An adaptive variable parameter composite oscillation welding control method and related device

Abstract

The application provides a composite swing welding control method with adaptive variable parameters and related equipment, and relates to the technical field of welding. The method comprises the following steps: acquiring real-time multi-physical field sensing data; inputting the real-time multi-physical field sensing data into a simulation model for predicting the behavior of a molten pool, and outputting a molten pool state parameter prediction value in a future preset time period; matching the composite swing trajectory parameters corresponding to the molten pool state parameter prediction value from a preset database; when a welding torch enters a residence area of a weld swing edge according to a macro reference trajectory, performing asymmetric spiral oscillation motion of a welding torch tip in a preset residence area according to a micro modulation trajectory until the welding torch tip leaves the preset residence area. By implementing the method, the problems of edge overheating, undercutting or lack of fusion caused by fixed parameters can be alleviated, and the weld forming quality and the stability of the welding process are improved.

Description

Technical Field

[0001] This application relates to the field of welding technology, and in particular to an adaptive variable parameter composite oscillation welding control method and related equipment. Background Technology

[0002] Welding technology, especially automated welding, is an indispensable key process in modern high-end manufacturing, widely used in aerospace, rail transportation, pressure vessels, marine engineering, and energy equipment. In these fields, the joining of medium and heavy plates is a common operation, and the welding quality directly determines the service performance, reliability, and safe lifespan of the entire structure. To achieve effective filling and fusion of wide welds in a single or a few welding processes, oscillating welding technology has emerged. By controlling the welding torch to periodically oscillate laterally perpendicular to the welding direction, this technology can significantly increase the weld width of a single pass, improve production efficiency, and enhance weld formation. Therefore, how to precisely control the oscillating welding process to ensure a dense, uniform, and defect-free weld has always been the core focus of research and practice in this field.

[0003] In related technologies, control systems typically employ offline programming and fixed parameter execution. Before the welding task begins, process engineers pre-set a set of fixed welding process parameters based on the workpiece material, plate thickness, bevel type, and empirical data. These parameters define the entire motion pattern of the welding torch, such as a constant oscillation frequency, a fixed oscillation amplitude, and a fixed dwell time at the edges of the weld. Throughout the welding process, the welding robot or special-purpose machine strictly follows this pre-set program to drive the welding torch, performing completely repetitive, periodic lateral oscillation movements from the start to the end of the weld. The logical basis of this control strategy is the assumption that welding conditions remain ideal and uniform throughout the process, completing the entire welding task through a set of "unchanging" actions.

[0004] However, welding is inherently a violent, dynamic, and non-equilibrium thermophysical process. As welding continues, a large amount of heat accumulates and is conducted within the workpiece, causing significant dynamic changes in the temperature field of the weld and its surrounding area. This cumulative heat effect makes the size, shape, and flow behavior of the molten pool unstable. When the welding torch moves to the edge of its oscillation stroke, if a preset, symmetrical, and fixed dwell motion is continued, excessive local heat concentration may lead to overheating of the molten pool metal, a surge in fluidity, and defects such as undercut and collapse under gravity. Conversely, if the initial parameters are set conservatively, insufficient heat input may result in incomplete fusion at the edges in the later stages of welding. Summary of the Invention

[0005] This application provides an adaptive variable parameter composite oscillation welding control method and related equipment to address the problem in related technologies that the use of fixed oscillation parameters makes it difficult to adapt to the dynamic thermal effects during the welding process, resulting in welding defects easily generated at the oscillation edge.

[0006] In a first aspect, this application provides an adaptive variable parameter composite oscillation welding control method, applied to a composite oscillation welding control system, the method comprising:

[0007] Acquire real-time multiphysics sensing data, which includes information from at least two heterogeneous sensors used to characterize the geometry of the molten pool, temperature field distribution, and arc stability.

[0008] The real-time multiphysics sensing data is input into a simulation model for predicting the behavior of the molten pool, and the predicted values ​​of the molten pool state parameters within a preset time period are output.

[0009] The composite oscillation trajectory parameters corresponding to the predicted values ​​of the molten pool state parameters are matched from the preset database. The composite oscillation trajectory parameters include macroscopic reference trajectory parameters and microscopic modulation trajectory parameters used to define the composite motion trajectory of the welding torch tip.

[0010] When the welding torch enters the preset dwelling area of ​​the weld swing edge according to the macro reference trajectory defined by the macro reference trajectory parameters, the tip of the welding torch is controlled to perform an asymmetric spiral oscillation motion in the preset dwelling area according to the micro modulation trajectory defined by the micro modulation trajectory parameters until it leaves the preset dwelling area.

[0011] By employing the above technical solution, the system can acquire multiphysics sensing data characterizing the molten pool state and predict future molten pool behavior using simulation models. Based on this prediction, the system can adaptively match composite oscillation trajectory parameters from the database, especially in the oscillation edge dwelling area where heat tends to concentrate. By executing asymmetric spiral oscillation motion, it can finely control local heat input and molten metal flow. This method can effectively alleviate edge overheating, undercut, or insufficient fusion caused by fixed parameters, improving weld formation quality and welding process stability.

[0012] In some embodiments, the step of acquiring real-time multiphysics sensing data specifically includes:

[0013] The surface image data of the molten pool is acquired by a visual sensor, and the molten pool surface image data includes the molten pool outline, surface texture and liquid metal flow characteristics.

[0014] Temperature field data of the welding area is acquired by an infrared thermal imaging sensor. The temperature field data of the welding area includes the center temperature of the molten pool, the temperature gradient distribution, and the range of the heat-affected zone.

[0015] Welding electrical parameter data is monitored by current and voltage sensors, including instantaneous current value, voltage fluctuation amplitude, and arc length change rate.

[0016] The molten pool surface image data, welding area temperature field data, and welding electrical parameter data are processed for time synchronization to generate time-aligned multiphysics sensing data.

[0017] By adopting the above technical solution, the system integrates multiple heterogeneous sensors such as vision, infrared thermal imaging, and electrical parameters to comprehensively perceive the state of the molten pool from multiple dimensions such as geometric shape, temperature distribution, and arc stability. It also performs time synchronization processing on these data to ensure the precise alignment of different physical field information on the time axis, thereby improving the accuracy and reliability of the entire adaptive control system in making predictions and decisions.

[0018] In some embodiments, the step of inputting the real-time multiphysics sensing data into a simulation model for predicting molten pool behavior and outputting predicted values ​​of molten pool state parameters for a future preset time period specifically includes:

[0019] The real-time multiphysics sensing data is subjected to noise reduction preprocessing to generate standardized sensing data, which is normalized data after removing outliers and noise interference.

[0020] Extract the time-series feature vector from the standardized sensor data. The time-series feature vector contains trend information and periodic patterns of dynamic changes in the molten pool.

[0021] The time-series feature vector is input into a deep learning-based molten pool behavior prediction model, which is a neural network model trained with historical welding data.

