Pipeline all-position automatic TIG welding method

Abstract

The invention relates to the field of welding process control, and discloses a pipeline all-position automatic TIG (Tungsten Inert Gas) welding method which comprises the following steps: collecting pipeline geometric parameters, welding position angles and material attribute data in real time through multi-source data fusion; constructing a Gaussian process regression dynamic response model, coupling a nonlinear mapping relationship among the process parameters, the molten pool morphology and the corrosion tendency index, and dynamically adjusting the weight of the model based on the welding position angle; a hierarchical strategy of Bayesian optimization and model prediction control is adopted to generate a parameter solution set meeting the fusion depth constraint and the corrosion threshold value, and the molten pool oscillation frequency is used as feedback to correct the current in real time; the heat accumulation evolution trend is predicted through a hidden Markov chain, and parameter closed-loop migration and trajectory compensation are achieved in combination with molten pool flow field coupling correction. The problems of uneven forming quality and corrosion risk caused by space displacement, dissimilar metal interface effect and dynamic disturbance in all-position welding are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of welding process control, and particularly to an all-position automatic TIG welding method for pipelines. Background Art

[0002] In the fields of petroleum, chemical industry, nuclear power, etc., all-position automatic TIG welding of pipelines is the core process to achieve high-quality connection of dissimilar metal circumferential welds. However, due to the continuous spatial variation of the welding position angle (0° - 360°), the molten pool behavior is affected by the coupling of gravity distribution, heat accumulation gradient, and dissimilar metal interface effect. The traditional welding process faces the following technical bottlenecks:

[0003] Existing methods mostly rely on static process parameter libraries or empirical rules, and it is difficult to dynamically adapt to the molten pool flow and thermal state at different spatial positions, resulting in forming defects such as insufficient penetration in the overhead welding area and excessive reinforcement in the flat welding area;

[0004] At the same time, for the corrosion tendency of dissimilar metal welding, the existing technology lacks a dynamic correlation model between process parameters and corrosion risk, and it is impossible to quantitatively and suppress the electrochemical corrosion driving force in real time during welding, resulting in a decrease in the corrosion resistance of the weld interface;

[0005] In addition, traditional closed-loop control strategies usually take a single feedback variable (such as weld width or arc voltage) as the adjustment basis, and it is difficult to synergistically optimize the forming quality, efficiency, and corrosion resistance, resulting in the risk of process instability under complex working conditions.

[0006] The above problems seriously restrict the quality consistency and long-term service reliability of all-position pipeline welding, and there is an urgent need to develop an integrated solution that combines dynamic modeling, multi-objective decision-making, and physical constraint closed-loop control. Summary of the Invention

[0007] Aiming at the deficiencies of the existing technology, the present invention provides an all-position automatic TIG welding method for pipelines, which solves the problems of uneven forming quality, uncontrollable corrosion risk, and process instability caused by spatial position changes, dissimilar metal interface effects, and dynamic disturbances in the existing all-position pipeline welding process.

[0008] To achieve the above objectives, the present invention is realized through the following technical solutions: an all-position automatic TIG welding method for pipelines, including the following steps:

[0009] Collect geometric parameters of the pipeline, welding position angle, and attribute data of the welding head;

[0010] Build a dynamic response model, and predict the penetration, weld width, reinforcement, and corrosion risk by real-time associating welding parameters with weld forming indexes and corrosion tendency indexes;

[0011] Perform multi-objective optimization based on the dynamic response model to generate welding parameters that meet the requirements of target forming quality and corrosion resistance;

[0012] Control the welding head to perform welding according to the optimized welding parameters, and perform closed-loop feedback control based on the real-time collected welding state data to dynamically adjust the welding parameters.

[0013] Preferably, the pipeline geometric parameters include the curvature radius and wall thickness, the welding position angle is an angular value from 0° to 360° based on the pipeline axis, and the attribute data of the welding head includes corrosion resistance, fracture toughness, and wire type.

[0014] Preferably, the dynamic response model is constructed in the following way:

[0015] The input variables of the dynamic response model include the pipeline curvature radius, welding position angle, welding current, welding speed, arc length, and the electrochemical potential difference between dissimilar metals. The pipeline curvature radius is the bending radius of the welded section of the pipeline, the welding position angle is an angular value from 0° to 360° based on the pipeline axis, and the arc length is the vertical distance between the tungsten electrode and the workpiece surface;

[0016] The output responses of the dynamic response model include penetration depth, weld width, reinforcement height, and corrosion tendency index. The penetration depth is the maximum penetration depth of the weld cross-section, the weld width is the maximum width of the weld surface, the reinforcement height is the height by which the weld surface exceeds the base metal, and the corrosion tendency index is calculated based on the electrochemical potential difference between dissimilar metals, welding current, welding speed, and molten pool temperature, and is used to characterize the corrosion risk of the weld interface;

[0017] The dynamic response model establishes a non-linear mapping relationship between welding parameters and output responses through the Gaussian process regression algorithm, and adjusts the model weights in real time based on the change of the welding position angle.

