] The entire welding process, with a weld length of 1200mm, took approximately 75 seconds, during which the system performed closed-loop optimization for approximately 1500 control cycles.

[0081] During step S4, after welding is completed, the 3D scanning module performs a final morphological scan of the weld, measuring an average weld reinforcement height of 0.12 mm and an average weld width of 2.3 mm, with no obvious surface defects. The system packages all process data into a digital weld quality archive, with a file size of approximately 15 MB, containing sensor data timing records, control command records, model prediction records, and final morphological data.

[0082] The following are two real-world examples of several abnormal operating conditions encountered during mass production: Case 1: Abnormal local assembly clearance During the welding of the 23rd heat exchange plate, when the welding reached the 850mm mark, the weld tracking module detected a sudden increase in the assembly gap to 0.35mm, exceeding the normal range of 0.1-0.2mm. The system immediately triggered the abnormal response mechanism. First, the welding quality prediction model predicts that if welding continues with the original parameters, the penetration depth at this point will drop to 1.55 mm, which is about 14% lower than the target value of 1.8 mm, posing a risk of incomplete penetration.

[0083] Subsequently, the optimization algorithm quickly solves the problem under the constraints and provides a compensation scheme within two control cycles (100ms): increasing the laser power from the current 2.55kW to 2.85kW, reducing the welding speed from 0.98m / min to 0.82m / min, and increasing the nitrogen flow rate from 30L / min to 35L / min to enhance the protection effect.

[0084] After the system implemented the aforementioned compensation parameters, the actual weld penetration depth in the abnormal section was subsequently measured at 1.73 mm. Although slightly lower than the target value, it was still within an acceptable range, successfully preventing the occurrence of welding defects. When the welding torch passed through the abnormal section, the system detected that the gap had returned to normal and automatically adjusted each parameter back to the standard value step by step.

[0085] Case 2: Fluctuations in Protective Atmosphere During the welding of the 41st heat exchange plate, due to airflow disturbance caused by the start-up of an adjacent piece of equipment in the workshop, the local atmosphere analysis module detected that the oxygen content in the cavity rose from 32ppm to 78ppm in about 3 seconds, exceeding the preset warning threshold of 50ppm.

[0086] The system immediately initiates the atmosphere compensation procedure: the flow rates of the two nozzles closest to the disturbance source in the multi-jet actuator module are increased from the base value of 5L / min to 12L / min, forming a directional air curtain to block the intrusion of external airflow; at the same time, the dynamic mixing module temporarily increases the total output flow rate from 30L / min to 42L / min.

[0087] After approximately 2 seconds of compensation adjustment, the oxygen content in the cavity dropped back below 45 ppm. During this period, the system simultaneously and temporarily reduced the welding speed from 1.0 m / min to 0.85 m / min to ensure sufficient shielding gas coverage time per unit length of weld. Once the atmosphere stabilized, all parameters automatically returned to normal values.

[0088] The test results after the plate was welded showed that the nitrogen content of the weld in the atmospheric fluctuation zone was 0.27%, which was very similar to the 0.29% in other normal zones, indicating that the compensation measures effectively prevented abnormal nitrogen loss.

[0089] Comparative Experiment: To verify the technical advantages of this invention, another batch of 10 heat exchange plates of the same specification were produced using a traditional fixed-parameter welding process for comparison. The traditional process uses constant parameter settings of 2.5kW laser power, 1.0m / min welding speed, and 25L / min nitrogen flow rate, which lacks real-time monitoring and dynamic adjustment capabilities.

[0090] The comparison results are shown in the table below:

[0091] As can be seen from the table above, although the welding time per piece increases slightly (approximately 4%), the welding quality is significantly improved, with a 16 percentage point increase in the first-pass yield and a 16 percentage point decrease in the rework rate. Considering the additional time and material costs required for rework, the overall production efficiency of the process of this invention is actually increased by approximately 12%. The above embodiments fully illustrate the feasibility and technical effects of the heat exchanger welding process based on intelligent nitrogen protection in actual industrial production scenarios, providing specific implementation support for the technical solutions described in the claims of this invention.

