Pipeline multilayer multi-pass welding track adaptive planning method

Abstract

The application relates to the technical field of pipeline welding automation, and discloses a pipeline multilayer multi-pass welding track adaptive planning method.The method comprises the following steps: collecting weld track pose, welding electric parameter and temperature information in real time, and constructing a sliding time window data sequence; establishing an interlayer constraint model and a track prediction model based on the sequence, generating a plurality of track deviation predictions through short-term recursive prediction and long-term sequence prediction; adopting a residual reciprocal weighted fusion strategy in combination with a weld forming quality evaluation function to obtain fused track deviation and evaluate uncertainty; superimposing the fused deviation on the original planning track, generating an adaptive correction track through a multi-objective optimization model; and controlling a welding robot to execute the correction track and feeding back and updating in real time.The application solves the problem of insufficient weld forming precision caused by the lack of dynamic correction in traditional static planning through double-time-scale prediction and track-process parameter collaborative optimization, and significantly improves welding quality and efficiency.

Description

Technical Field

[0001] This invention relates to the field of pipeline welding automation and robot trajectory control technology, and in particular to an adaptive planning method for multi-layer, multi-pass pipeline welding trajectories. Background Technology

[0002] With the continuous growth of global energy demand, the construction scale of long-distance and high-pressure pipelines is constantly expanding. The quality of pipeline welding is directly related to the safety and economy of the transportation system. The large wall thickness of high-strength pipeline steel such as X80 in modern pipeline engineering requires multi-layer and multi-pass welding processes to complete the bevel filling. This long-term and multi-layer welding process results in a significant heat accumulation effect. During the welding process, the pipeline will experience radial thermal expansion and bevel angle changes. The fluctuation of the residual height of the previous weld layer will directly affect the starting position of the subsequent weld layer, resulting in a cumulative deviation between the actual weld trajectory and the preset trajectory.

[0003] Currently, the industry mainly uses teach-in programming or offline programming to plan welding trajectories. Teach-in programming requires manual dragging of the welding torch to record trajectory points, resulting in low efficiency, poor accuracy, and an inability to adapt to variations in bevels on-site. Offline programming generates trajectories based on ideal bevel geometry parameters, failing to consider assembly clearance deviations, blunt edge height errors, and welding thermal deformation, leading to significant trajectory deviations during actual execution. More importantly, existing methods are all static planning models, unable to dynamically correct based on real-time data. Once deviations occur, they can only be resolved through manual inspection and rework after welding is complete.

[0004] In recent years, some studies have attempted to introduce visual sensors or laser scanners for weld seam tracking. However, these methods can only passively respond to deviations at the current moment and lack the ability to predict future trajectory deviations. On the other hand, thermal deformation prediction techniques based on finite element analysis are computationally time-consuming and cannot meet the requirements of real-time planning. Furthermore, most existing studies treat trajectory planning and welding process parameters independently, without establishing a coupled optimization mechanism between the two.

[0005] Therefore, there is an urgent need to develop an adaptive planning method for multi-layer, multi-pass welding trajectories in pipelines to fundamentally solve the problems of trajectory deviation accumulation and process parameter mismatch. Summary of the Invention

[0006] To overcome the above shortcomings, this invention provides an adaptive planning method for multi-layer, multi-pass welding trajectories of pipelines, aiming to improve the problem that traditional welding trajectory planning mostly adopts static programming methods, which lacks dynamic correction of real-time deviations, resulting in insufficient weld formation accuracy.

[0007] This invention provides the following technical solution: an adaptive planning method for multi-layer, multi-pass welding trajectories in pipelines, comprising the following steps:

[0008] S1. Real-time acquisition of weld trajectory pose, wire feed speed, welding electrical parameters and temperature information of the welded pipeline; acquisition of point cloud data of the pipeline bevel using a 3D laser scanner; extraction of the bevel opening angle, blunt edge height and circumferential distribution of the butt joint gap; acquisition of interlayer temperature field distribution using an infrared temperature measurement system; measurement of the residual height value of the previous weld layer using a visual sensor; and formation of a sliding time window data sequence from the acquired data according to the time sequence. The length of the sliding time window is adaptively adjusted according to the thermal cycle of the previous weld layer.

[0009] S2. Establish an interlayer constraint model for multi-layer and multi-pass welding of pipelines, including: calculating the theoretical filling cross-sectional area of ​​the current weld layer based on the geometric characteristics of the bevel, determining the trajectory starting offset of the current weld layer based on the measured value of the residual height of the previous weld layer, and establishing the coupling constraint relationship between the interlayer temperature gradient and the welding heat input.

[0010] Based on the sliding time window data sequence and interlayer constraint model, a trajectory prediction model is constructed. Multiple future trajectory deviation prediction results are generated through short-term recursive prediction and long-term sequence prediction. The short-term recursive prediction uses welding current fluctuation, arc offset and molten pool oscillation frequency as input to predict the instantaneous trajectory deviation within 0.5-2 seconds. The long-term sequence prediction predicts the trajectory drift along the circumference of the pipe based on the cumulative heat input and the trend of pipe thermal deformation.

[0011] S3. Integrate the prediction results of the multiple trajectory deviations, and apply the weld formation quality evaluation function: The weld width deviation, excess height deviation and interlayer overlap rate corresponding to each prediction result are calculated. The fusion prediction trajectory deviation is obtained by weighting the residual reciprocal and the uncertainty of the trajectory deviation is evaluated. Interlayer continuity constraints are applied to ensure that the overlap of adjacent welds is in the range of 30%-50%.

[0012] S4. The fused predicted trajectory deviation is superimposed onto the original planned welding trajectory to establish a multi-objective optimization model. The objective function is to minimize the weighted sum of weld width fluctuation, interpass temperature difference, and total welding time. Constraints include welding line energy limitations. Matching relationship between rate and welding speed Inter-floor dwell time constraints Spatial displacement constraints for circumferential welding, combined with welding heat input, structural constraints and multi-layer, multi-pass continuity requirements, and an optimization algorithm is used to generate an adaptive correction trajectory;

[0013] S5. Control the welding robot to perform welding according to the corrected trajectory, monitor the weld width, reinforcement height and fusion state in real time, update the parameter correction coefficients in the interlayer constraint model according to the actual forming results, and collect the execution trajectory and welding parameters in real time, and update the collected information to the sliding time window data sequence.

[0014] Preferably, in step S2, the method for establishing the inter-layer constraint model includes:

[0015] According to the bevel opening angle Blunt edge height and the gap between the joints Calculate the first Theoretical fill cross-sectional area of ​​layer weld and trajectory coverage width requirements ;

[0016] Measurement of the first Actual reinforcement height of layer weld bead Calculate the starting offset of the current weld layer trajectory. ,in This is the amount of fusion compensation;

[0017] Establish interlayer temperature constraints: when At that time, the inter-floor dwell time is forcibly extended. Or reduce the welding speed of the current layer to .

[0018] Preferably, in step S2, the input parameters for the short-term recursive prediction include the welding current fluctuation amplitude. Arc deflection angle and molten pool oscillation frequency The LSTM neural network is used to predict the trajectory deviation in the X, Y, and Z directions within the next 0.5-2 seconds.

[0019] The long-term sequence prediction is based on a thermo-mechanical coupled finite element model. The cumulative linear energy distribution and welding sequence are input, the thermal deformation displacement field in the 0-360° direction of the pipe circumference is calculated, and the displacement component of the weld centerline is extracted as the long-term trajectory drift prediction value.

[0020] Preferably, in step S4, the constraints of the multi-objective optimization model include:

[0021] Welding line energy constraint: ;

[0022] Rate-welding speed matching relationship: ;

[0023] Inter-floor dwell time constraints: ;

[0024] Circumferential welding spatial displacement constraints: considering the kinematic limitations of pipe rotation or welding torch posture adjustment;

[0025] Multi-layer, multi-pass continuity constraint: overlap between adjacent weld passes satisfy .