[0018] Preferably, the real-time adjustment of the dynamic response model is achieved in the following way:

[0019] Periodically collect the molten pool morphology and temperature data during the welding process, and dynamically update the hyperparameters of the Gaussian process regression model through the recursive least squares algorithm;

[0020] The update of the hyperparameters is based on the error feedback between the input variables and the measured output response values;

[0021] The molten pool morphology data extracts the molten pool geometric parameters through image processing, and is used to compare with the model prediction values to correct the prediction deviation of the dynamic response model.

[0022] Preferably, the steps of performing multi-objective optimization based on the dynamic response model include:

[0023] In the rough optimization stage, based on the Bayesian optimization algorithm, a global search is conducted on the combination of welding current and welding speed to generate a Pareto solution set including penetration depth, bead width, reinforcement height, and corrosion tendency index. Parameter combinations that satisfy the penetration depth not being less than 90% of the target penetration depth and the corrosion tendency index being lower than the preset threshold are selected as candidate solutions;

[0024] In the fine optimization stage, based on the model predictive control algorithm, real-time correction is performed on the candidate solutions. Taking the molten pool oscillation frequency as the feedback variable, the welding current is dynamically adjusted through proportional-integral control, where the molten pool oscillation frequency is obtained through spectral analysis of the arc voltage fluctuation signal.

[0025] Preferably, the calculation formula for the corrosion tendency index is:

[0026]

[0027] where C is the corrosion tendency index; A weld is the weld cross-sectional area, which is calculated in real time through the molten pool morphology image; I is the welding current; v is the welding speed; R gas is the gas constant; ΔΦ is the electrochemical potential difference between dissimilar metals, which is obtained through the pre-stored values in the material database; T pool is the molten pool temperature, which is collected in real time by an infrared thermometer.

[0028] Preferably, the step of performing closed-loop feedback control based on the real-time collected welding state data to dynamically adjust the welding parameters includes:

[0029] Predict the initial welding parameters for the next welding position angle based on the hidden Markov chain. The state transition probability of the hidden Markov chain is constrained by the heat accumulation gradient, where the heat accumulation gradient is calculated from the difference in heat input between adjacent position angles and the angle increment;

[0030] Dynamically adjust the welding current according to the real-time collected molten pool oscillation frequency. The molten pool oscillation frequency is obtained through fast Fourier transform analysis of the arc voltage fluctuation signal, and the adjustment process satisfies the proportional-integral control law;

[0031] The heat accumulation gradient constraint is used to limit the parameter migration amplitude between adjacent welding position angles, avoiding weld formation defects caused by sudden changes in local heat accumulation.

[0032] Preferably, the calculation formula for the heat accumulation gradient is:

[0033]

[0034] where, represents the cumulative heat input of the current welding position angle θ, I(τ) is the instantaneous value of the welding current, U(τ) is the instantaneous value of the arc voltage, and Δθ is the angle increment;

[0035] The state transition probability of the hidden Markov chain is negatively correlated with the matching degree of the heat accumulation gradient, which is specifically expressed as:

[0036]

[0037] where λ is the heat gradient weight coefficient, which is calibrated through welding experiments; and respectively represent the heat accumulation gradients of the current position angle θ i and the next position angle θ i+1 ; ||·|| 2 represents the Euclidean norm.

[0038] Preferably, the step of performing closed-loop feedback control based on the real-time collected welding state data to dynamically adjust the welding parameters further includes:

[0039] Dynamically correcting the welding trajectory based on the molten pool flow velocity field, which is calculated by coupling the hydrodynamic equation and the electrochemical potential difference, and its control equation is:

[0040]

[0041] where u is the molten pool flow velocity vector; ρ pool is the molten pool density; v pool is the molten pool dynamic viscosity; σ pool is the molten pool conductivity; is the electric potential gradient; ΔΦ is the electrochemical potential difference between dissimilar metals;

[0042] The correction amount of the welding trajectory is proportional to the lateral component of the flow velocity field, and the correction direction is perpendicular to the welding progress direction.

[0043] The present invention also provides a pipeline all-position automatic TIG welding control device, including:

[0044] A data acquisition module for obtaining pipeline geometric parameters, welding position angles, and welding head attributes;

[0045] A dynamic modeling module for constructing and updating a multi-physical field coupling dynamic response model;

[0046] An optimization control module for performing multi-objective optimization and closed-loop parameter adjustment;

[0047] A welding execution module for driving the welding head to complete welding according to the optimized parameters.

[0048] The present invention provides a pipeline all-position automatic TIG welding method. It has the following beneficial effects:

[0049] 1. Through the spatial weight adjustment mechanism of multi-source heterogeneous data fusion and dynamic response model, the present invention overcomes the forming quality fluctuations caused by differences in gravity distribution and heat accumulation at different positions (flat welding, vertical welding, overhead welding) of the pipeline circumferential weld in traditional welding processes, realizes the adaptive matching of welding parameters with the position angle, and significantly improves the process stability and forming consistency of all-position welding.

[0050] 2. Based on the multi-objective hierarchical optimization strategy, the present invention incorporates both molten pool morphology parameters and corrosion tendency index into the optimization objectives, breaks through the technical bottleneck that it is difficult to balance interface strength and corrosion resistance simultaneously in dissimilar metal welding by traditional single-objective optimization methods, and achieves the Pareto optimal balance of quality-performance through the collaboration of Bayesian global search and model prediction local correction.