[0022] The molten pool behavior prediction model calculates the geometric parameters, temperature distribution parameters, and flow state parameters of the molten pool within a preset time period in the future, and generates predicted values ​​of the molten pool state parameters including confidence scores.

[0023] By adopting the above technical solution, the system first preprocesses and extracts temporal features from the sensor data, transforming the raw data into effective information that reflects dynamic trends. Then, a trained deep learning model is used to process these features. This model can learn and master the complex nonlinear laws of the welding process, thereby accurately predicting the future state of the molten pool. Furthermore, the prediction results are accompanied by a confidence score, enabling the system to assess the reliability of the predictions, thus achieving smarter and more robust decision-making, improving the intelligence level of the control strategy and its adaptability to uncertainty.

[0024] In some embodiments, prior to the step of matching the composite oscillation trajectory parameters corresponding to the predicted values ​​of the molten pool state parameters from a preset database, the method further includes:

[0025] A mapping relationship library between welding process parameters and oscillation trajectory parameters is constructed, and the mapping relationship library contains the optimal trajectory parameter combinations under different materials, plate thicknesses and welding positions;

[0026] The trajectory parameter search space is determined based on the mapping relationship library corresponding to the material properties and structural features of the current welding task. The trajectory parameter search space is the range of parameter values ​​that satisfy the process constraints.

[0027] A fuzzy matching algorithm is used to find historical cases within the trajectory parameter search space that are closest to the predicted values ​​of the molten pool state parameters. These historical cases include verified successful welding parameter combinations.

[0028] A set of candidate composite oscillation trajectory parameters is generated based on the historical cases.

[0029] By adopting the above technical solution, the system first constrains the parameter search space based on the current welding task before matching parameters, avoiding blind searching. Then, a fuzzy matching algorithm is used to find successful historical cases within the narrowed space that are closest to the predicted molten pool state. This method does not require exact matching, but rather seeks the "most similar" solution, greatly enhancing the system's generalization ability and adaptability to new working conditions. It improves the efficiency and intelligence of parameter matching, enabling the system to quickly respond and find validated, high-quality control strategies.

[0030] In some embodiments, after the step of controlling the welding torch tip to perform asymmetric helical oscillation motion in the preset dwell area according to the micro-modulation trajectory defined by the micro-modulation trajectory parameters, the method further includes:

[0031] Real-time monitoring of the molten pool width variation within the preset dwell zone, the molten pool width variation reflecting the degree of heat accumulation in the edge region;

[0032] When the change in the width of the molten pool exceeds a preset threshold, the amplitude attenuation coefficient of the spiral oscillation motion is adjusted according to a preset ratio. The amplitude attenuation coefficient is used to control the deceleration rate of the spiral motion radius.

[0033] The dwell time of the welding torch tip in the preset dwell area is adjusted according to the adjusted amplitude attenuation coefficient.

[0034] By adopting the above technical solution, the system introduces a real-time closed-loop fine-tuning mechanism when performing precise helical oscillation motion. By monitoring the actual changes in the width of the molten pool within the dwell zone, the system can directly assess the immediate effect of edge heat accumulation. Once heat accumulation overshoot is detected, the system can immediately adjust the amplitude decay rate and dwell time of the helical oscillation, directly intervening in local heat input. This real-time feedback and correction within the oscillation cycle enables more precise dynamic control of the edge fusion process, further improving the ability to suppress welding defects and the immediacy of control.

[0035] In some embodiments, after the step of controlling the welding torch tip to perform asymmetric helical oscillation motion in the preset dwell area according to the micro-modulation trajectory defined by the micro-modulation trajectory parameters, the method further includes:

[0036] A trajectory optimization feedback mechanism based on reinforcement learning is established, which includes reward function design and policy network update.

[0037] The weld formation quality indicators during the execution of asymmetric helical oscillation motion are collected. The weld formation quality indicators include weld reinforcement uniformity, undercut depth and surface ripple coefficient.

[0038] Calculate the comprehensive reward value for the current oscillation cycle, which is composed of a weighted average of weld formation quality index, heat input uniformity index, and edge fusion state index;

[0039] When the comprehensive reward value is lower than the historical average, the control strategy of the micro-modulation trajectory is updated by the strategy gradient algorithm to generate optimized spiral oscillation parameters.

[0040] The optimized spiral oscillation parameters are applied in real time to the next oscillation cycle.

[0041] By adopting the above technical solution, the system introduces a reinforcement learning-based self-optimization mechanism, enabling it to continuously learn and improve. By evaluating the weld formation quality and calculating the comprehensive reward value after each oscillation cycle, the system can quantify the effectiveness of its control strategy. When the effect is unsatisfactory, it can autonomously update the control strategy of the micro-trajectory using a policy gradient algorithm. This "post-mortem review and continuous iteration" learning mode allows the system to continuously accumulate experience and improve itself, thereby achieving long-term optimization of the welding process and a higher level of intelligence.

[0042] In some embodiments, prior to the step of the welding torch entering the preset dwell zone of the weld oscillation edge according to the macro reference trajectory defined by the macro reference trajectory parameters, the method further includes:

[0043] The spatial boundary of the preset dwelling area is determined based on the geometric features of the weld and the heat accumulation distribution characteristics. The spatial boundary defines the key area where special motion control needs to be performed.

[0044] By analyzing the defect distribution patterns in historical welding data, high-risk locations in the oscillation trajectory are identified. These high-risk locations are areas prone to defects such as undercut and lack of fusion.

[0045] Based on the current welding speed, oscillation amplitude, and material thermophysical properties, calculate the critical heat accumulation radius of the oscillation edge, which is the distance threshold at which overheating symptoms begin to appear at the edge of the molten pool;

[0046] Using the high-risk location as the center and the critical heat accumulation radius as the benchmark, a dynamic residence zone boundary function is established. The dynamic residence zone boundary function outputs the residence zone range that is adjusted in real time with the welding process.

[0047] The dwell area is divided into a core dwell area and a transition buffer zone. The core dwell area performs a complete spiral oscillation motion, while the transition buffer zone performs a smooth transition motion with gradually changing amplitude, ensuring the continuity of the welding torch's trajectory when entering and leaving the dwell area.

[0048] By adopting the above technical solution, the system can dynamically and accurately define key areas requiring special control. Instead of using a fixed dwell zone, it calculates the range of the dwell zone in real time using a dynamic boundary function based on thermophysical properties and historical defect data, ensuring that precise spiral oscillation control is always applied precisely to the most needed location. Furthermore, dividing the dwell zone into a core area and a transition zone guarantees a smooth and continuous trajectory of the welding torch as it enters and leaves this area, avoiding the introduction of new instabilities due to abrupt changes in motion, thereby improving the overall accuracy and stability of the control.