[0026] Preferably, the real-time acquisition includes:

[0027] Multi-point pose sampling of welded pipe seams is performed, including collecting the coordinates of the weld centerline and the normal information of the pipe surface;

[0028] The wire feed speed is measured using an encoder or optical sensor, and the acquisition frequency is consistent with the update frequency of the sliding time window.

[0029] Welding electrical parameters are collected using voltage and current sensors, and weld temperature information is collected using thermocouples or infrared sensors.

[0030] The collected data are sorted by timestamp to form a unified data structure, and then organized into a continuous sequence using a sliding time window to facilitate input into the subsequent trajectory prediction model.

[0031] Preferably, the construction of the trajectory prediction model includes:

[0032] Normalization and missing value imputation are performed on the sliding time window data sequence to ensure the integrity of the input data;

[0033] The normalized data is input into the short-term recursive prediction model, and multi-step prediction is performed based on historical small window data to output the short-term trajectory deviation sequence at several future sampling times.

[0034] The same sequence is input into the long-term sequence prediction model, and trend features are extracted based on the entire sliding time window sequence to output the long-term trajectory deviation sequence;

[0035] The short-term and long-term forecast results are time-aligned and fused according to weighting coefficients, which are dynamically adjusted by the historical residual size and uncertainty index, to form multiple future trajectory deviation forecast results.

[0036] Preferably, the fusion prediction trajectory deviation includes:

[0037] The residuals of the deviations of multiple predicted trajectories are calculated separately and compared with the actual historical trajectories to obtain the residual sequence.

[0038] The fusion trajectory bias is calculated by weighting the residual sequence inversely, and the fusion value at each time point is obtained by weighted averaging of the residuals of each prediction result;

[0039] The variance of each predicted trajectory at each time node is calculated to form a trajectory uncertainty sequence. The corrected trajectory is then adjusted by weighting according to the uncertainty sequence, with the correction magnitude increased for points in high uncertainty areas and the original correction value maintained for points in low uncertainty areas.

[0040] The fused trajectory deviation and uncertainty indicators are stored as an updatable data structure for use in the next step of trajectory generation correction.

[0041] Preferably, the generation of the adaptive correction trajectory includes:

[0042] The deviation of the fused predicted trajectory is superimposed on the original planned trajectory to form a preliminary corrected trajectory;

[0043] Based on the initial trajectory correction, and in conjunction with welding heat input limitations, the welding speed and wire feed rate for each weld bead segment are constrained and calculated.

[0044] Check the spatial spacing between multiple weld layers to ensure that the weld layers are continuous and do not overlap. Sequentially mark the multiple weld trajectory segments and check the connection between adjacent trajectory endpoints.

[0045] Combining pipeline geometry constraints and multi-layer, multi-pass continuity requirements, the spacing between trajectory points is adjusted to meet continuity and heat input constraints. An optimization algorithm is used to locally adjust the trajectory points, generating the final executable corrected trajectory. The corrected trajectory is output as a continuous point sequence, with each point containing its position, welding parameters, and corresponding time sequence.

[0046] Preferably, the corrected trajectory welding process includes:

[0047] The robot is controlled to move point by point according to the final corrected trajectory, with each trajectory point corresponding to welding parameter instructions;

[0048] Real-time acquisition of robot joint pose and end effector position, and correction of trajectory deviation through sensor feedback;

[0049] The collected information is compared with the corrected trajectory to generate a trajectory deviation vector, and the sliding time window data sequence is updated for the next round of prediction.

[0050] The core technical concept of this invention lies in:

[0051] (1) Establish a multi-source data acquisition system that integrates bevel geometry, welding heat input and interlayer temperature field, and construct an interlayer constraint model to quantitatively calculate the filling volume, excess height compensation and heat input coupling relationship;

[0052] (2) Construct a dual time-scale trajectory prediction system: short-term prediction is used to deal with instantaneous disturbances in the welding process (current fluctuations, arc deviation), and long-term prediction is used to deal with circumferential drift caused by heat accumulation (based on thermo-mechanical coupling analysis).

[0053] (3) The trajectory deviation is integrated with the weld formation quality as the guide, and the trajectory deviation is directly related to the weld width, reinforcement height and interlayer overlap rate to ensure the weldability of the corrected trajectory.

[0054] (4) Achieve multi-objective collaborative optimization of trajectory correction and welding process parameters (line energy, wire feeding matching, interpass temperature difference, spatial displacement), and update the prediction model through welding execution feedback.

[0055] Compared with existing technologies, the innovation of this invention lies in the fact that it establishes an interlayer constraint model for multi-layer and multi-pass welding of pipelines for the first time, proposes a trajectory prediction fusion strategy guided by welding process, and realizes the collaborative optimization of trajectory and process parameters.

[0056] The present invention has the following beneficial effects:

[0057] By combining dual-timescale prediction with interlayer constraint model, trajectory deviations caused by heat accumulation can be predicted in advance. Compared with traditional static planning methods, this significantly improves trajectory planning accuracy and weld formation stability, and reduces weld width fluctuations significantly.

[0058] By using a weld formation quality evaluation function to weightedly fuse multiple predicted trajectories, the overlap rate of adjacent welds is maintained at a reasonable rate, the interlayer non-fusion rate is significantly reduced, and the welding qualification rate is significantly improved.

[0059] A multi-objective optimization model was established, which includes weld width fluctuation, interpass temperature difference, welding time, trajectory deviation and heat input constraints. While correcting the trajectory, the model automatically adjusts process parameters such as welding speed and wire feed rate, realizing the coordinated optimization of trajectory and process parameters, and avoiding the loss of linear energy or abnormal interpass temperature caused by trajectory changes.

[0060] Through the rolling update mechanism of parameter correction coefficients and adaptive adjustment of thermal cycle, the system can learn and optimize model parameters based on the actual execution results of the previous weld layer, and the prediction error gradually converges, enhancing the system's adaptability and adapting to different pipe specifications and welding conditions.

[0061] The trajectory planning time is significantly shortened compared to traditional teaching programming, the welding rework rate is significantly reduced, and production efficiency is effectively improved while reducing labor costs and production cycle. Attached Figure Description

[0062] Figure 1 This is a flowchart of an adaptive planning method for multi-layer, multi-pass welding trajectories in pipelines proposed in this invention.

[0063] Figure 2 This is a flowchart illustrating the data acquisition and sliding time window construction process of an adaptive planning method for multi-layer, multi-pass welding trajectories in pipelines proposed in this invention.

[0064] Figure 3 This is a flowchart illustrating the trajectory prediction model construction of an adaptive planning method for multi-layer, multi-pass welding trajectories in pipelines proposed in this invention.

[0065] Figure 4 This is a flowchart of trajectory deviation fusion and uncertainty evaluation for an adaptive planning method for multi-layer, multi-pass welding trajectories of pipelines proposed in this invention.

[0066] Figure 5This is a flowchart of the adaptive correction trajectory generation process for an adaptive planning method for multi-layer, multi-pass welding trajectories in pipelines proposed in this invention. Detailed Implementation

[0067] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0068] Example 1:

[0069] In a first embodiment of the present invention, the present invention provides an adaptive planning method for multi-layer, multi-pass welding trajectories in pipelines, such as... Figures 1-5 As shown, it includes the following steps:

[0070] S1. Real-time acquisition of weld trajectory pose, wire feed speed, welding electrical parameters and temperature information of the welded pipeline, and forming a sliding time window data sequence according to the time series of the acquired data;

[0071] Furthermore, real-time data collection includes:

[0072] Multi-point pose sampling of welded pipe seams is performed, including collecting the coordinates of the weld centerline and the normal information of the pipe surface;

[0073] The wire feed speed is measured using an encoder or optical sensor, and the acquisition frequency is consistent with the update frequency of the sliding time window.