[0051] 3. By using the heat accumulation gradient constraint of the hidden Markov chain and the coupled correction mechanism of the molten pool flow field, the present invention can predict the evolution trend of the welding thermal cycle in real time and compensate for the trajectory deviation caused by the molten pool flow, effectively suppressing the weld forming defects caused by dynamic disturbances such as sudden changes in the thermal physical properties of the base material and fluctuations in the shielding gas, and improving the process robustness under complex working conditions.

[0052] 4. Through the dynamic modeling and closed-loop feedback of the corrosion tendency index, the present invention transforms the passive anti-corrosion mode of traditional post-welding inspection into online risk warning and parameter adjustment during the welding process, blocks the accumulation of the driving force of electrochemical corrosion in advance, and reduces the risks of intergranular corrosion and stress corrosion cracking at the dissimilar metal welding interface. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 is a schematic flow chart of the method of the present invention;

[0054] Figure 2 is a schematic structural diagram of the control device of the present invention.

[0055] Among them, 10 is the data acquisition module; 20 is the dynamic modeling module; 30 is the optimization control module; 40 is the welding execution module. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0056] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0057] Please refer to the attached Figure 1, the present invention provides a full-position automatic TIG welding method for pipelines, which realizes the collaborative control of the forming quality and corrosion resistance of full-position pipeline welding by integrating dynamic response modeling, multi-objective optimization, and closed-loop feedback control. The implementation manner of the technical solution is described in detail below in combination with specific steps.

[0058] As Figure 1 shown, the full-position automatic TIG welding method for pipelines may include the following steps:

[0059] S1. Collect the geometric parameters of the pipeline, the welding position angle, and the attribute data of the welding head;

[0060] S2. Build a dynamic response model, and predict the penetration depth, bead width, reinforcement height, and corrosion risk by real-time associating welding parameters with weld forming indexes and corrosion tendency indexes;

[0061] S3. Perform multi-objective optimization based on the dynamic response model to generate welding parameters that meet the requirements of target forming quality and corrosion resistance;

[0062] S4. Control the welding head to perform welding according to the optimized welding parameters, and perform closed-loop feedback control based on the real-time collected welding state data to dynamically adjust the welding parameters.

[0063] The following is a detailed description of each step in the method of the present invention, and the specific implementation principles, technical details, and processes of each step are comprehensively elaborated.

[0064] For step S1, in this embodiment, the data collection step is realized by multi-source heterogeneous sensor fusion technology, which specifically includes the acquisition and preprocessing of the geometric parameters of the pipeline, the welding position angle, and the attribute data of the welding head, providing basic data support for subsequent dynamic modeling and optimization control.

[0065] The acquisition objects of the pipeline geometric parameters cover the curvature radius R curv and wall thickness T w of the welded section of the pipeline. Among them, the curvature radius R curv is defined as the bending radius of the welded section of the pipeline, which is used to characterize the influence of local geometric deformation of the pipeline on welding heat conduction and molten pool flow; the wall thickness T w is the thickness of the pipeline base material, which directly affects the threshold range of the welding heat input.

[0066] Preferably, the curvature radius is calculated by performing three-dimensional point cloud reconstruction on the outer surface of the pipeline using a laser scanner, and the wall thickness is continuously measured along the welding path using an ultrasonic thickness gauge.

[0067] The acquisition of the welding position angle θ is based on the pipe axis as the spatial reference, defined as the angular value of the welding head's circumferential position relative to the axis reference plane along the pipe, with a value range of 0° to 360°. Among them, 0° corresponds to the top position of the pipe, and 180° corresponds to the bottom position.

[0068] Preferably, the welding position angle is obtained in real time through a high-precision encoder. The encoder is installed on the rotating shaft of the welding actuator and outputs an angular signal synchronously with the circumferential movement of the welding head along the pipe. The real-time acquisition of this angular value can provide a spatial position basis for the weight adjustment of position-related parameters (such as the gravity distribution of the molten pool and the arc deflection characteristics) in the dynamic response model.

[0069] The attribute data of the welding head includes the corrosion resistance grade, fracture toughness index of the welding wire, and the welding wire model, which are used to quantify the compatibility and interfacial behavior between the welding material and the base material.

[0070] Preferably, the corrosion resistance grade is calculated based on the contents of Cr, Mo, and Ni elements in the welding wire composition database. The fracture toughness index is mapped to a dimensionless score value through the pre-stored welding wire tensile test data (such as elongation and reduction of area). The acquisition of the welding wire model is achieved by scanning the embedded RFID tag of the welding head or by matching the corresponding physical and chemical parameters from the cloud material database according to the welding process document.

[0071] Furthermore, the data acquisition step also includes the relative pose parameters between the welding head and the pipe, such as the arc length L arc , that is, the vertical distance from the tip of the tungsten electrode to the surface of the workpiece.

[0072] Preferably, the arc length is jointly calculated through the displacement feedback signal of the welding head servo motor and the preset tungsten electrode extension length. Its dynamic change amount will be used as a key input variable for the subsequent dynamic response model to correct the arc heat flux density distribution.