[0049] Secondly, this application provides a composite oscillating welding control system, the system comprising: one or more processors and a memory;

[0050] The memory is coupled to the one or more processors. The memory is used to store computer program code, which includes computer instructions. The one or more processors call the computer instructions so that the system can implement the adaptive variable parameter composite oscillation welding control method provided in the above embodiments, which will not be described in detail here.

[0051] Thirdly, this application provides a computer-readable storage medium including instructions that, when executed on a composite oscillation welding control system, enable the system to implement an adaptive variable parameter composite oscillation welding control method provided in the above embodiments, which will not be elaborated further here.

[0052] Fourthly, this application provides a computer program product that, when running on a composite oscillating welding control system, enables the system to implement an adaptive variable parameter composite oscillating welding control method provided in the above embodiments, which will not be elaborated here.

[0053] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0054] 1. A smart closed loop of "perception-prediction-decision" has been constructed, completely breaking away from the traditional welding model that relies on fixed parameters or delayed feedback. Instead of passively responding to changes, it integrates real-time data from multiple dimensions, including vision, thermal field, and electric arc, and uses deep learning models to perform time-series analysis on this information, thereby accurately predicting the state of the molten pool in the near future. This forward-looking insight allows the system to intervene before welding defects form, calling upon the optimal control strategy from the experience base through fuzzy matching.

[0055] 2. A composite motion mode of "macroscopic reference trajectory + microscopic modulation trajectory" was designed. Especially in the high-defect zone of the weld edge, it does not simply remain stationary, but performs an asymmetric spiral oscillation motion to finely stir the molten pool and equalize heat. Furthermore, the system can dynamically identify the "dwelling zone" requiring fine control based on the heat accumulation model and historical data, and introduce a real-time feedback mechanism to fine-tune the amplitude and duration of the spiral oscillation according to the instantaneous changes in the molten pool width. This control method, which combines a complex motion trajectory with dynamically changing key areas, achieves precise control of the welding process in both spatial and temporal dimensions.

[0056] 3. A self-optimizing loop based on reinforcement learning is introduced. The system quantitatively evaluates the forming quality after each oscillation cycle. When the control effect does not meet expectations, the control strategy for the micro-trajectory can be autonomously updated using a policy gradient algorithm. This enables the control system to continuously learn and iteratively optimize, adapting to changes in process conditions to pursue long-term stable high-quality welding. Attached Figure Description

[0057] Figure 1 This is a flowchart illustrating an adaptive variable parameter composite oscillation welding control method in an embodiment of this application.

[0058] Figure 2 This is another flowchart illustrating an adaptive variable parameter composite oscillation welding control method in an embodiment of this application;

[0059] Figure 3This is an exemplary scenario diagram of the composite oscillation welding control system controlling the tip of the welding torch to perform composite oscillation in an embodiment of this application;

[0060] Figure 4 This is a schematic diagram of the physical structure of a composite swing welding control system in an embodiment of this application. Detailed Implementation

[0061] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to any or all possible combinations including one or more of the listed items.

[0062] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0063] In the embodiments of this application, the composite swing welding control system can be built based on a domestically produced SoC (System-on-Chip) chip, such as the CMC (Control Motion Chip) series chip. This SoC chip integrates a high-performance processor core, memory, a pulse generator for motion control, an ADC (Analog-to-Digital Converter) for signal acquisition, and various communication interfaces. By embedding PLC (Programmable Logic Controller) functions, multi-axis motion control functions, and data processing functions into a single chip's soft core, the system's integration and data interaction speed can be improved. Its response speed is more than twice that of traditional general-purpose controllers, providing computing power support and hardware foundation for the real-time sensing, high-speed prediction, and dynamic trajectory adjustment of this application.

[0064] For ease of understanding, the method provided in this implementation is described in process below. Please refer to [link / reference]. Figure 1 This is a flowchart illustrating an adaptive variable parameter composite oscillation welding control method in an embodiment of this application.

[0065] S101. Acquire real-time multiphysics sensing data.

[0066] Among them, real-time multiphysics sensing data refers to the collection of heterogeneous sensor information that is collected in real time during the welding process and can reflect the different physical field states of the welding area. It is used to characterize the geometry of the molten pool, the temperature field distribution, and the arc stability. The geometry of the molten pool refers to the external manifestation of the shape and size (such as the width, length, and depth of the molten pool) of the molten metal formed during the welding process. The temperature field distribution refers to the overall presentation of the temperature values ​​and variation patterns at different locations within the welding area. The arc stability refers to the degree to which the arc maintains a stable state in terms of combustion and energy output during the welding process.

[0067] Specifically, the core of the composite oscillating welding control system for acquiring real-time multiphysics sensing data is to comprehensively collect physical field information closely related to the welding process through various heterogeneous sensors. Because the welding process is a complex dynamic thermophysical process, relying solely on a single type of sensing data cannot accurately and comprehensively reflect the true state of the molten pool and welding area. Therefore, it is necessary to integrate information from at least two sets of heterogeneous sensors that can characterize the geometry of the molten pool, temperature field distribution, and arc stability. Optionally, the system can deploy visual sensors, adjusting their angle and focal length to clearly capture the molten pool area, continuously acquiring surface image data of the molten pool during welding. This data includes the molten pool outline, surface texture, and liquid metal flow characteristics. Infrared thermal imaging sensors can be installed near the welding equipment to ensure their detection range covers the entire welding area, acquiring real-time temperature field data of the welding area, including the molten pool center temperature, temperature gradient distribution, and heat-affected zone range. Current and voltage sensors are connected to the welding power source to monitor current and voltage changes during welding in real time, acquiring welding electrical parameter data, including instantaneous current values, voltage fluctuation amplitude, and arc length change rate. The acquired molten pool surface image data, welding area temperature field data, and welding electrical parameter data are processed for time synchronization, ensuring that data from different sensors are aligned in the time dimension through a unified timestamp, generating time-aligned multiphysics sensing data.

[0068] S102. Input real-time multiphysics sensing data into a simulation model for predicting the behavior of the molten pool, and output the predicted values ​​of the molten pool state parameters within a preset time period in the future.

[0069] Among them, the simulation model refers to the mathematical or algorithmic model used to predict the behavior of the molten pool. It is trained with historical welding data and can simulate the dynamic changes of the molten pool based on the input real-time multiphysics sensor data.

[0070] Specifically, the composite oscillating welding control system inputs real-time multiphysics sensor data into the simulation model to obtain predicted values ​​of the molten pool state parameters. This is because the welding process is dynamic and lag-dependent; if control is based solely on the current molten pool state, it may be unable to respond promptly to impending changes in the molten pool, leading to welding defects. By predicting the molten pool behavior over a preset time period through the simulation model, the system can anticipate potential states of the molten pool, such as whether it will excessively expand due to heat accumulation or whether the temperature distribution will become unbalanced. The simulation model is trained on historical welding data and has learned the patterns of molten pool state changes with multiphysics sensor data. It can accurately simulate the dynamic changes of the molten pool based on the input real-time data, thereby outputting reliable predicted values ​​of the molten pool state parameters.