[0074] Welding electrical parameters are collected using voltage and current sensors, and weld temperature information is collected using thermocouples or infrared sensors.

[0075] The collected data are sorted by timestamp to form a unified data structure, and then organized into a continuous sequence using a sliding time window to facilitate input into the subsequent trajectory prediction model.

[0076] Specifically, weld pose acquisition involves multi-point pose sampling of the weld seam in the welded pipeline, acquiring the coordinates of the weld centerline and the normal information of the pipeline surface. The set of pose sampling points is denoted as... Each sampling point Includes weld centerline coordinates and the corresponding pipe surface normal vector The acquisition of coordinate and normal information is accomplished using a 3D position sensor or laser scanning device, and the output raw data structure is as follows: .

[0077] The wire feed rate is acquired by measuring the wire feed rate using an encoder or optical sensor, and the output is a time series. ,in For a moment The corresponding wire speed. The acquisition frequency is consistent with the sliding time window update frequency to ensure data continuity and the integrity of the prediction model input.

[0078] Welding electrical parameter acquisition: Welding electrical parameters are acquired through voltage and current sensors, and the output is a sequence. ,in For welding voltage, The current is the welding current. The electrical parameter acquisition equipment should ensure synchronization with the weld position acquisition, and the data timestamps should be consistent.

[0079] Weld temperature acquisition involves collecting weld temperature information using thermocouples or infrared sensors to form a temperature sequence. ,in For the welding time The instantaneous temperature value. Temperature data is acquired synchronously with pose and electrical parameters to ensure data integrity.

[0080] The sliding time window data sequence construction sorts various types of collected data according to timestamps to form a unified data structure. The data is organized using a sliding time window, with a time window length of [value missing]. The update step size is 1 / 2 each time the slide is 1 / 2. This forms the continuous time series input matrix: ;in For the first The input data matrix for each time window is used as the input for the subsequent trajectory prediction model.

[0081] The sliding time window length is adaptively adjusted to address the dynamic changes in heat accumulation effects during multi-layer, multi-pass welding of pipelines. This invention employs an adaptive window length adjustment strategy based on the thermal cycle period of the previous weld layer. Thermal cycle period. Defined as the time required from the start of welding heating to cooling to the specified interpass temperature, it is a key parameter characterizing the welding thermal process.

[0082] In a specific implementation, the method for measuring the thermal cycle period is as follows: During the welding process, the temperature distribution of the weld and heat-affected zone is continuously monitored using an infrared temperature measurement system, and the welding start time t_heating and the interpass temperature drop are recorded. (The temperature is 150-200℃, depending on the material) The calculation yields:

[0083] ;

[0084] For multi-pass welding, take the average thermal cycle period of all weld passes in that layer:

[0085] ;

[0086] in For the first Total number of weld passes.

[0087] Sliding time window length Adaptive adjustment based on the actual measured thermal cycle period of the previous weld layer:

[0088] ;

[0089] Where β is the adjustment coefficient, typically ranging from 1.2 to 1.5. The settings ensure that the time window covers the complete thermal cycle process and the subsequent thermal accumulation effect, enabling the long-term prediction model to fully capture the development trend of pipeline thermal deformation. The following factors should be considered when setting the value:

[0090] (1) Pipe wall thickness: The greater the wall thickness, the longer the heat conduction time. Take the larger value (1.4-1.5).

[0091] (2) Environmental conditions: forced cooling A smaller value (1.2-1.3) can be taken.

[0092] (3) Number of weld passes: When welding multiple passes Use 1.3-1.4 to ensure multiple heat accumulations are captured.

[0093] Compared with a fixed window length, the adaptive strategy has the following advantages: (1) It dynamically adapts to changes in welding thermal state, avoiding the problem of excessively long windows (insufficient data) in the first layer or excessively short windows (missing heat accumulation information) in subsequent layers; (2) It adapts to different pipe specifications, materials and welding process parameters, improving the versatility of the prediction model; (3) It automatically compensates for the influence of external factors such as environmental heat dissipation and auxiliary cooling on the thermal cycle.

[0094] Example: For an X80 pipeline steel pipe with a diameter of φ610×10mm, the initial time window length is set to [value missing] during the first layer of root pass welding. (Based on empirical values). After the first layer is completed, the temperature is measured using temperature monitoring. (Due to better heat dissipation at the bottom layer and longer thermal circulation), the system automatically adjusts the time window length of the second layer to... The second layer has three filler welds, and the average thermal cycle period was measured. (Heat accumulation accelerates cooling), therefore the length of the third layer time window is adjusted to... Through this layer-by-layer adaptive adjustment, the system can consistently maintain a match between the time window and the actual thermal process, ensuring the accuracy of long-term predictions. Compared to a fixed window length scheme, the adaptive scheme reduces trajectory prediction errors by approximately 15-20%.

[0095] Data output and input process, data acquisition module output matrix To the trajectory prediction module. The data update process is as follows: every time new time data is collected... Add it to the sliding time window, while discarding the earliest time. This ensures the sliding window length remains constant. The input / output formula can be expressed as: .

[0096] Through the above implementation, step S1 forms a complete input data structure and sliding time window sequence, providing continuous, synchronous, and multi-parameter input for subsequent S2 trajectory prediction, S3 deviation fusion, and S4 trajectory correction. This ensures that the spatial, temporal, and parameter dimensions of each collected feature are fully disclosed, and clarifies the data input / output process and formula parameters.

[0097] S2. Construct a trajectory prediction model based on sliding time window data sequence, and generate multiple future trajectory deviation prediction results through short-term recursive prediction and long-term sequence prediction.

[0098] Furthermore, constructing the trajectory prediction model includes:

[0099] Normalization and missing value imputation are performed on the sliding time window data sequence to ensure the integrity of the input data;

[0100] The normalized data is input into the short-term recursive prediction model, and the trajectory deviation prediction for several future sampling times is output through the multi-parameter sequence collected in the previous step.

[0101] By inputting the same sequence into a long-term sequence prediction model, the trajectory trend over a long period of time can be obtained through time series pattern recognition.

[0102] The short-term and long-term forecasts are time-aligned to generate multiple future trajectory deviation forecasts.

[0103] The construction of the trajectory prediction model further includes:

[0104] The short-term recursive prediction model performs multi-step predictions based on historical small window data and outputs a short-term trajectory deviation sequence.

[0105] Long-term series prediction models extract trend features based on the entire sliding time window sequence and output a long-term trajectory deviation sequence;

[0106] The short-term and long-term forecast results are merged according to weighting coefficients, which are dynamically adjusted by the size of historical residuals and uncertainty indicators.

[0107] Specifically, the input data is preprocessed to obtain the sliding time window data sequence from step S1:

[0108] ;

[0109] in Indicates the current timestamp. The length of the sliding time window. The coordinates of the weld center are... Normal to the pipe surface The wire feed speed, and These are welding current and voltage, respectively. This refers to the weld temperature.

[0110] Normalize the sequence: ;in This represents any input parameter. and These represent the historical minimum and maximum values, respectively. Missing values ​​are handled using linear interpolation or forward imputation to ensure data continuity and integrity.

[0111] Short-term recursive prediction model, input normalized sliding time window sequence :

[0112] ;

[0113] in Indicates the number of short-term forecast steps. For short-term trajectory deviation prediction sequences, This represents a recursive prediction function that recursively calculates the deviation of the future trajectory using historical data from a small window.

[0114] Short-term recursive prediction models rely on the output of each step as the input of the next step to achieve multi-step prediction:

[0115] ;

[0116] in , The output of the previous prediction step. This is the current input data.

[0117] Long-term series prediction models, for the entire sliding time window series Extracting time series trend features: where For long-term prediction of steps, This is a long-term trajectory deviation sequence. The overall trajectory change trend is captured by sequence pattern recognition.