[0073] To ensure data reliability, a sliding time window filtering algorithm is introduced in the acquisition step to denoise the original signal. Preferably, for low-frequency slowly varying parameters such as the radius of curvature R curv and the wall thickness T w , mean filtering is used to eliminate measurement jitter; for high-frequency dynamic parameters such as the welding position angle θ and the arc length L arc , the Kalman filter is used to fuse multi-sensor data to suppress random noise interference.

[0074] For step S2, in this embodiment, the construction of the dynamic response model is achieved through multi-physics field coupling modeling and real-time data-driven technology, aiming to establish a non-linear mapping relationship between welding parameters, weld forming quality, and corrosion resistance, providing a prediction basis for the optimal control of all-position welding.

[0075] The input variables of the dynamic response model cover geometric parameters, process parameters, and material property parameters. Among them, the pipe curvature radius R curv characterizes the local bending degree of the welded section of the pipe, and its value is obtained from the laser scanning data in step S1, which is used to correct the distribution characteristics of the arc heat flux on the curved surface geometry; the welding position angle θ is based on the pipe axis and dynamically reflects the spatial position of the welding head in the circumferential direction, providing a position trigger condition for model weight adjustment.

[0076] Preferably, the process parameters include welding current I, welding speed v, and arc length L arc , where the arc length is defined as the perpendicular distance from the tip of the tungsten electrode to the workpiece surface, and is jointly solved through the servo motor displacement feedback and the preset tungsten electrode extension length, which is used to quantify the spatial distribution of the arc energy density. The material property parameters include the electrochemical potential difference ΔΦ between dissimilar metals, and its value is calculated based on the composition data of the welding wire and the base metal in step S1 through the electrode potential difference, which is directly related to the corrosion driving force at the weld interface.

[0077] The output responses of the model include penetration depth D, weld width W, reinforcement height H, and corrosion tendency index C. Among them, the penetration depth D characterizes the maximum penetration depth of the weld cross-section, the weld width W is the maximum width of the weld surface, and the reinforcement height H is the height by which the weld exceeds the base metal. The three jointly quantify the forming quality; the corrosion tendency index C is calculated by the following formula:

[0078]

[0079] where, A weld is the cross-sectional area of the weld, which is extracted in real time through the edge detection algorithm of the molten pool morphology image; R gas is the gas constant, with a value of 8.314 J / (mol·K); T pool is the molten pool temperature, which is collected in real time by an infrared thermometer. This formula quantifies the corrosion risk at the dissimilar metal welding interface by coupling the electrochemical potential difference, heat input, and molten pool morphology parameters.

[0080] In this embodiment, the dynamic response model is based on the Gaussian process regression as the mathematical framework, and its essence is to model the probability distribution from the welding parameter space to the response space. The input variable set of the model is defined as X = [R curv , θ, I, v, L arc , ΔΦ] T , and the output response set is Y = [D, W, H, C] T . The Gaussian process regression assumes that the output response follows a multivariate Gaussian distribution:

[0081]

[0082] where, K(X, X′) is the kernel function matrix, is the noise variance, and I is the identity matrix. Preferably, the kernel function adopts a combination form of a radial basis function and white noise:

[0083]

[0084] In the formula, is the signal variance, l is the length scale, and δ ij is the Kronecker function.

[0085] In this embodiment, the dynamic adjustment of the model weights is achieved through the scale factor of the covariance matrix, and the kernel function parameters are specifically adjusted according to the interval division of the welding position angle θ:

[0086] Interval division: The welding position angle θ is divided into a flat welding area (0° ≤ θ < 90°), a vertical welding area (90° ≤ θ < 270°), and an overhead welding area (270° ≤ θ < 360°).

[0087] Differentiated scale factors are applied to the covariance matrices of the output responses in different intervals. For example, in the overhead welding area (where the gravity sag effect on the molten pool is significant), the covariance weight of the penetration depth D and the reinforcement height H is increased to enhance the sensitivity of the molten pool morphology prediction.

[0088] For each output response y k ∈ {D, W, H, C}, its kernel function parameters are adjusted to:

[0089]

[0090] where w k (θ) is the weight function related to the position angle. Preferably, the weight function is defined by piecewise linear interpolation or an empirical function based on historical data.

[0091] To ensure the prediction accuracy of the model, the hyperparameter update process of the dynamic response model is closely coupled with the real-time data acquisition in step S1.

[0092] Preferably, the molten pool morphology images and infrared temperature data are periodically collected, and the hyperparameters of the Gaussian process regression model are iteratively optimized through the recursive least squares algorithm. The molten pool morphology images are obtained by a laser vision sensor, and the measured values of the penetration depth D, the weld width W, and the reinforcement height H are extracted through an edge detection algorithm. After comparing with the model prediction values, an error signal is generated to drive the hyperparameter update. The error feedback mechanism enables the model to adapt to the dynamic disturbances (such as base metal heat accumulation, shielding gas fluctuations, etc.) during the welding process, ensuring the consistency between the prediction results and the measured data.

[0093] In this embodiment, the hyperparameters (σ f , l, σ n ) of the model are updated in real time through the recursive least squares (RLS) algorithm, and the specific process includes:

[0094] 1. Data stream input: The molten pool morphology image (laser vision sensor) and the molten pool temperature T are collected every 0.1 s pool (infrared thermometer), and the measured output response Y is extracted after preprocessing meas =[D meas ,W meas ,H meas ,C meas T .