[0071] Optionally, the acquired real-time multiphysics sensing data is preprocessed by using filtering algorithms (such as Kalman filtering and wavelet filtering) to remove noise and outliers. The processed data is then standardized to meet the input requirements of the simulation model. Temporal feature vectors are extracted from the standardized sensing data, and time series analysis methods (such as the sliding window method) are used to mine the dynamic change trend information and periodic patterns of the molten pool. The temporal feature vectors are input into a deep learning-based molten pool behavior prediction model (such as an LSTM neural network model), which has been trained and validated using a large amount of historical welding data. The model calculates the input temporal feature vectors and outputs the molten pool geometric parameters (such as predicted molten pool width and depth), temperature distribution parameters (such as predicted molten pool center temperature and temperature gradient), and flow state parameters (such as predicted liquid metal flow velocity) for a future preset time period. Simultaneously, confidence scores are generated for each predicted value, forming predicted molten pool state parameters.

[0072] In some embodiments of this application, the system's control logic and algorithms, including preprocessing of sensor data, calculation of the molten pool behavior prediction model, and decision generation of composite oscillation trajectory parameters, can all be programmed and deployed in a development environment conforming to the IEC 61131-3 international standard. For example, configuration software such as ConfDes can be used, and complex prediction models and data analysis algorithms can be written using structured text (ST) language. Modular control flows, such as PID control loops, can be constructed using function block diagrams (FBD). The PLC logic actuator built into the SoC chip is responsible for sequential control and logical judgment, while the motion control module directly generates high-speed pulse commands (e.g., up to 1MHz) to drive the motor based on PLC instructions and database parameters, achieving high-precision composite oscillation control of the welding torch tip. The PLC logic module and the motion control module achieve high-speed data interconnection by sharing data addresses, ensuring a high degree of coordination between decision-making and execution.

[0073] Furthermore, electrical parameter sensing data is not only used for macroscopic prediction but also for real-time closed-loop feedback control. For example, the system can execute AVC (Automatic Voltage Control) functionality. Specifically, this involves high-speed acquisition of signals from the arc voltage sensor via an ADC controller integrated into the SoC chip. After signal conditioning and filtering, mathematical features such as peak and valley values ​​of the voltage signal are extracted to determine the current arc length. Subsequently, a PID controller built into the control system compares the real-time arc voltage value with the target arc voltage set in the database and calculates the adjustment amount based on the error. This adjustment drives the welding torch lifting axis (i.e., the AVC axis) in real-time for fine-tuning, thereby maintaining the optimal arc length and further improving the stability of the welding process.

[0074] S103. Match the composite oscillation trajectory parameters corresponding to the predicted values ​​of the molten pool state parameters from the preset database.

[0075] Among them, the preset database refers to a pre-built database that stores the mapping relationship between welding process parameters and oscillation trajectory parameters, as well as historical successful welding cases, including the optimal parameter combination under different welding conditions; the composite oscillation trajectory parameters refer to the set of parameters used to define the composite motion trajectory of the welding torch tip, including macroscopic reference trajectory parameters and microscopic modulation trajectory parameters; the macroscopic reference trajectory parameters refer to the parameters used to define the macroscopic motion trajectory of the welding torch tip, which determines the overall oscillation path and range of the welding torch; the microscopic modulation trajectory parameters refer to the parameters used to define the microscopic motion adjustment of the welding torch tip based on the macroscopic trajectory, which determines the fine motion mode of the welding torch in a specific area (such as the preset dwell area).

[0076] Specifically, the composite oscillation welding control system matches composite oscillation trajectory parameters from a preset database because different molten pool states require corresponding oscillation trajectory parameters to control the welding torch movement in order to achieve good weld formation. The preset database stores a large number of verified mapping relationships between welding process parameters and oscillation trajectory parameters, as well as historical successful welding cases. This data covers the optimal parameter combinations under various welding conditions such as different materials, plate thicknesses, and welding positions. After the system obtains the predicted value of the molten pool state parameters, it needs to search the database for the parameter combination that best matches the predicted value. This is because the predicted value reflects the future state of the molten pool. Only by matching the corresponding composite oscillation trajectory parameters can the welding torch ensure the correctness of the overall welding path through the macroscopic reference trajectory during subsequent movements, while simultaneously making fine adjustments in key areas through microscopic modulation of the trajectory, thereby effectively controlling the molten pool state and avoiding welding defects such as undercut and lack of fusion.

[0077] Optionally, the system pre-constructs a mapping database of welding process parameters and oscillation trajectory parameters, collects successful welding cases under different materials (e.g., carbon steel, alloy steel), different plate thicknesses (e.g., 10mm, 20mm), and different welding positions (e.g., flat welding, vertical welding), extracts the welding process parameters and corresponding optimal oscillation trajectory parameters, establishes the mapping relationship, and stores it in a preset database; based on the material properties (e.g., material type, yield strength) and structural characteristics (e.g., plate thickness, bevel form) of the current welding task, it determines the corresponding trajectory parameter search space in the mapping database, which limits the range of parameter values ​​that meet the current welding process constraints; using a fuzzy matching algorithm (e.g., fuzzy comprehensive evaluation method based on similarity), it compares the predicted values ​​of the molten pool state parameters with historical cases in the trajectory parameter search space and calculates the similarity; it selects the historical case with the highest similarity and extracts the corresponding composite oscillation trajectory parameters (including macroscopic reference trajectory parameters and microscopic modulation trajectory parameters) from this case as the matching result.

[0078] Furthermore, before executing step S104, the system determines the specific range of the preset dwell area. In this embodiment, the dwell zone is dynamically generated. Specific steps include: determining the spatial boundary of the preset dwell zone based on weld geometry and heat accumulation distribution characteristics; defining the key area requiring special motion control; identifying high-risk locations in the oscillation trajectory by analyzing the defect distribution patterns in historical welding data, where high-risk locations are areas prone to defects such as undercut and lack of fusion; calculating the critical heat accumulation radius of the oscillation edge based on the current welding speed, oscillation amplitude, and material thermophysical properties, where the critical heat accumulation radius is the distance threshold at which overheating symptoms begin to appear at the edge of the molten pool; establishing a dynamic dwell zone boundary function centered on the high-risk location and using the critical heat accumulation radius as a reference, where the dynamic dwell zone boundary function outputs a dwell zone range that adjusts in real-time with the welding process; dividing the dwell zone range into a core dwell zone and a transition buffer zone, where the core dwell zone performs a complete spiral oscillation motion, and the transition buffer zone performs a smooth transition motion with gradually changing amplitude, ensuring the continuity of the welding torch's trajectory when entering and leaving the dwell zone.