[0092] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A heat exchanger welding process based on intelligent nitrogen protection, employing an integrated welding system, which includes an intelligent nitrogen protection subsystem, a multi-sensor fusion monitoring subsystem, and a central intelligent control subsystem; This process is automatically executed by a central intelligent control subsystem, and is characterized by including the following steps: S1. Path pre-scanning and modeling: Control the three-dimensional scanning module in the multi-sensor fusion monitoring subsystem to perform a full-line scan of the weld seam to be welded on the heat exchanger workpiece and obtain the three-dimensional morphology data of the weld seam; Based on this data, the central intelligent control subsystem generates an initial welding path and estimates the welding heat input parameters. S2. Establish a dynamic protective environment: Activate the intelligent nitrogen protection subsystem, set the initial nitrogen mixing ratio and total flow rate according to the welding material type and the heat input parameters; activate the local environmental cavity and the multi-jet execution module to isolate the welding area from the atmosphere, and confirm through the atmosphere monitoring unit that the oxygen content in the isolated area has dropped below the set threshold. S3. Closed-loop adaptive control: The welding power supply is started, the welding torch moves along the planned path and welding begins; during the welding process, the following closed-loop control flow is executed cyclically: S31. The multi-sensor fusion monitoring subsystem collects and uploads real-time data on the shape, atmosphere, thermal field and location of the welding area. S32. The central intelligent control subsystem inputs the real-time data into a preset welding quality prediction model and calculates the predicted quality index under the current parameters. S33. With the goal of achieving optimal weld formation and performance, the optimal combination of process parameters for the next control cycle is solved by an optimization algorithm. The parameters include at least welding power, welding speed, and nitrogen mixing ratio. S34. The optimized parameter instructions are synchronously sent to the welding power source and the intelligent nitrogen protection subsystem to adjust the welding process. S4. Post-weld evaluation and data archiving: After welding is completed, the weld is scanned to obtain the weld forming parameters; the sensor data, control commands and result data of the entire welding process are linked and stored to generate a digital archive of welding quality.

2. The process according to claim 1, characterized in that, The intelligent nitrogen protection subsystem includes a nitrogen source, a dynamic mixing module, a multi-channel jet execution module, and a local environmental cavity. The inlet of the dynamic mixing module is connected to a nitrogen source and at least one auxiliary inert gas source, and its outlet is connected to the multi-channel jet execution module through a pipeline. The dynamic mixing module includes a mass flow controller for accurately controlling the mixing ratio and an online analyzer for feedback of gas composition. The multi-jet actuation module includes multiple miniature electromagnetic valve-controlled nozzles arranged around the welding torch head and on the back of the weld. The local environmental cavity is a flexible, sealed cover that can move with the welding torch; The intelligent nitrogen protection subsystem has at least two controlled operating states: First protection state: The dynamic mixing module outputs gas according to a preset basic ratio, and all nozzles of the multi-jet execution module spray out at a basic flow rate to establish and maintain the initial protective environment before welding. Second protection state: According to the instructions in step S33, the dynamic mixing module adjusts the mixing ratio in real time, and the selected nozzle in the multi-jet execution module adjusts its opening and flow rate according to the position of the molten pool and the distribution of the heat field, for dynamic local strengthening protection during the welding process.

3. The process according to claim 1, characterized in that, The multi-sensor fusion monitoring subsystem includes: The welding morphology vision module is a ring laser 3D scanner used to acquire high-precision 3D point cloud data of weld width, misalignment, weld width and reinforcement height before, during and after welding. A local atmosphere analysis module is integrated into the local environment cavity to monitor the oxygen content, nitrogen content, and humidity at key locations within the cavity in real time. The molten pool thermal field monitoring module includes an infrared thermal imager, used to collect temperature field distribution data of the weld pool and its heat-affected zone; The weld seam tracking module is used to identify the offset between the welding torch and the weld seam center in real time. The central intelligent control subsystem receives multi-source heterogeneous data from the above modules and performs spatiotemporal synchronization and fusion processing.

4. The process according to claim 1, characterized in that, The welding quality prediction model mentioned in step S32 is a multi-physics coupling model based on a deep neural network; The input layer feature vector of this model includes at least: real-time weld width and misalignment, oxygen content of the protective atmosphere, maximum temperature of the molten pool and temperature gradient, as well as the current welding power and speed; The model is trained using historical process data to establish a nonlinear mapping relationship between the input features and welding quality evaluation indicators; the welding quality evaluation indicators are the predicted penetration depth, porosity, and nitrogen loss rate. The central intelligent control subsystem makes optimization decisions based on the output of the model using a model predictive control algorithm.