[0095] 2. Error calculation: Calculate the residual e of the model prediction value Y pred and the measured value Y meas as e = Y meas -Y pred .

[0096] 3. Hyperparameter update: Update the hyperparameter vector Θ = [σ f ,l,σ n T :

[0097]

[0098] where φ k is the non - linear basis function mapping of the input variable X k , and P k is the covariance matrix, which is iteratively calculated through the inverse covariance update formula.

[0099] For step S3, in this embodiment, the multi - objective optimization step is realized through a hierarchical optimization strategy, combining the global search and local correction mechanisms to generate a welding parameter combination that meets the collaborative requirements of forming quality and corrosion resistance. The optimization process uses the dynamic response model constructed in step S2 as the prediction engine, and solves the Pareto optimal solution set in a two - stage progressive manner of rough optimization and fine optimization, and dynamically corrects the parameters based on real - time feedback data to ensure the process stability of all - position welding.

[0100] In this embodiment, the rough optimization stage aims to globally search for the welding parameter combination through the Bayesian optimization algorithm to generate a Pareto optimal solution set that simultaneously meets the requirements of forming quality and corrosion resistance.

[0101] The parameter space definition and objective function construction are as follows:

[0102] Taking the welding current I (unit: A) and the welding speed v (unit: mm / s) as the optimization variables, a two - dimensional parameter space is defined The boundary values are preset according to the base metal thickness and wire diameter.

[0103] ​​A multi-objective optimization problem is established, and the objective function includes the weighted sum of the penetration depth D, the weld width W, the reinforcement height H, and the corrosion tendency index C:

[0104]

[0105] Among them, D target , W target , H max are the preset forming index thresholds; α, β, γ, δ are weight coefficients, and the priorities are calibrated through the Analytic Hierarchy Process (AHP).

[0106] The Bayesian optimization algorithm is implemented as follows:

[0107] Taking the dynamic response model in step S2 as the surrogate model, a Gaussian process regression relationship from the welding parameters (I, v) to the objective function FF is constructed, and the kernel function uses the Matérn 5 / 2 function:

[0108]

[0109] Among them, σ f is the signal variance, l is the length scale, and it is initialized by maximum likelihood estimation.

[0110] The Expected Improvement (EI) is used as the acquisition function to balance exploration and development:

[0111]

[0112] Among them, is the current optimal objective value. The EI function is maximized through the Differential Evolution algorithm (DE), and new sampling points x new = argmaxEI(x) are iteratively generated until the preset number of iterations (such as 50 times) is reached.

[0113] The Pareto solution set is screened as follows:

[0114] The candidate solutions need to satisfy the penetration depth D ≥ 0.9D target and the corrosion tendency index C ≤ C th .

[0115] The solutions that satisfy the constraints are quickly non-dominated sorted (NSGA-II algorithm), and the Pareto front solution set is screened out and stored in the candidate solution queue.

[0116] In this embodiment, in the fine optimization stage, the candidate solutions are corrected in real time based on the Model Predictive Control (MPC) framework. Taking the molten pool oscillation frequency as the feedback variable, the welding current is dynamically adjusted to suppress the dynamic disturbances during the welding process.

[0117] The model predictive control architecture is as follows:

[0118] Taking the dynamic response model of step S2 as the internal prediction model, establish the dynamic relationship between the welding current I and the molten pool oscillation frequency f osc :

[0119] f osc (k + 1) = f osc (k) + K·(I(k) - I nom ) + ∈(k)

[0120] where K is the dynamic gain coefficient, calibrated through a step response experiment; ∈(k) is the process noise.

[0121] In each control period t, solve the following optimization problem:

[0122]

[0123] where H p is the prediction horizon, H c is the control horizon, and λ is the control weight coefficient.

[0124] The proportional-integral (PI) control law is implemented as follows:

[0125] The molten pool oscillation frequency f osc is obtained through the spectral analysis of the arc voltage signal U(t). Perform a fast Fourier transform (FFT) on U(t) and extract the main frequency component:

[0126]

[0127] where represents the FFT operation, and the frequency band range [f min , f max is preset according to the welding process.

[0128] Adopt the discrete PI control law to dynamically adjust the welding current:

[0129]

[0130] where e(t) = f target - f osc (t), K p and K i are calibrated through the Ziegler-Nichols tuning method, and Δt is the control period.

[0131] The corrosion tendency index constraint is verified as follows:

[0132] During the fine optimization process, calculate the corrosion tendency index C of the current parameter combination in real time. If C > C th , then trigger the re-optimization mechanism:

[0133] Re-optimization trigger condition:

[0134]

[0135] Parameter fallback strategy: Re-select parameter combinations that satisfy C ≤ C th from the candidate solution queue and reset the state of the MPC controller.