[0079] In some embodiments of this application, the preset database can be a structured welding process database, which is integrated or linked to the welding control system. This database not only stores macroscopic and microscopic trajectory parameters but also systematically includes a large number of validated welding process packages, the content of which originates from the experience data of senior welding technicians and the results of numerous process experiments. Each process package records in detail a complete set of parameters for a specific working condition (such as base material type, plate thickness, groove type, welding position, etc.), including but not limited to welding current, arc voltage, welding speed, wire feed speed, gas flow rate, and the corresponding optimal macroscopic reference trajectory parameters (such as oscillation amplitude, frequency) and microscopic modulation trajectory parameters (such as the amplitude, attenuation coefficient, and asymmetry coefficient of helical oscillation, etc.). The database design supports the storage of dozens or even more process files and allows for on-demand access by PLC logic and motion control modules through an efficient data sharing mechanism, realizing the automated selection and optimization of welding processes.

[0080] S104. When the welding torch enters the preset dwelling area of ​​the weld swing edge according to the macro reference trajectory defined by the macro reference trajectory parameters, the tip of the welding torch is controlled to perform an asymmetric spiral oscillation motion in the preset dwelling area until it leaves the preset dwelling area according to the micro modulation trajectory defined by the micro modulation trajectory parameters.

[0081] Among them, the preset dwell zone refers to a specific area located at the edge of the weld swing, which is pre-defined. In this area, the movement of the welding torch needs to be specially controlled to ensure the welding quality. The asymmetric spiral oscillation motion refers to the oscillation motion with an asymmetric spiral trajectory executed by the tip of the welding torch in the preset dwell zone. This motion can be used to finely control the local heat input and the flow of molten metal.

[0082] Specifically, the composite oscillating welding control system controls the welding torch to perform asymmetric spiral oscillation motion when it enters the preset dwell zone because the preset dwell zone at the edge of the weld oscillation is a high-incidence area for welding defects. In traditional welding control, the welding torch typically dwells in the edge area for a fixed time. If the heat input is too high, it will lead to overheating of the molten pool, metal loss, and undercut or collapse; if the heat input is insufficient, the edge will not fuse. The macroscopic reference trajectory can only ensure the correctness of the overall oscillation path of the welding torch, and cannot meet the needs of fine control of the dwell zone. Therefore, when the welding torch enters the preset dwell zone, it is necessary to control the welding torch to perform asymmetric spiral oscillation motion according to the micro-modulation trajectory parameters. This motion mode can finely control the local heat input in the dwell zone by adjusting the asymmetric distribution of the spiral and the oscillation frequency, while stirring the molten metal to promote uniform mixing of the molten pool and avoid excessive heat concentration or uneven distribution, thereby effectively suppressing the generation of welding defects until the welding torch leaves the dwell zone and then returns to the normal macroscopic oscillation trajectory.

[0083] Optionally, the system can pre-set the spatial boundary parameters of the preset dwell zone. These parameters are determined based on the weld geometry (such as groove width and edge angle) and heat accumulation distribution characteristics. When the welding torch moves along the macroscopic reference trajectory, the welding torch position coordinates are monitored in real time. When the welding torch position coordinates are detected to enter the spatial boundary range of the preset dwell zone, the system calls the micro-modulation trajectory parameters, which include the starting amplitude, ending amplitude, oscillation frequency, and asymmetry coefficient of the spiral oscillation. Based on the micro-modulation trajectory parameters, an asymmetric spiral oscillation motion control command is generated to control the welding torch drive mechanism (such as a servo motor) to drive the welding torch tip to move according to the command. During the movement, the radius and direction of the spiral are adjusted in real time to achieve asymmetric oscillation. The welding torch position is continuously monitored. When the welding torch position coordinates exceed the spatial boundary range of the preset dwell zone, the asymmetric spiral oscillation motion is stopped, and the welding torch is controlled to resume moving along the macroscopic reference trajectory. The asymmetry can manifest as the central axis of the helical oscillation shifting towards the welding direction or away from the already welded weld; and / or, the amplitude of the helical oscillation trajectory differs along and perpendicular to the welding direction, forming an elliptical helix; and / or, the pitch or amplitude of the helix changes asymmetrically when entering and leaving the dwell zone. This design aims to more precisely control the heat distribution at the edges, for example, preheating the base material in front while avoiding overheating the already solidified weld behind.

[0084] Optionally, the system can also establish a dynamic recognition model of the preset dwell zone. Combining the molten pool state data (such as the edge position of the molten pool) and the welding torch movement trajectory during the real-time welding process, it can determine in real time whether the welding torch has entered the preset dwell zone. This model can fine-tune the boundary of the dwell zone according to the dynamic changes during the welding process (such as the expansion of the molten pool due to heat accumulation). When the model determines that the welding torch has entered the preset dwell zone, the system extracts the micro-modulation trajectory parameters from the matched composite oscillation trajectory parameters and analyzes the motion law of asymmetric spiral oscillation (such as the equation parameters of the spiral and the oscillation period). The micro-modulation trajectory parameters are converted into control signals that can be recognized by the welding torch drive system. The speed and direction of the servo motor are controlled by pulse width modulation (PWM) technology, thereby driving the tip of the welding torch to perform asymmetric spiral oscillation motion. During the movement of the welding torch, the actual position of the welding torch is obtained in real time through a position feedback sensor (such as an encoder) and compared with the preset trajectory. If there is a deviation, the control signal is adjusted in time to correct it until the welding torch leaves the preset dwell zone and restores the macroscopic reference trajectory motion.

[0085] For a clearer explanation of the invention, please refer to [link / reference]. Figure 3 This is an exemplary scenario diagram illustrating how the composite oscillation welding control system controls the welding torch tip to perform composite oscillation in an embodiment of this application. Figure 3As shown, welding is performed along the "welding direction". The tip of the welding torch typically moves along a "macro reference trajectory", which causes it to oscillate laterally within the weld.

[0086] When the welding torch moves along the macroscopic reference trajectory to the "weld edge" and enters the "preset dwell zone," its motion mode switches. The system will control the tip of the welding torch to perform a fine "asymmetric helical oscillation motion." For example... Figure 3 As shown in the "preset dwell zone," the motion trajectory is not a simple circle or a symmetrical spiral, but rather an asymmetrical spiral shape with continuously changing curvature and inward convergence. This "asymmetry" design enables non-uniform, directional, and precise control of heat input to the edge region.

[0087] Specifically, this asymmetry can be achieved in at least one of the following ways:

[0088] Asymmetrical shape: such as Figure 3 As shown, the spiral trajectory exhibits a shape with one side being smooth and convex, and the other side narrowing and concave, forming a contour similar to a heart shape or a kidney shape. By properly orienting the trajectory, the smooth, convex side can be used for large-scale uniform preheating or fusion of the bevel sidewall, ensuring the fusion width; while the narrow, concave part of the trajectory can be quickly turned and precisely pointed to key positions such as the root of the bevel for brief focused heating to ensure the fusion depth, while avoiding prolonged stay in a single area.