5. The process according to claim 1, characterized in that, The optimal combination of process parameters described in step S33 is solved by an optimization algorithm, specifically by using the gradient descent method for online iteration. Its optimization objective function is: min J = α·(Dt - Dp)² + β·Pp + γ·|N| Where Dt is the target melting depth, Dp is the model-predicted melting depth, Pp is the model-predicted porosity, N is the nitrogen loss rate calculated by the model, and α, β, and γ are weighting coefficients set according to product requirements. In each control cycle, the algorithm starts with the current process parameters, calculates the partial derivatives of the objective function with respect to each controllable parameter, including welding power, welding speed, and nitrogen ratio, and updates the parameters along the gradient descent direction until it finds the parameter combination that minimizes the objective function, which is then output as the optimal command.

6. The process according to claim 1, characterized in that, The welding process in step S3 also includes a defect early warning sub-step based on acoustic emission signals: Acoustic emission signals during the welding process are collected using acoustic emission sensors; Wavelet packet transform analysis is performed on the acoustic emission signal to extract signal energy features within a specific frequency band, which is 100kHz-300kHz. When the energy characteristic continuously exceeds the preset threshold, it is determined that the welding process is prone to spatter or microcracks. The central intelligent control subsystem then triggers a level one alarm and calls upon a preset set of emergency parameters for intervention, including reducing power and adjusting gas flow.

7. The process according to claim 1, characterized in that, The auxiliary inert gas used in the intelligent nitrogen protection subsystem is helium or argon; when the welding material is high-nitrogen stainless steel, the initial nitrogen ratio is set to 95%-99.5%; during the welding process, if the local atmosphere analysis module detects a continuous decrease in nitrogen content, or the welding quality prediction model outputs a high nitrogen loss rate prediction value, the central intelligent control subsystem dynamically instructs the dynamic mixing module to increase the nitrogen ratio by 1%-3%.

8. The process according to claim 1, characterized in that, Following step S4, step S5 is also included: calibration verification and model update; Specifically, this involves periodically using standard test pieces to verify welding, conducting non-destructive testing and mechanical property tests on the welded test pieces, and obtaining actual welding quality data. The actual data is compared with the data predicted by the model during the welding process, and the prediction error is calculated. If the error exceeds the allowable range, the complete data chain generated in this verification is used as a new sample. The complete data chain includes sensor data, process parameters, and actual results. The internal weight parameters of the welding quality prediction model are fine-tuned through the backpropagation algorithm to achieve self-learning and accuracy evolution of the model.

9. A heat exchanger welding system based on intelligent nitrogen protection, characterized in that, include: The intelligent nitrogen protection subsystem is used to provide a protective gas with dynamically adjustable composition and flow rate, and to establish a local protective environment in the welding area; The multi-sensor fusion monitoring subsystem is used to collect real-time data on the shape, atmosphere, thermal field, and location of the welding area. The central intelligent control subsystem is used to receive data from the multi-sensor fusion monitoring subsystem, calculate predicted quality indicators through a preset welding quality prediction model, and solve for the optimal combination of process parameters through an optimization algorithm. A welding execution unit is used to perform welding operations according to the instructions of the central intelligent control subsystem; The central intelligent control subsystem is electrically connected to the intelligent nitrogen protection subsystem, the multi-sensor fusion monitoring subsystem, and the welding execution unit, respectively. The central intelligent control subsystem stores an executable program that, when run, automatically executes the following process flows: path pre-scanning and modeling, establishing a dynamic protection environment, closed-loop adaptive control, post-weld evaluation and data archiving; or may also include calibration verification and model updating.

10. The system according to claim 9, characterized in that, The system also includes a remote interaction interface; Through this interface, the system can receive welding task files, material parameter library updates, or model algorithm update packages sent by the host computer or cloud platform. Meanwhile, the system can encrypt and upload welding process data and the generated welding quality digital archives for remote monitoring, big data analysis, or supply chain quality traceability.

Images