[0136] To ensure optimization efficiency, the parallel computing architecture of the Bayesian optimization algorithm is deployed in the edge computing unit, and the asynchronous sampling strategy is used to reduce the iteration delay. Preferably, the time window length of the model predictive control matches the change rate of the welding position angle θ to ensure that parameter correction is completed before the position angle switch. The hierarchical optimization strategy effectively balances the multi-objective conflicts of welding quality, efficiency, and corrosion resistance through the cooperation of global exploration and local development, providing a reliable basis for the process parameter decision-making of all-position welding.

[0137] For step S4, in this embodiment, the closed-loop feedback control step is implemented through a multi-modal sensing data fusion and dynamic parameter migration mechanism, and the welding parameters are dynamically adjusted based on the real-time collected welding state data to suppress the spatial variability and dynamic disturbances during the all-position welding process. The control process is based on the optimized parameters generated in step S3, combined with hidden Markov chain prediction, molten pool oscillation frequency feedback, and molten pool flow field coupling correction, to form a multi-level closed-loop control architecture.

[0138] The core of the closed-loop feedback control lies in the parameter prediction and migration mechanism driven by the hidden Markov chain (HMM). The state space of the hidden Markov chain is defined as the discretized partition of the welding position angle θ (for example, each 10° is a state), and the state transition probability is constrained by the heat accumulation gradient. The heat accumulation gradient is calculated by the difference in heat input between adjacent position angles and the angle increment, and its formula is:

[0139]

[0140] where represents the cumulative heat input of the current position angle θ, Δθ is the angle increment (typical value is 3°), and by restricting the change range of the heat accumulation gradient, the parameter mutation between adjacent welding position angles is avoided, which may cause the molten pool to become unstable. The state transition probability of the hidden Markov chain is negatively correlated with the matching degree of the heat accumulation gradient, specifically expressed as:

[0141]

[0142] where λ is the heat gradient weight coefficient, which is calibrated through welding experiments and is used to quantify the constraint strength of the heat accumulation continuity on the parameter migration. and respectively represent the current position angle θi and the next position angle θ i+1 thermal accumulation gradient; ||·|| 2 represents the Euclidean norm.

[0143] During the real-time control process, the molten pool oscillation frequency f osc as a key feedback variable, is obtained through the fast Fourier transform (FFT) analysis of the arc voltage fluctuation signal, such as the calculation in step S3.

[0144] Based on the deviation between the molten pool oscillation frequency and the target value f target the proportional-integral (PI) control law is adopted to dynamically adjust the welding current I:

[0145]

[0146] where I nom is the nominal current value optimized in step S3, K p and K i are control coefficients, which are tuned through step response experiments. The closed-loop control of the molten pool oscillation frequency is directly related to the stability of the molten pool flow. High-frequency oscillation indicates increased turbulence in the molten pool, and the current needs to be reduced to suppress spatter; low-frequency oscillation indicates insufficient fluidity of the molten pool, and the current needs to be increased to improve wettability.

[0147] Furthermore, the closed-loop feedback control integrates a welding trajectory correction mechanism driven by the molten pool flow velocity field u. The molten pool flow velocity field is calculated by coupling the hydrodynamic equation and the electrochemical potential difference effect, and its control equation is:

[0148]

[0149] where ρ pool , v pool , σ pool are the density, dynamic viscosity, and conductivity of the molten pool, respectively, from the pre-stored molten pool physical property database; is the electric potential gradient, which is calculated in real time by the arc shape model; ΔΦ is the electrochemical potential difference between dissimilar metals, which is obtained through material data matching in step S1. The lateral component u y (the component perpendicular to the welding direction) is used to generate the trajectory correction amount ΔP, and its correction direction is strictly limited to be perpendicular to the welding forward direction. The correction formula is:

[0150] ΔP = k·u y ·n ⊥

[0151] where k is the proportionality coefficient, and n ⊥ is the normal vector of the welding forward direction. The weld offset caused by the molten pool flow is compensated through trajectory correction to ensure the consistency of the weld formation.

[0152] Generally speaking, the present invention realizes precise regulation of the forming quality and corrosion resistance of all-position welding through data-driven dynamic modeling, multi-objective optimization and closed-loop feedback collaborative control. First, multi-source sensor fusion is used to collect pipeline geometric parameters, welding position angles and wire properties data in real time, and an association database between welding parameters and molten pool dynamic response is constructed. Based on this, a Gaussian process regression algorithm is used to establish a dynamic response model, non-linearly map process parameters such as welding current and speed to penetration, bead width, reinforcement and corrosion tendency index, and dynamically update the model weights through the recursive least squares algorithm to adapt to spatial position changes. Subsequently, combining Bayesian optimization and model predictive control, on the basis of globally searching the Pareto solution set that satisfies the penetration and corrosion constraints, the parameters are corrected in real time with the molten pool oscillation frequency as the feedback variable to ensure process stability under dynamic conditions. Finally, through hidden Markov chain prediction, thermal accumulation gradient constraint and molten pool flow field coupling correction, closed-loop migration and trajectory compensation of welding parameters are realized, suppressing forming defects and corrosion risks caused by gravity distribution, sudden change of thermal accumulation and dissimilar metal interface effect in all-position welding, and systematically solving the problem of quality consistency in all-position welding of pipelines.