[0089] Speed ​​asymmetry: The linear velocity of the welding torch during this spiral motion can be non-constant. For example, it can decelerate on the side of the smooth, convex trajectory to increase the arc heating time; while it can accelerate when the trajectory narrows and turns rapidly to reduce local heat input.

[0090] Central asymmetry: The geometric center of the entire spiral oscillation can deviate from the theoretical endpoint of the macroscopic trajectory, for example, by fine-tuning towards the unwelded area or downward (in the direction of gravity) to preheat the base material or counteract the molten pool's fall.

[0091] By executing this composite trajectory until the welding torch leaves the dynamic dwell area, the edge area is fully fused while effectively avoiding defects such as undercut, burn-through, or collapse caused by excessive local heat concentration.

[0092] In one specific embodiment, with Figure 3 Taking the cardioid spiral shown as an example, its two-dimensional coordinates (x(t'), y(t')) within the dynamic dwell region can be described by the following exemplary system of equations:

[0093] x(t') = A × (1 + k×cos(ωt')) × cos(ωt') × exp(-αt');

[0094] y(t') = B × (1 + k×cos(ωt')) × sin(ωt') × exp(-αt');

[0095] Where t' is the time when timing begins within the dynamic dwell zone; A and B are the basic amplitudes in the X and Y directions, respectively, with an elliptical spiral formed when A≠B; k is the asymmetry coefficient, determining the degree of concavity in the cardioid; ω is the oscillation angular frequency; exp(-αt') is the exponential decay term, and α is the decay coefficient, used to control the speed of inward convergence of the spiral. By adaptively adjusting these parameters (A, B, k, ω, α), various forms of asymmetric spiral trajectories can be generated to cope with different melt pool prediction states.

[0096] It should be noted that, Figure 3 The “preset dwell area” schematically indicated in the diagram is intended to represent a general functional area that requires special control. In other embodiments of the invention, the precise boundary of this area can be calculated and dynamically adjusted in real time according to the welding process, rather than being fixed.

[0097] In the above embodiments, the system can acquire multiphysics sensing data characterizing the state of the molten pool and use simulation models to predict future molten pool behavior. Based on this prediction, the system can adaptively match composite oscillation trajectory parameters from the database, especially in the oscillation edge dwelling area where heat tends to concentrate. By executing asymmetric spiral oscillation motion, it can finely control local heat input and molten metal flow. This effectively alleviates problems such as edge overheating, undercut, or insufficient fusion caused by fixed parameters, improving weld formation quality and welding process stability.

[0098] The following provides a more detailed description of the process of the method provided in this implementation. Please refer to [link / reference]. Figure 2 This is another flowchart illustrating an adaptive variable parameter composite oscillation welding control method in an embodiment of this application.

[0099] S201. Visual sensors, infrared thermal imaging sensors, and current and voltage sensors are used to collect image data of the molten pool surface, temperature field data of the welding area, and welding electrical parameter data, respectively.

[0100] Specifically, this step is the initial stage for the composite oscillating welding control system to acquire multi-physics field sensing data. The core is to collect key data from different dimensions during the welding process using three types of heterogeneous sensors. The vision sensor must be directly facing the molten pool area. By adjusting the lens focal length and exposure parameters, it ensures clear capture of the dynamic changes in the molten pool, such as the expansion and contraction of the molten pool edge and the flow patterns of surface ripples, thereby generating molten pool surface image data. The infrared thermal imaging sensor must cover the entire welding operation area, with its detection range including the molten pool, heat-affected zone, and surrounding base material. By receiving the infrared radiation intensity of different areas, it converts it into temperature values, forming temperature field data for the welding area, which can intuitively reflect the temperature difference between the center and edge of the molten pool and the range changes of the heat-affected zone. The current and voltage sensors must be connected in series or parallel in the circuit between the welding power source and the welding torch to collect instantaneous current fluctuations and voltage stability during the welding process in real time. Combining the correlation between arc length and electrical parameters, the arc length change rate is derived, generating welding electrical parameter data.

[0101] S202. Perform time synchronization processing on the molten pool surface image data, welding area temperature field data, and welding electrical parameter data to generate time-aligned multiphysics sensing data.

[0102] This step is crucial for resolving the time discrepancy issue in data from multi-source heterogeneous sensors. The core is to eliminate data time misalignment caused by startup delays and differences in sampling frequencies between different sensors, ensuring accurate correspondence between the molten pool surface state, temperature distribution, and electrical parameters at the same time point in subsequent analysis. Because visual sensors, infrared thermal imaging sensors, and current / voltage sensors operate on different principles, their data acquisition time intervals and startup response speeds may differ. For example, current / voltage sensors have high sampling frequencies and short data generation intervals, while visual sensors, affected by image processing speed, have longer data generation intervals. If unsynchronized data is used directly, situations may arise where "electrical parameters at one moment correspond to the molten pool image at another moment," leading to deviations in subsequent molten pool state analysis. Therefore, time synchronization processing is necessary. Using the timestamp of a specific sensor (usually the current / voltage sensor with the highest sampling frequency) as a benchmark, the data from the other two types of sensors are interpolated or filtered to ensure a one-to-one correspondence between the three types of data on the time axis, ultimately generating time-aligned multiphysics sensing data.

[0103] S203. Extract time-series feature vectors after denoising preprocessing of real-time multiphysics sensing data.

[0104] Among them, noise reduction preprocessing refers to the operation of removing noise and abnormal data caused by external interference (such as arc light and vibration) from real-time multiphysics field sensing data by using algorithms such as filtering and outlier removal; time-series feature vector refers to vector data extracted from the noise-reduced multiphysics field sensing data that can reflect the dynamic change trend of the molten pool (such as the expansion trend of the molten pool width over time) and periodic patterns (such as the periodic fluctuation of the molten pool temperature with the oscillation of the welding torch).

[0105] Specifically, this step is the core link in transforming raw multiphysics sensor data into effective feature information, and it is divided into two key parts: noise reduction preprocessing and temporal feature extraction. In the noise reduction preprocessing stage, due to the complex welding site environment, the molten pool images acquired by the visual sensor may be affected by changes in arc light intensity and dust obstruction, resulting in noise. Infrared thermal imaging data may be affected by ambient temperature fluctuations, resulting in outliers. Current and voltage data may be affected by power grid fluctuations, resulting in spike interference. Therefore, appropriate noise reduction methods need to be adopted for different types of data to ensure data authenticity. In the temporal feature extraction stage, it is necessary to mine the dynamic patterns of the data based on the time dimension, because the state of the molten pool (such as width and temperature) changes with welding time and welding torch oscillation cycle. Analyzing data at only a single time point cannot reflect its changing trend. However, temporal feature vectors can integrate dynamic information over a period of time, providing input data that can characterize the changing patterns of the molten pool for subsequent molten pool behavior prediction models.

[0106] S204. Input the time-series feature vector into the deep learning-based molten pool behavior prediction model, calculate the molten pool geometric parameters, temperature distribution parameters, and flow state parameters within a preset time period in the future, and generate predicted values ​​of molten pool state parameters including confidence scores.