[0118] By fusing short-term and long-term forecast results and aligning the short-term and long-term forecast sequences over time, multiple future trajectory deviation prediction results are generated. : ;in and These are the short-term and long-term forecast weighting coefficients, derived from historical residuals. and uncertainty indicators Dynamic calculation: ;in and These represent the uncertainty standards for short-term and long-term forecast sequences, respectively, ensuring automatic weight adjustment and improving forecast reliability.

[0119] Data output and process, generating fused trajectory deviation prediction results The data is output to the trajectory fusion and adaptive correction module in the next step, S3, and can also be fed back to update the sliding time window data sequence.

[0120] ;

[0121] This data update ensures that the prediction model can perform closed-loop iterations during continuous data collection, enabling adaptive trajectory planning.

[0122] Example of establishing an inter-layer constraint model:

[0123] Mathematical expression of the inter-layer constraint model: The inter-layer constraint model established in this invention includes three core elements: theoretical fill cross-sectional area... trajectory starting offset And interlayer temperature gradient constraints.

[0124] (1) Calculation of the theoretical infill cross-sectional area A_n:

[0125] For multi-layer, multi-pass welding of V-groove pipes, the cross-sectional area of ​​the nth layer is calculated based on the groove geometry parameters. The cross-sectional area of ​​the first layer (bottom layer) is: ;in For the gap between the joints, The height of the blunt edge. The bevel opening angle.

[0126] The calculation of the fill cross-sectional area of ​​the nth layer (n ≥ 2) needs to take into account the residual height of the previous layer and the fill height of the new weld layer: ;

[0127] in For the first Weld bead width For the first Layer weld coverage width, For the first Layer fill height. Determined based on bevel shape and filling strategy:

[0128]

[0129] in For the first The cumulative fill depth after the layer welding is completed. This is the accumulated depth of the previous layer.

[0130] Track coverage width Based on lap ratio requirements and weld bead quantity allocation:

[0131] ;

[0132] in The width of the bevel crown. Plan the number of weld passes for this layer. This refers to the overlap coefficient (30-50% overlap between adjacent weld passes).

[0133] (2) Calculation of trajectory initial offset ΔZ_n:

[0134] The initial offset of the trajectory is used to compensate for the excess height deviation of the previous layer, ensuring good fusion between the new weld layer and the previous layer.

[0135] ;

[0136] in For the first The actual height of the floor is measured (by visual sensors or laser scanning). This is the standard excess height value (generally 2-3mm for pipe welding). This is the amount of fusion compensation.

[0137] Fusion compensation amount The determination takes into account the relationship between welding current and penetration depth:

[0138] ;

[0139] in The welding current (in amperes) is given by this empirical formula based on arc physics and molten pool behavior. Higher welding currents result in deeper penetration, requiring greater compensation to avoid incomplete fusion defects. The value range is generally 0.2-0.5mm.

[0140] (3) Interlayer temperature gradient constraint:

[0141] Excessive interpass temperature can lead to coarsening of the microstructure in the heat-affected zone, the formation of hardened structures, or hot cracks. Interpass temperature threshold. Based on the martensitic transformation temperature of the material And the process specifications are determined. For X80 pipeline steel, ≈ 380℃, considering a safety margin, Use 280-350℃.

[0142] When measured When this happens, the system will take the following mandatory measures:

[0143] Option 1: Extend the inter-floor interval: ;

[0144] in This is a time extension factor, determined based on the pipe's heat dissipation rate, with a typical value of 2-3 s / ℃. For example... , Then it will be forcibly extended. .

[0145] Option 2: Reduce welding speed:

[0146] ;

[0147] This formula is based on the square root relationship between heat input and temperature. Decreasing speed reduces heat input per unit length, thus controlling heat accumulation. For example... , , ,but:

[0148] ;

[0149] Option 3: Combination Strategy

[0150] when At the same time, the interval time is extended and the welding speed is reduced to control the interpass temperature in a dual manner.

[0151] Through the synergistic effect of these three constraint models, the system can ensure weld formation quality and microstructure properties while guaranteeing welding efficiency.

[0152] Taking the second layer of filler weld as an example:

[0153] The bevel opening angle was measured by laser scanning. = 60° ± 1.5°, blunt edge height Gap between the joints ;

[0154] Calculate the theoretical infill cross-sectional area of ​​the second layer. Requires trajectory coverage width ;

[0155] Measure the actual remaining height of the first floor (Standard value 2.5mm), calculate the initial offset of the trajectory:

[0156] Temperature when the first layer is completed After 180 seconds, the interlayer spacing decreased. ;

[0157] Based on temperature constraints, the maximum welding speed for the second layer is... (Limited linear energy not exceeding 1.5 kJ / mm).

[0158] Example of dual-scale prediction:

[0159] Short-term forecast (starting from level 2):

[0160] Input: Current fluctuations in the first 10 seconds Arc offset angle ;

[0161] LSTM model output: X-direction deviation +0.3 mm, Y-direction -0.1 mm, Z-direction +0.2 mm within the next 1 second.

[0162] Long-term forecast (second layer, full circumference):

[0163] Finite element model calculation results: radial expansion of 0 mm at 0° position, 0.8 mm at 90° position, 1.2 mm at 180° position, and 0.7 mm at 270° position (showing an elliptic trend).

[0164] S3. Integrate the prediction results of multiple trajectory deviations, calculate the fused predicted trajectory deviation by weighting the residuals, and evaluate the uncertainty of the trajectory deviation;

[0165] Furthermore, the fusion prediction trajectory bias includes:

[0166] The residuals of the deviations of multiple predicted trajectories are calculated separately and compared with the actual historical trajectories to obtain the residual sequence.

[0167] The fusion trajectory bias is calculated by weighting the residual sequence inversely, and the fusion value at each time point is obtained by weighted averaging of the residuals of each prediction result;

[0168] During the fusion process, the variance of the deviation of each node is calculated simultaneously to form a trajectory deviation uncertainty index, which provides data input for adaptive trajectory correction.

[0169] The fused trajectory deviation and uncertainty indicators are stored as an updatable data structure for use in the next step of trajectory generation correction.

[0170] The bias in the fusion prediction trajectory further includes:

[0171] Calculate the variance of each predicted trajectory at each time point to form a trajectory uncertainty sequence;

[0172] The correction trajectory is weighted and adjusted based on the uncertainty sequence, with the correction magnitude increased for points in high uncertainty areas and the original correction value maintained for points in low uncertainty areas.

[0173] The uncertainty sequence and the corrected trajectory are updated synchronously for closed-loop iteration.

[0174] Specifically, the input data structure consists of short-term and long-term trajectory deviation prediction result sets generated in step S2: ;in Indicates the sequence number of different prediction models (e.g., short-term recursive model, long-term sequence model, hybrid model, etc.); For the first The model in the first The trajectory deviation vector predicted at each time point; This represents the total number of nodes within the time window. The deviation from the historical true trajectory is represented as: ;in These are the actual coordinates of the welding locations collected historically. To plan the trajectory coordinates.

[0175] Residual calculation: For each predicted trajectory deviation, calculate the residual between it and the actual trajectory deviation.

[0176] ;

[0177] in Indicates the first The predicted trajectory in the first The residual scalar values ​​at each time point are used for subsequent weighting. The residual sequence is represented as: ;

[0178] The residual inverse weighted fusion calculation involves weighting and fusing multiple prediction results based on the residual sequence. The fused prediction value at each time point is expressed as: The weights are defined as the reciprocals of the residuals: ; To prevent the stability constant from having a denominator of zero, this weighting method numerically assigns higher weights to predictions with smaller errors and lower weights to predictions with larger errors.

[0179] Uncertainty index calculation for each time point Calculate the output variance of all prediction models at this node to measure the uncertainty of trajectory deviation prediction:

[0180] ;in Indicates time node The trajectory deviation variance value is used to reflect the differences between different models.