[0153] The following embodiments take the all-position welding of dissimilar metal circumferential welds of oil pipelines as the application scenario, and specifically illustrate the implementation process of the technical solution of the present invention:

[0154] Welding object: Circumferential weld welding of X65 steel pipeline with an outer diameter of 508 mm and a wall thickness of 12 mm and Inconel 625 nickel-based alloy pipeline, and the welding position angle covers 0° (top) to 360° (bottom).

[0155] Step S1: Data acquisition and preprocessing

[0156] Pipeline geometric parameters: The curvature radius R of the welding section is obtained by a laser scanner curv = 2540 mm, and the wall thickness T is measured by an ultrasonic thickness gauge w = 12 ± 0.2 mm.

[0157] Welding position angle: The six-axis welding robot is equipped with an encoder to record the θ value in real time, with a dynamic range of 0° to 360° and a resolution of 0.1°.

[0158] Wire properties: The wire model is ERNiCrMo-3, and its corrosion resistance grade (PREN = 45) and fracture toughness (elongation rate ≥ 30%) are matched from the material database.

[0159] Arc length: L is calculated through the displacement feedback of the servo motor arc = 3.2 mm, and the data is denoised by Kalman filtering.

[0160] Step S2: Construction and update of dynamic response model

[0161] Input and output definitions:

[0162] Input variables: X = [R curv , θ, I, v, L arc , ΔΦ], where ΔΦ = 0.15V (electrode potential difference between X65 steel and Inconel 625).

[0163] Output responses: penetration depth D, bead width W, reinforcement height H, corrosion tendency index C.

[0164] Model initialization: Train a Gaussian process regression model using historical welding data, with the kernel function being RBF, and the initial hyperparameters σ f = 1.2, l = 0.5.

[0165] Real-time update: Collect infrared images of the molten pool (temperature T pool = 1650 ± 50°C) and topography data every 100 ms, update the hyperparameters through recursive least squares, and increase the weight coefficient in the overhead welding area (θ = 180°) by 30%.

[0166] Step S3: Multi-objective optimization and parameter correction

[0167] Coarse optimization: Bayesian optimization searches the parameter space of I ∈ [150, 220] A and v ∈ [4, 8] mm / s, and filters out the Pareto solution set (I = 180, v = 6, satisfying D = 5.8 mm ≥ 0.9D target , C = 0.72 ≤ C th = 0.8).

[0168] Fine optimization: The MPC controller uses the oscillation frequency f osc = 120 Hz (target value f target = 100 Hz) as feedback, and adjusts the current to I = 185 A through PI control (K p = 0.5, K i = 0.1) to suppress the turbulence of the molten pool.

[0169] Step S4: Closed-loop feedback control and trajectory correction

[0170] Hidden Markov chain prediction: At the position of θ = 90°, calculate the state transition probability based on the heat accumulation gradient Q(θ) = 850, and predict the initial current I = 182 A at the next position angle θ = 95°.

[0171] Molten pool flow correction: Solve the hydrodynamics equation to obtain the transverse flow velocity u y = 0.15 m / s, generate a trajectory correction amount ΔP = 0.3 mm (proportional coefficient k = 2 s / m), and adjust the lateral offset of the welding torch.

[0172] The all-position automatic TIG welding control device described below can be referred to in correspondence with the all-position automatic TIG welding method described above.

[0173] Please refer to the appendix Figure 2 , the present invention also provides an all-position automatic TIG welding control device, including:

[0174] A data acquisition module 10 for obtaining pipeline geometric parameters, welding position angles, and welding head attributes;

[0175] A dynamic modeling module 20 for constructing and updating a multi-physical field coupling dynamic response model;

[0176] An optimization control module 30 for performing multi-objective optimization and closed-loop parameter adjustment;

[0177] A welding execution module 40 for driving the welding head to complete welding according to the optimized parameters.

[0178] The device of this embodiment can be used to execute the method embodiment above. The principle and technical effect are similar and will not be elaborated here.

[0179] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A pipeline full-position automatic TIG welding method, characterized in that: The following steps are involved: Collect pipeline geometry parameters, welding position angles and weld joint attribute data; Build a dynamic response model to predict the depth of penetration, weld width, excess height and corrosion risk by real-time correlation of welding parameters with weld formation indicators and corrosion tendency index; Perform multi-objective optimization based on the dynamic response model to generate welding parameters that meet target forming quality and corrosion resistance requirements; The welding head is controlled to perform welding according to the optimized welding parameters, and closed-loop feedback control is performed according to the welding state data collected in real time to dynamically adjust the welding parameters.

2. The pipeline all-position automatic TIG welding method according to claim 1 is characterized in that: The pipeline geometric parameters include the radius of curvature and the wall thickness, the welding position angle is an angle value of 0° to 360° based on the pipeline axis, and the attribute data of the welding head include corrosion resistance, fracture toughness and welding wire model.