[0107] Among them, the deep learning-based molten pool behavior prediction model refers to a model that uses a neural network architecture (such as LSTM, Transformer) and is trained on a large amount of historical welding data to predict the future state of the molten pool based on the input time-series feature vector; the future preset time period refers to the pre-set time interval corresponding to the molten pool state that the model needs to predict (such as the future 100ms); the molten pool geometric parameters refer to the parameters used to describe the shape and size of the molten pool (such as the maximum width and average depth of the molten pool in the future time period); the temperature distribution parameters refer to the parameters used to describe the temperature of the molten pool and its surroundings (such as the minimum temperature at the center of the molten pool and the rate of change of the temperature gradient in the future time period); and the flow state parameters refer to the parameters used to describe the flow of liquid metal in the molten pool (such as the average flow velocity of liquid metal and the frequency of change of flow direction in the future time period).

[0108] Specifically, this step is the core of achieving a "perception-prediction" closed loop. The core is to utilize a well-trained deep learning model to transform temporal feature vectors into predictions of the future molten pool state. Because the welding process is dynamic and lag-dependent, directly adjusting welding torch parameters based on the current molten pool state may lead to defects due to control lag. Therefore, it is necessary to predict the molten pool state within a preset time period and formulate control strategies in advance. The deep learning-based molten pool behavior prediction model learns the mapping relationship between "temporal features" and "future molten pool state" in historical welding data, capturing the nonlinear patterns of molten pool state changes, such as the lag in molten pool width changes with electrical parameter fluctuations and the gradual trend of temperature distribution with heat accumulation, thus outputting accurate prediction results. Simultaneously, confidence scoring helps the system judge the reliability of the prediction results. If the confidence level is too low (e.g., below 0.6), the system can trigger data re-acquisition or model fine-tuning to ensure the accuracy of subsequent parameter matching.

[0109] In some embodiments, the prediction of molten pool state parameters can be achieved in several ways: Optionally, the first step is to load a pre-trained LSTM-based molten pool behavior prediction model. This model has been trained using historical data from over 1000 different welding conditions (different materials, plate thicknesses, and welding positions). During training, the model parameters are optimized through backpropagation using temporal feature vectors as input and actual molten pool state parameters corresponding to a preset future time period (e.g., 100ms) as labels, ensuring that the model prediction error is less than 5%. The second step is to standardize the extracted temporal feature vectors (e.g., normalize them to the 0-1 range) and input them into the input layer of the LSTM model. The model learns the dynamic dependencies of temporal features through hidden layers (set to 3 layers, 128 nodes per layer). The output layer is a fully connected layer, which outputs the predicted values ​​of the molten pool geometric parameters (width, depth, area), temperature distribution parameters (center temperature, temperature gradient, heat-affected zone temperature), and flow state parameters (flow velocity, flow direction) within the next 100ms. In the third step, the model's uncertainty assessment module calculates the confidence score corresponding to each predicted parameter based on the deviation between the predicted value and the prediction results of similar historical cases (e.g., using the Monte Carlo dropout method to assess uncertainty). The predicted parameters and confidence scores are integrated to generate the predicted values ​​of the molten pool state parameters.

[0110] S205. Construct a mapping library between welding process parameters and oscillation trajectory parameters.

[0111] Specifically, this step is the foundational step for providing data support for subsequent parameter matching. Its core is to establish a precise correlation between "process conditions" and "trajectory parameters" by collecting, organizing, and verifying a large number of welding cases. Because the requirements for the oscillation trajectory of the molten pool vary greatly under different welding process conditions—for example, thick plate welding requires a larger oscillation amplitude to ensure penetration, while vertical welding requires a shorter edge dwell time to prevent the molten pool from flowing—without a mapping database, the system would need to repeatedly try and fail to determine suitable trajectory parameters, leading to low efficiency and a high risk of defects. Therefore, when building the mapping database, it is necessary to cover a variety of common welding process parameter combinations. The oscillation trajectory parameters corresponding to each combination must be verified through actual welding (such as weld formation quality inspection and mechanical property testing) to ensure that they can achieve good welding results. Simultaneously, the database needs to record the welding results of each case (such as whether there is undercut or lack of fusion) to provide a quality reference for subsequent fuzzy matching.

[0112] S206. Determine the trajectory parameter search space based on the mapping relationship library corresponding to the material properties and structural characteristics of the current welding task.

[0113] Among them, the material properties of the current welding task refer to the material-related characteristics of the workpiece to be welded, including material type (such as Q345 steel, 304 stainless steel), material composition (such as carbon content, alloy element ratio), thermophysical properties (such as thermal conductivity, melting point), etc.; structural characteristics refer to the structural-related characteristics of the workpiece to be welded, including plate thickness, bevel type (such as V-groove, U-groove), weld length, joint type (such as butt joint, corner joint), etc.; trajectory parameter search space refers to the set of oscillation trajectory parameter value ranges in the mapping relation library that match the material properties and structural characteristics of the current welding task and meet the process constraints (such as the oscillation amplitude not exceeding 80% of the bevel width).

[0114] Specifically, this step is a prerequisite for achieving accurate parameter matching. Its core is to filter out a range of parameters that meet the specific conditions of the current welding task from a vast mapping database, avoiding blind searches. Different material properties and structural characteristics impose strict constraints on the oscillation trajectory parameters. For example, stainless steel has a low thermal conductivity and is prone to heat accumulation, requiring a limitation on the dwell time of the oscillation edge (a micro-modulation trajectory parameter). Therefore, the trajectory parameter search space must exclude parameters with excessively long dwell times. Thick plate workpieces require a larger oscillation amplitude (a macro-based baseline trajectory parameter) to ensure penetration, and the search space must include a range of parameters with large oscillation amplitudes. By determining the trajectory parameter search space, on the one hand, the range of subsequent fuzzy matching can be narrowed, reducing computational resource consumption and improving matching speed; on the other hand, parameters that do not conform to the current process constraints can be excluded, avoiding welding defects caused by parameters exceeding reasonable ranges, and ensuring the feasibility of subsequent matching parameters.

[0115] S207. Use a fuzzy matching algorithm to find the historical cases that are closest to the predicted values ​​of the molten pool state parameters in the trajectory parameter search space, and generate the corresponding candidate composite swing trajectory parameter set.

[0116] Among them, historical cases refer to verified successful cases stored in the mapping relationship library, which include welding process parameters, oscillation trajectory parameters, molten pool state parameters and welding quality results; the candidate composite oscillation trajectory parameter set refers to a set of multiple candidate parameter combinations (usually containing 1-5 sets of optimal parameters) extracted from the closest historical cases found, which includes macroscopic reference trajectory parameters and microscopic modulation trajectory parameters.