[0181] The degree of divergence is predicted by a trajectory uncertainty sequence composed of the variances at each time point:

[0182] .

[0183] To ensure that the trajectory deviation fusion process balances prediction accuracy and weld formation quality, this invention introduces a weld formation quality evaluation function Q. This function comprehensively considers three key forming indicators: weld width deviation, reinforcement height deviation, and interlayer overlap rate, and is defined as follows:

[0184] ;

[0185] in:

[0186] This refers to the actual measured weld width. The target weld width is determined based on the bevel geometry and filler requirements.

[0187] This represents the actual measured weld bead height. This is the standard allowance value; for multi-layer, multi-pass welding of pipelines, it is generally taken as 2-3 mm.

[0188] The actual overlap rate between adjacent weld passes is defined as the ratio of the overlap width to the weld pass width. For standard overlap ratios, use 0.3-0.5 to ensure good fusion;

[0189] , , Let be the weighting coefficient, satisfying The parameters are set according to the importance of each parameter in the welding process.

[0190] For multi-layer, multi-pass welding of pipelines, the typical weighting is set as follows: , , .in The weld width has the highest weight because it directly affects the filling efficiency and interlayer continuity. Secondly, because the excess height affects the starting position and fusion quality of subsequent weld layers; Ensure that the overlap rate meets the process requirements.

[0191] During the fusion process, the system calculates the quality evaluation function value corresponding to each predicted trajectory. ( = 1, 2, …,N), will With residual Combined to form a comprehensive weight:

[0192]

[0193] in This is the numerically stable term (taken as 0.01). This is an uncertainty indicator. The formula reflects a triple weighting strategy: (1) residual reciprocal weighting, where the smaller the residual, the higher the weight; (2) quality-oriented weighting, The smaller (the better the quality), the higher the weight; (3) Uncertainty adjustment: the weight of predictions with high uncertainty is reduced.

[0194] The final fusion weights are obtained after normalization:

[0195] ;

[0196] The trajectory deviation after fusion is:

[0197] ;

[0198] This quality-oriented fusion strategy ensures that the corrected trajectory meets the weld formation quality requirements while minimizing prediction deviation, avoiding defects such as large width fluctuations, uneven reinforcement height, or unqualified lap ratio.

[0199] Example illustration: During the fusion process of the second layer filler weld, the short-term prediction results were measured. = 9.2 mm = 2.3 mm = 0.42, corresponding to Long-term forecast results were obtained = 10.1 mm = 2.6 mm = 0.38, corresponding to Combining residuals = 0.3 mm = 0.5 mm and uncertainty = 0.1、 = 0.15, the calculated comprehensive weight =0.64、 = 0.36, the final fusion trajectory is more biased towards the short-term prediction results, which effectively ensures the quality of weld formation.

[0200] Uncertainty-based trajectory correction adjustment involves weighted correction of the fused trajectory deviation based on the uncertainty sequence. Let the original fused trajectory deviation be... The corrected trajectory deviation is The calculation method is as follows: ;in, The adjustment coefficient is used to enhance the correction magnitude in areas of high uncertainty and maintain stable correction in areas of low uncertainty. The value can be set based on empirical rules or a fixed threshold.

[0201] Data output and closed-loop structure: The final output data includes: fused predicted trajectory deviation sequence.

[0202] ;

[0203] Uncertainty indicator series: The two sets of data are stored in a unified data structure and used as input for trajectory correction in step S4. Simultaneously, residuals and uncertainty indicators are fused to update the weight distribution within the sliding time window, enabling adaptive adjustments when subsequent window data is input into the prediction model.

[0204] The data output process involves calculating the residual sequences for each model. Next, the residual sequences are weighted inversely and fused to generate a more reliable trajectory deviation. Simultaneously, the variance is calculated based on the prediction results of each model to obtain an uncertainty sequence reflecting the reliability of the predictions. Finally, the fused trajectory deviation is combined with uncertainty information, weighted and corrected, and the adjusted deviation value is output. And pass it on to the next step.

[0205] S4. The fusion prediction trajectory deviation is superimposed on the original planned welding trajectory, and an adaptive correction trajectory is generated by using an optimization algorithm, taking into account welding heat input, structural constraints and multi-layer and multi-pass continuity requirements.

[0206] Furthermore, generating the adaptive correction trajectory includes:

[0207] The deviation of the fused predicted trajectory is superimposed on the original planned trajectory to form a preliminary corrected trajectory;

[0208] Based on the initial trajectory correction, and in conjunction with welding heat input limitations, the welding speed and wire feed rate for each weld bead segment are constrained and calculated.

[0209] Combining pipeline geometry constraints and multi-layer, multi-channel continuity requirements, optimization algorithms are used to locally adjust trajectory points, generating the final executable corrected trajectory;

[0210] The corrected trajectory is output as a continuous point sequence, with each point containing the position, welding parameters, and corresponding time sequence.

[0211] Generating an adaptive correction trajectory further includes:

[0212] Check the spatial spacing between multiple weld layers to ensure that the weld layers are continuous and do not overlap.

[0213] Sequential marking and connection checks of adjacent trajectory endpoints are performed on multiple welding trajectory segments to ensure that the trajectory continuity meets the weld structure requirements;

[0214] The spacing between trajectory points is adjusted to meet the requirements of continuity and thermal input constraints, and an executable trajectory sequence is formed.

[0215] Specifically, in step S4, after the fusion prediction trajectory deviation is superimposed on the original planned welding trajectory, the welding path can be quantitatively corrected at the spatial coordinate level.

[0216] Let the original planned trajectory be: ;

[0217] The fusion prediction trajectory bias is: ;

[0218] The initial corrected trajectory is defined as follows:

[0219] ;

[0220] in Indicates the first The position coordinates of a trajectory point in a three-dimensional coordinate system. This is the time series corresponding to this point. This refers to the pose deviation correction amount output by the fusion prediction module. This superposition process unifies the trajectory errors between the sensing and planning ends within the same spatial reference frame for correction.

[0221] The generation of the adaptive correction trajectory must be combined with welding heat input constraints. The welding heat input is calculated using: ;in Heat input per unit length (J / mm), For arc thermal efficiency, Arc voltage (V), The welding current (A) is used. The welding speed is measured in mm / s. The system operates based on the allowable heat input range provided by the process database. During the optimization phase, the welding speed corresponding to each weld segment was adjusted. and wire feed rate Perform constraint calculations.

[0222] The constraints are expressed as follows: In the optimization solution, by adjusting and The range of values ​​is used to ensure that the thermal input constraints of the trajectory segment are satisfied.

[0223] To address structural constraints and the requirements for multi-layer, multi-pass continuity, the system is based on the number of weld layers. and the number of weld passes per layer Establish spatial topology mapping: ;in The first Layer The spatial location of each weld joint in polar coordinates. System check of interlayer spacing conditions: And check the transverse spacing between weld beads: If the above conditions are not met, then the coordinates of adjacent trajectory points will be locally offset and corrected during the optimization algorithm iteration.

[0224] A multi-objective optimization model is established, as the trajectory correction process needs to simultaneously consider multiple objectives such as weld quality, heat input control, and welding efficiency. The multi-objective optimization model established in this invention aims to minimize weld width fluctuation, interpass temperature difference, and total welding time, while simultaneously constraining trajectory offset and structural parameters. The complete objective function is defined as follows:

[0225] ;

[0226] The definitions of each item are as follows:

[0227] (1) Weld width fluctuation item :

[0228] Weld bead width consistency directly affects filling efficiency and appearance quality, and is defined as:

[0229] ;

[0230] in For the first Predicted weld width corresponding to each trajectory point This represents the average weld width of this layer. This represents the total number of trajectory points. Minimizing this term ensures that the weld bead width is uniformly distributed along the circumference.