3. The pipeline all-position automatic TIG welding method according to claim 1 is characterized in that: The dynamic response model is constructed in the following way: The input variables of the dynamic response model include pipeline curvature radius, welding position angle, welding current, welding speed, arc length and electrochemical potential difference between dissimilar metals, wherein the pipeline curvature radius is the bending radius of the welding section pipeline, the welding position angle is an angle value of 0° to 360° based on the pipeline axis, and the arc length is the vertical distance between the tungsten electrode and the workpiece surface; The output response of the dynamic response model includes penetration depth, weld width, residual height and corrosion tendency index, wherein the penetration depth is the maximum penetration depth of the weld cross section, the weld width is the maximum width of the weld surface, the residual height is the height of the weld surface exceeding the parent material, and the corrosion tendency index is calculated based on the electrochemical potential difference of dissimilar metals, welding current, welding speed and molten pool temperature, and is used to characterize the corrosion risk of the weld interface; The dynamic response model establishes a nonlinear mapping relationship between welding parameters and output responses through a Gaussian process regression algorithm, and adjusts the model weight in real time based on changes in the welding position angle.

4. The pipeline full-position automatic TIG welding method according to claim 3 is characterized in that: The real-time adjustment of the dynamic response model is achieved by: The molten pool morphology and temperature data during the welding process are collected periodically, and the hyperparameters of the Gaussian process regression model are dynamically updated through the recursive least squares algorithm; The updating of the hyperparameters is based on error feedback between the input variables and the measured output response values; The molten pool morphology data is processed to extract molten pool geometric parameters for comparison with model prediction values ​​to correct the prediction deviation of the dynamic response model.

5. The pipeline all-position automatic TIG welding method according to claim 1 is characterized in that: The step of performing multi-objective optimization based on the dynamic response model comprises: In the rough optimization stage, a global search is performed on the combination of welding current and welding speed based on the Bayesian optimization algorithm to generate a Pareto solution set including penetration depth, weld width, residual height and corrosion tendency index. The parameter combination that satisfies the requirements of penetration depth not less than 90% of the target penetration depth and corrosion tendency index less than the preset threshold is selected as the candidate solution. In the fine optimization stage, the candidate solutions are corrected in real time based on the model predictive control algorithm. The welding current is dynamically adjusted through proportional-integral control with the molten pool oscillation frequency as the feedback variable. The molten pool oscillation frequency is obtained by spectral analysis of the arc voltage fluctuation signal.

6. The pipeline all-position automatic TIG welding method according to claim 5 is characterized in that: The calculation formula of the corrosion tendency index is: Where C is the corrosion tendency index; A weld is the cross-sectional area of ​​the weld, which is calculated in real time through the molten pool morphology image; I is the welding current; v is the welding speed; R gas is the gas constant; ΔΦ is the electrochemical potential difference between dissimilar metals, which is obtained through the pre-stored value in the material database; T pool is the molten pool temperature, which is collected in real time by an infrared thermometer.

7. The pipeline all-position automatic TIG welding method according to claim 1 is characterized in that: The step of performing closed-loop feedback control according to the real-time collected welding state data to dynamically adjust the welding parameters includes: Predicting the initial welding parameters of the next welding position angle based on a hidden Markov chain, wherein the state transition probability of the hidden Markov chain is constrained by a heat accumulation gradient, wherein the heat accumulation gradient is calculated by the difference in heat input between adjacent position angles and the angle increment; Dynamically adjust the welding current according to the real-time collected molten pool oscillation frequency, the molten pool oscillation frequency is obtained by fast Fourier transform analysis of the arc voltage fluctuation signal, and the adjustment process satisfies the proportional-integral control law; The heat accumulation gradient constraint is used to limit the parameter migration amplitude between adjacent welding position angles to avoid weld formation defects caused by sudden changes in local heat accumulation.

8. The pipeline all-position automatic TIG welding method according to claim 7 is characterized in that: The calculation formula of the heat accumulation gradient is: in, It represents the cumulative heat input at the current welding position angle θ, I(τ) is the instantaneous value of welding current, U(τ) is the instantaneous value of arc voltage, and Δθ is the angle increment; The state transition probability of the hidden Markov chain is negatively correlated with the heat accumulation gradient matching degree, which can be specifically expressed as: Among them, λ is the thermal gradient weight coefficient, which is calibrated by welding experiments; and Respectively represent the current position angle θ i and the next position angle θ i+1 The heat accumulation gradient of ||·|| 2 represents the Euclidean norm.

9. The pipeline all-position automatic TIG welding method according to claim 7, characterized in that: The step of performing closed-loop feedback control according to the real-time collected welding state data to dynamically adjust the welding parameters also includes: The welding trajectory is dynamically corrected based on the molten pool flow velocity field, which is calculated by coupling the fluid mechanics equation with the electrochemical potential difference. The control equation is: Where u is the melt pool flow velocity vector; ρ pool is the molten pool density; v pool is the dynamic viscosity of the molten pool; pool is the molten pool conductivity; is the potential gradient; ΔΦ is the electrochemical potential difference between dissimilar metals; The correction amount of the welding trajectory is proportional to the transverse component of the flow velocity field, and the correction direction is perpendicular to the welding forward direction.

10. A pipeline full-position automatic TIG welding control device, used to execute the method according to any one of claims 1 to 9, characterized in that: include: Data acquisition module, used to obtain pipeline geometric parameters, welding position angle and welding head properties; Dynamic modeling module, builds and updates multi-physics coupling dynamic response models; Optimize control module to perform multi-objective optimization and closed-loop parameter adjustment; The welding execution module drives the welding head to complete welding according to the optimized parameters.

Images