[0231] (2) Interlayer temperature difference term :

[0232] Uneven interlayer temperature distribution leads to microstructure inhomogeneity and residual stress concentration, defined as:

[0233] ;

[0234] in, Number of weld passes in this layer and The first The highest and lowest temperatures of the layer when the weld is completed. This parameter minimizes the uniformity of the temperature field within the same layer.

[0235] (3) Total welding time item :

[0236] Improving efficiency by shortening the total welding time while ensuring quality is defined as:

[0237] ;

[0238] in For the first The length of the segment trajectory, For the corresponding welding speed, For the first Inter-level dwell time. This item includes both movement time and waiting time.

[0239] (4) Trajectory offset :

[0240] The corrected trajectory should be as close as possible to the original planned trajectory, avoiding excessive adjustments, and is defined as follows: ;

[0241] in } is the first The adjusted trajectory point positions This is the initial planned location. The adjustment range for this control trajectory.

[0242] (5) Heat input constraint terms :

[0243] To prevent excessive line energy, a soft constraint term is added:

[0244] ;

[0245] in For the first Linear energy of the segment and This represents the upper and lower limits of the linear energy. This term acts as a penalty, increasing the objective function value when the linear energy exceeds the limits.

[0246] Weighting coefficient settings:

[0247] Weighting coefficient , , , , Based on the welding process priority setting, Σ α_i = 1 must be satisfied. A typical weight allocation is as follows:

[0248] =0.30 (consistency of weld width is the most important).

[0249] =0.25 (Temperature control ensures tissue properties);

[0250] =0.15 (for efficiency reasons, the weight is relatively low);

[0251] =0.20 (Trajectory smoothness, to avoid drastic adjustments);

[0252] =0.10 (hot input soft constraint).

[0253] Constraints:

[0254] The optimization process must meet the following hard constraints:

[0255] (1) Welding line energy constraint: For X80 steel, generally ;

[0256] (2) Wire feed rate matching: ,in The density of the deposited metal, For welding wire density, To fill the cross-sectional area, This refers to the cross-sectional area of ​​the welding wire;

[0257] (3) Inter-floor dwell time: Ensure the previous layer is cooled to a safe temperature;

[0258] (4) Spatial displacement constraints: Considering the robot joint angle limitations Angular velocity limit ;

[0259] (5) Multi-layer, multi-pass continuity constraint: overlap between adjacent weld passes satisfy Ensure good fusion and no unfilled areas.

[0260] Through this multi-objective collaborative optimization strategy, the system can achieve a balance between quality, efficiency, and executability, generating a corrected trajectory that meets engineering realities.

[0261] The optimization algorithm employs a constrained least squares-based local trajectory adjustment model. Let the final set of trajectory points be: The objective function to be optimized is:

[0262] ;

[0263] in This indicates the hot input constraint penalty term. This represents the structural continuity penalty term. These are the weighting coefficients. The algorithm solves the problem using gradient descent or a quasi-Newton method to obtain the final executable corrected trajectory that satisfies the thermal input and structural constraints. In the output phase, the corrected trajectory is output as a continuous point sequence: Each point contains spatial location, welding speed, wire feed rate, heat input value, and time sequence information. This output sequence is directly input to the motion control module to achieve continuous trajectory execution at the end of the welding torch.

[0264] Data input / output process: Input: Original planned trajectory data Prediction bias data Process constraint parameters (including) ), structural parameters (including Processing: Execute trajectory overlay, constraint detection, local optimization, and trajectory continuity check. Output: Generate an executable modified trajectory that satisfies the constraints. The data is stored in the trajectory buffer of the control terminal for use by the motion module.

[0265] S5. Control the welding robot to perform welding according to the corrected trajectory, and collect the execution trajectory and welding parameters in real time, and update the collected information to the sliding time window data sequence;

[0266] Furthermore, the correction trajectory for welding includes:

[0267] The robot is controlled to move point by point according to the final corrected trajectory, with each trajectory point corresponding to welding parameter instructions;

[0268] Real-time acquisition of robot joint pose and end effector position, and correction of trajectory deviation through sensor feedback;

[0269] The collected information is compared with the corrected trajectory to generate a trajectory deviation vector, and the sliding time window data sequence is updated for the next round of prediction.

[0270] Specifically, in step S5, the welding robot is controlled to perform the welding operation according to the corrected trajectory. The system calls the corrected trajectory sequence output in step S4:

[0271] ;

[0272] This sequence serves as the input to the robot's motion control module. It includes the end-effector spatial coordinates, welding speed, wire feed rate, heat input, and corresponding timestamp information at each moment.

[0273] The control module executes time-series commands for each point in the trajectory point sequence. Perform real-time interpolation calculations to generate control command vectors: ;in This represents the angle vectors of each joint of the robot. Represents the joint angular velocity vector. For welding current, For welding voltage, This corresponds to the wire feed rate. The control system sends commands to each joint motor via the servo drive unit, enabling the end effector to move point by point in space.

[0274] During trajectory execution, the sensor module collects the robot's joint positions, end-effector pose, and welding parameters in real time, forming a real-time dataset. ;in Indicates the actual angle of the joint. Indicates the actual angular velocity. These are the real-time acquired current, voltage, and wire feed rate, respectively.

[0275] Based on the robot's forward kinematics model: ; Calculate the actual spatial position of the end point The function This represents the mapping from joint space to Cartesian space. The system maps the actual pose to the corresponding points in the corrected trajectory: Perform difference analysis to calculate the trajectory deviation vector: ; Modulus of the deviation vector:

[0276] Used to quantify the current position error.

[0277] During the data processing phase, the system constructs a sliding time window sequence to record trajectory execution data within the most recent time period. Let the time window length be... The sampling period is The number of data points contained within the time window is The sliding time window data sequence is defined as follows:

[0278] ;

[0279] in Stored the most recent The system records trajectory deviation and welding parameter information at each moment. At the arrival of each new sampling period, the system appends the latest sampled data to the end of the sequence and deletes the oldest data entry, thus achieving a sliding update of the time window.

[0280] The update process of this data sequence can be represented as: ;in This indicates that the earliest record within the time window is removed to ensure a constant time window length. The updated sliding time window sequence serves as the input to the next round of trajectory prediction model (corresponding to steps S2 and S3), used to dynamically correct future welding path deviations.

[0281] Data input and output process: Input data: executable correction trajectory data from step S4. Welding power supply control parameters Robot kinematic model and joint limit parameters.

[0282] Processing steps: Execute trajectory command interpolation and motion control; acquire joint and end-effector poses and welding electrical parameters in real time; calculate trajectory deviation vectors and generate deviation sequences; update sliding time window data structure. .

[0283] Output data: Real-time execution trajectory deviation vector Updated sliding time window sequence This serves as the input for the next cycle prediction and trajectory generation module.

[0284] To improve the system's adaptability and robustness, this invention designs a parameter rolling update mechanism based on actual forming results. The system uses actual welding data obtained from visual sensors, laser scanning, and temperature monitoring to perform online corrections on key parameters in the interlayer constraint model, achieving a gradual transition from an empirical model to a data-driven model.

[0285] Definition of parameter correction coefficient:

[0286] (1) Correction factor for filled cross-sectional area :

[0287] This coefficient reflects the deviation between theoretical calculations and actual filling: ;

[0288] in For the first The actual measured cross-sectional area of ​​the filling layer is obtained through laser scanning or visual reconstruction. These are theoretically calculated values ​​based on bevel geometry. The predicted cross-sectional area of ​​the nth layer infill is corrected as follows:

[0289] ;

[0290] This correction compensates for factors not considered in the theoretical model, such as wire deposition efficiency, spatter loss, and penetration depth differences.

[0291] (2) Correction coefficient for residual height :

[0292] This coefficient is used to predict the development trend of residual height deviation: ;

[0293] Used to correct the trajectory starting offset: ;

[0294] This indicates a tendency for the remaining height to be too large, requiring an increase in the offset. This indicates that the offset is too small and the offset needs to be reduced.

[0295] (3) Thermal cycle period correction factor :

[0296] This coefficient reflects the deviation between the actual thermal process and the prediction:

[0297] ;

[0298] Used to adjust the length of the next time window:

[0299] ;

[0300] This modification enables the system to adapt to external factors such as changes in ambient temperature and differences in heat dissipation conditions.

[0301] (4) Temperature threshold correction factor :

[0302] This coefficient is used to adjust the interlayer temperature control based on actual tissue performance test results.

[0303] ;

[0304] If no tissue abnormalities are found in the previous layer (as indicated by hardness testing or metallographic examination), and near Appropriately relax temperature restrictions ( If an anomaly occurs, the restrictions will be tightened. ).

[0305] Parameter correction strategy:

[0306] To avoid the impact of single measurement errors on system stability, the exponentially weighted moving average (EWMA) method is used:

[0307] ;

[0308] Where λ is the forgetting factor, which controls the weight of historical data and the latest data. This indicates that it relies entirely on historical data (without adaptive processing). This indicates complete reliance on the latest data (susceptible to noise). In practical applications... A value of 0.3-0.5 is used to balance stability and adaptability.

[0309] Correction coefficient initialization:

[0310] During the first layer of welding, all correction factors are initialized to 1 (unbiased assumption): .

[0311] Starting from the second layer, the system is updated gradually based on measured data. After 3-5 iterations, the system converges to the actual process conditions.

[0312] Example illustration:

[0313] Taking the welding of a φ610×10mm pipe as an example, after the first layer is completed, laser scanning measures... Theoretical calculations ,calculate This indicates that the actual deposition rate is 8.3% higher than the theoretical value, possibly due to a larger welding wire diameter or a higher deposition efficiency than the standard value.

[0314] After applying the correction factor, the predicted cross-sectional area of ​​the second layer infill was revised from the theoretical value of 50 mm² to... Adjust the welding speed and wire feed rate based on this value. Measurements were taken after the second layer was completed. , Update using EWMA :

[0315] ;

[0316] Used in the third layer As a correction coefficient. After 5 iterations, the correction coefficient stabilized at... The prediction error of the filled cross-sectional area has been reduced from 12% initially to less than 3%.

[0317] The same correction mechanism is applied to residual height, thermal cycling, and temperature control, enabling the system to gradually adapt to actual welding conditions. This parameter self-learning capability allows the invention to be applied to different pipe specifications, materials, and welding processes without requiring recalibration of model parameters for each situation.

[0318] Closed-loop iterative convergence:

[0319] Theoretical analysis and experimental verification show that this parameter correction mechanism has good convergence. Assume the true values ​​of the actual parameters are... The initial deviation is ,go through The deviation after the next iteration is:

[0320] ;

[0321] Forgetting factor At that time, after 5 iterations, the initial deviation decayed to Even with large initial deviations, it can converge quickly. This ensures the system's adaptability and robustness.

[0322] Example 2:

[0323] This invention relates to an automated welding system for long-distance oil and gas pipelines. The welding robot performs multi-layer, multi-pass welding of the pipeline's circumferential weld seam in a field environment. During welding, it must move continuously along the outer circumference of the pipeline, sequentially welding multiple weld layers and passes. Due to the combined effects of welding heat input, assembly errors, weld shrinkage deformation, and external disturbances, the spatial pose of the pipeline weld seam dynamically changes, causing cumulative trajectory deviations when the robot welds along a preset trajectory, affecting weld formation accuracy and inter-layer continuity. Traditional trajectory planning methods are mostly based on static models or single-time measurement data, lacking the ability to dynamically respond to real-time acquired data and predict future trajectory deviations, and cannot predict and correct for thermal deformation trends. Furthermore, under multi-layer, multi-pass continuous welding conditions, there is a coupling relationship between the spatial geometric constraints between weld passes and the welding heat input. Without adaptive constraint control, overlapping or uneven spacing between weld passes can easily occur, leading to weld structure instability. Existing systems also have shortcomings in the reliability of the prediction model and the effective utilization of feedback data, making it difficult to achieve closed-loop adaptive correction of the welding process. To address the aforementioned problems, this invention provides an adaptive planning method for multi-layer, multi-pass welding trajectories in pipelines, the structure of which is as follows: Figure 1 As shown. The specific implementation process of this method is as follows:

[0324] In step S1, by acquiring the weld trajectory pose, wire feed speed, welding electrical parameters, and temperature information in real time, the dynamic data state during the welding process can be obtained, forming a sliding time window data sequence. This sequence is used to describe the time-varying characteristics of the welding process, enabling the subsequent trajectory prediction model to have a time-continuous input.

[0325] In step S2, a trajectory prediction model is constructed using a sliding time window data sequence. By combining short-term recursive prediction and long-term sequence prediction results, it can simultaneously reflect both short-term dynamic disturbances and long-term trend drifts in welding. This dual-scale prediction structure can cover trajectory deviation changes at different time scales, improving the reliability of the prediction results.

[0326] In step S3, the prediction results of multiple trajectory deviations are fused, and the fused predicted trajectory deviation is generated by weighting the results using the inverse of the residuals. This process enables credibility assessment between the results of different prediction models, suppresses the influence of single model errors, and outputs the uncertainty evaluation results of the deviation to guide subsequent correction decisions.

[0327] In step S4, the deviation of the fused predicted trajectory is superimposed onto the original planned welding trajectory. Based on this, considering welding heat input constraints, structural geometric constraints, and the continuity requirements of multi-layer and multi-pass welding, an adaptive correction trajectory is generated through an optimization algorithm. This trajectory satisfies welding process constraints in both spatial and heat input aspects, forming an executable dynamic trajectory instruction set for robot control.

[0328] In step S5, the welding robot performs welding operations according to the corrected trajectory, collects the execution trajectory and welding parameters in real time, monitors the actual trajectory deviation, and feeds updated data back to the sliding time window data sequence, forming a closed-loop correction mechanism. This closed-loop control enables the system to have continuous adaptive adjustment capabilities, realizing dynamic correction and trajectory self-learning in the welding process.

[0329] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0330] Table 1: Comparison of this invention with traditional static programming methods (X80 pipe multi-layer welding)

[0331] Evaluation indicators Traditional methods Method of the present invention Improvement range Weld bead width variation (mm) ±0.6 ±0.2 Reduced by 67% Interlayer non-fusion rate (%) 8.3 1.2 Reduced by 85% Welding rework rate (%) 12 3 Reduced by 75% Trajectory planning time (minutes) 15 2.3 shorten by 85% Welding pass rate (%) 91 97.8 Increased by 7.5%

[0332] Effect Analysis:

[0333] (1) The weld width fluctuation is significantly reduced: through the interlayer constraint model and dual-scale prediction, it effectively adapts to the geometric deviation of the groove and the thermal accumulation deformation;

[0334] (2) Significant reduction in non-fusion rate: The welding process-oriented fusion strategy ensures an optimal overlap rate of 30-50%;

[0335] (3) Reduced rework rate: The synergistic optimization of the trajectory and process parameters ensures the weldability of the corrected trajectory;

[0336] (4) Improved planning efficiency: Automated planning replaces manual teaching.

Claims

1. An adaptive planning method for multi-layer, multi-pass welding trajectories in pipelines, characterized in that, Includes the following steps: S1. Real-time acquisition of weld trajectory pose, wire feed speed, welding electrical parameters and temperature information of the welded pipeline; acquisition of point cloud data of the pipeline bevel using a 3D laser scanner; extraction of the bevel opening angle, blunt edge height and circumferential distribution of the butt joint gap; acquisition of interlayer temperature field distribution using an infrared temperature measurement system; measurement of the residual height value of the previous weld layer using a visual sensor; and formation of a sliding time window data sequence from the acquired data according to the time sequence. The length of the sliding time window is adaptively adjusted according to the thermal cycle of the previous weld layer. S2. Establish an interlayer constraint model for multi-layer and multi-pass welding of pipelines, including: calculating the theoretical filling cross-sectional area of ​​the current weld layer based on the geometric characteristics of the bevel, determining the trajectory starting offset of the current weld layer based on the measured value of the residual height of the previous weld layer, and establishing the coupling constraint relationship between the interlayer temperature gradient and the welding heat input. Based on the sliding time window data sequence and interlayer constraint model, a trajectory prediction model is constructed. Multiple future trajectory deviation prediction results are generated through short-term recursive prediction and long-term sequence prediction. The short-term recursive prediction uses welding current fluctuation, arc offset and molten pool oscillation frequency as input to predict the instantaneous trajectory deviation within 0.5-2 seconds. The long-term sequence prediction predicts the trajectory drift along the circumference of the pipe based on the cumulative heat input and the trend of pipe thermal deformation. S3. Integrate the prediction results of the multiple trajectory deviations, and apply the weld formation quality evaluation function: The weld width deviation, excess height deviation and interlayer overlap rate corresponding to each prediction result are calculated. The fusion prediction trajectory deviation is obtained by weighting the residual reciprocal and the uncertainty of the trajectory deviation is evaluated. Interlayer continuity constraints are applied to ensure that the overlap of adjacent welds is in the range of 30%-50%. S4. The fused predicted trajectory deviation is superimposed onto the original planned welding trajectory to establish a multi-objective optimization model. The objective function is to minimize the weighted sum of weld width fluctuation, interpass temperature difference, and total welding time. Constraints include welding line energy limitations. Matching relationship between rate and welding speed Inter-floor dwell time constraints Spatial displacement constraints for circumferential welding, combined with welding heat input, structural constraints and multi-layer, multi-pass continuity requirements, and an optimization algorithm is used to generate an adaptive correction trajectory; S5. Control the welding robot to perform welding according to the corrected trajectory, monitor the weld width, reinforcement height and fusion state in real time, update the parameter correction coefficients in the interlayer constraint model according to the actual forming results, and collect the execution trajectory and welding parameters in real time, and update the collected information to the sliding time window data sequence.

2. The adaptive planning method for multi-layer, multi-pass welding trajectory of a pipeline according to claim 1, characterized in that, In step S2, the method for establishing the inter-layer constraint model includes: According to the bevel opening angle Blunt edge height and the gap between the joints Calculate the first Theoretical fill cross-sectional area of ​​layer weld and trajectory coverage width requirements ; Measurement of the first Actual reinforcement height of layer weld bead Calculate the starting offset of the current weld layer trajectory. ,in This is the amount of fusion compensation; Establish interlayer temperature constraints: when At that time, the inter-floor dwell time is forcibly extended. Or reduce the welding speed of the current layer to .

3. The adaptive planning method for multi-layer, multi-pass welding trajectory of a pipeline according to claim 1, characterized in that, In step S2, the input parameters for the short-term recursive prediction include the welding current fluctuation amplitude. Arc deflection angle and molten pool oscillation frequency The LSTM neural network is used to predict the trajectory deviation in the X, Y, and Z directions within the next 0.5-2 seconds. The long-term sequence prediction is based on a thermo-mechanical coupled finite element model. The cumulative linear energy distribution and welding sequence are input, the thermal deformation displacement field in the 0-360° direction of the pipe circumference is calculated, and the displacement component of the weld centerline is extracted as the long-term trajectory drift prediction value.

4. The adaptive planning method for multi-layer, multi-pass welding trajectory of a pipeline according to claim 1, characterized in that, In step S4, the constraints of the multi-objective optimization model include: Welding line energy constraint: ; Rate-welding speed matching relationship: ; Inter-floor dwell time constraints: ; Circumferential welding spatial displacement constraints: considering the kinematic limitations of pipe rotation or welding torch posture adjustment; Multi-layer, multi-pass continuity constraint: overlap between adjacent weld passes satisfy .

5. The adaptive planning method for multi-layer, multi-pass welding trajectory of a pipeline according to claim 1, characterized in that, The real-time data acquisition includes: Multi-point pose sampling of welded pipe seams is performed, including collecting the coordinates of the weld centerline and the normal information of the pipe surface; The wire feed speed is measured using an encoder or optical sensor, and the acquisition frequency is consistent with the update frequency of the sliding time window. Welding electrical parameters are collected using voltage and current sensors, and weld temperature information is collected using thermocouples or infrared sensors. The collected data are sorted by timestamp to form a unified data structure, and then organized into a continuous sequence using a sliding time window to facilitate input into the subsequent trajectory prediction model.

6. The adaptive planning method for multi-layer, multi-pass welding trajectory of a pipeline according to claim 1, characterized in that, The trajectory prediction model includes: Normalization and missing value imputation are performed on the sliding time window data sequence to ensure the integrity of the input data; The normalized data is input into the short-term recursive prediction model, and multi-step prediction is performed based on historical small window data to output the short-term trajectory deviation sequence at several future sampling times. The same sequence is input into the long-term sequence prediction model, and trend features are extracted based on the entire sliding time window sequence to output the long-term trajectory deviation sequence; The short-term and long-term forecast results are time-aligned and fused according to weighting coefficients, which are dynamically adjusted by the historical residual size and uncertainty index, to form multiple future trajectory deviation forecast results.

7. The adaptive planning method for multi-layer, multi-pass welding trajectory of a pipeline according to claim 1, characterized in that, The fusion prediction trajectory deviation includes: The residuals of the deviations of multiple predicted trajectories are calculated separately and compared with the actual historical trajectories to obtain the residual sequence. The fusion trajectory bias is calculated by weighting the residual sequence inversely, and the fusion value at each time point is obtained by weighted averaging of the residuals of each prediction result; The variance of each predicted trajectory at each time node is calculated to form a trajectory uncertainty sequence. The corrected trajectory is then adjusted by weighting according to the uncertainty sequence, with the correction magnitude increased for points in high uncertainty areas and the original correction value maintained for points in low uncertainty areas. The fused trajectory deviation and uncertainty indicators are stored as an updatable data structure for use in the next step of trajectory generation correction.

8. The adaptive planning method for multi-layer, multi-pass welding trajectory of a pipeline according to claim 1, characterized in that, The generation of the adaptive correction trajectory includes: The deviation of the fused predicted trajectory is superimposed on the original planned trajectory to form a preliminary corrected trajectory; Based on the initial trajectory correction, and in conjunction with welding heat input limitations, the welding speed and wire feed rate for each weld bead segment are constrained and calculated. Check the spatial spacing between multiple weld layers to ensure that the weld layers are continuous and do not overlap. Sequentially mark the multiple weld trajectory segments and check the connection between adjacent trajectory endpoints. Combining pipeline geometry constraints and multi-layer, multi-pass continuity requirements, the spacing between trajectory points is adjusted to meet continuity and heat input constraints. An optimization algorithm is used to locally adjust the trajectory points, generating the final executable corrected trajectory. The corrected trajectory is output as a continuous point sequence, with each point containing its position, welding parameters, and corresponding time sequence.

9. The adaptive planning method for multi-layer, multi-pass welding trajectory of a pipeline according to claim 1, characterized in that, The corrected trajectory welding process includes: The robot is controlled to move point by point according to the final corrected trajectory, with each trajectory point corresponding to welding parameter instructions; Real-time acquisition of robot joint pose and end effector position, and correction of trajectory deviation through sensor feedback; The collected information is compared with the corrected trajectory to generate a trajectory deviation vector, and the sliding time window data sequence is updated for the next round of prediction.

Images