[0117] Specifically, this step is a crucial link in achieving the "prediction-decision" closed loop. The core is to use a fuzzy matching algorithm to filter historical cases within the trajectory parameter search space that best match the predicted values ​​of the current molten pool state parameters, and then extract candidate parameters. Due to uncertainties in the welding process (such as differences in workpiece surface cleanliness and ambient temperature fluctuations), the predicted values ​​of the molten pool state parameters are rarely completely consistent with the molten pool state parameters in historical cases. If precise matching is used, corresponding cases may not be found. However, fuzzy matching, by calculating similarity (e.g., converting molten pool geometric parameters, temperature distribution parameters, and flow state parameters into vectors and calculating the Euclidean distance between vectors; the smaller the distance, the higher the similarity), can find historical cases closest to the predicted values. The oscillation trajectory parameters in these cases have been verified through actual welding and can effectively handle molten pool states similar to the predicted values. Therefore, extracting these parameters to form a candidate composite oscillation trajectory parameter set can provide reliable parameter selection for subsequent welding torch control. Furthermore, the candidate set contains multiple sets of parameters, which can be further optimized based on actual welding results.

[0118] S208. When performing asymmetric spiral oscillation motion, monitor the change value of the molten pool width in the preset dwell zone in real time.

[0119] Specifically, during the asymmetric helical oscillation motion of the welding torch according to micro-modulated trajectory parameters, the weld pool width is a core indicator reflecting the state of heat accumulation in the edge region. Excessive heat input will cause the weld pool to expand due to overheating, increasing the width variation and potentially leading to undercut or collapse. Insufficient heat input will result in a small or even stable weld pool width variation, potentially causing incomplete fusion defects. Therefore, the system needs to monitor the weld pool width variation in real time, continuously capturing its dynamic fluctuations to provide direct data for subsequent parameter adjustments. This monitoring process must be synchronized with the helical oscillation motion to ensure that the weld pool state at each stage of the motion is detected promptly, preventing defects caused by monitoring lag.

[0120] Optionally, the first step involves integrating a high-frame-rate vision sensor (frame rate not less than 30fps) into the composite oscillating welding control system, focusing its field of view on the preset dwell area and the surrounding molten pool area to ensure clear capture of the molten pool edge contour; the second step involves triggering the vision sensor to continuously acquire molten pool images when the helical oscillation motion starts, generating a frame of molten pool image data every 5ms; the third step involves processing each frame of molten pool image using an edge detection algorithm (such as the Canny algorithm), extracting the molten pool contour and calculating the width value, subtracting the width value of the current frame from the width value of the previous frame to obtain the real-time molten pool width change value, and simultaneously transmitting the change value to the system control module in real time.

[0121] S209. When the change in the width of the molten pool exceeds the preset threshold, the amplitude attenuation coefficient of the spiral oscillation motion is adjusted according to the preset ratio.

[0122] Specifically, the preset threshold is the core standard for the system to determine whether the heat input of the molten pool is unbalanced. When the change in molten pool width exceeds the upper limit of the preset threshold, it indicates that there is excessive heat accumulation in the dwell zone, and the molten pool is expanding rapidly. If not adjusted in time, it is prone to defects such as undercut and collapse. At this time, the amplitude attenuation coefficient needs to be increased according to the preset ratio to accelerate the deceleration rate of the spiral oscillation radius, so that the movement range of the welding torch in the dwell zone gradually shrinks, reducing the local arc action time and thus reducing heat input. If the system sets a lower limit of the preset threshold, when the change in molten pool width is lower than the lower limit (i.e., the molten pool width increases slowly or stagnates), it indicates insufficient heat input. The amplitude attenuation coefficient needs to be reduced according to the preset ratio to slow down the attenuation rate of the spiral radius, extend the dwell time of the welding torch in the effective heating area, increase local heat input, and ensure full fusion of the molten pool. The entire adjustment process must be based on the preset ratio to avoid drastic fluctuations in the welding torch movement trajectory due to sudden changes in the coefficient, which would affect the stability of the weld formation.

[0123] Optionally, the first step involves pre-entering the preset threshold for the change in molten pool width (e.g., upper limit 1mm, lower limit 0.2mm) and the preset adjustment ratio for the amplitude attenuation coefficient (e.g., increasing by 15% when exceeding the limit, and decreasing by 10% when below the lower limit) in the parameter configuration module of the composite oscillation welding control system for the current welding task. The second step involves the system comparing the monitored change in molten pool width with the preset threshold in real time. When the change reaches 1.2mm (exceeding the 1mm upper limit), an adjustment command is triggered. The third step involves obtaining the current amplitude attenuation coefficient (e.g., the initial coefficient is 0.3), calculating the adjustment amount (0.3×15%=0.045) according to the preset ratio of 15%, updating the coefficient to 0.345, and transmitting the updated coefficient to the welding torch motion control unit to adjust the spiral oscillation trajectory in real time.

[0124] S210. Adjust the dwell time of the welding torch tip in the preset dwell area according to the adjusted amplitude attenuation coefficient.

[0125] Specifically, adjusting the amplitude attenuation coefficient directly alters the trajectory characteristics of the helical oscillation motion. When the coefficient increases, the helical radius attenuates more rapidly, and the effective heating range of the welding torch within the dwell zone shrinks quickly. If the original dwell time is maintained, insufficient heat input may result later. Conversely, when the coefficient decreases, the helical radius attenuates more slowly, and the effective heating range is maintained for a longer period. If the dwell time remains unchanged, excessive heat input accumulation may occur. Therefore, the dwell time must be adjusted synchronously with the adjusted amplitude attenuation coefficient: when the coefficient increases (heat input needs to be reduced), the dwell time should be appropriately shortened to avoid the welding torch dwelling too long on a small radius trajectory, leading to localized overheating; when the coefficient decreases (heat input needs to be increased), the dwell time should be appropriately extended to ensure the welding torch has sufficient time to heat the molten pool on a larger radius trajectory, achieving full fusion. The adjustment of the dwell time must be consistent with the changing trend of the amplitude attenuation coefficient, forming a coordinated control of "trajectory characteristics - dwell time," further improving the accuracy of heat input in the dwell zone.

[0126] Optionally, the first step is to establish a correlation formula between "amplitude attenuation coefficient and dwell time" in the system. For example, the dwell time T = initial dwell time T0 × (1 + Δk), where Δk is the change in amplitude attenuation coefficient (Δk is positive when the coefficient increases and negative when the coefficient decreases), and the proportional coefficient in the welding process calibration formula is used. The second step is to adjust the amplitude attenuation coefficient from 0.3 to 0.345 (Δk = 0.045) and the initial dwell time T0 = 200ms, and substitute it into the formula to calculate the adjusted dwell time T = 200 × (1 + 0.045) = 209ms. The third step is for the system to send the adjusted dwell time to the motion control module, which controls the welding torch to perform spiral oscillation motion within the preset dwell area for a total duration of 209ms, and automatically leave the dwell area after the motion ends.

[0127] S211. Establish a trajectory optimization feedback mechanism based on reinforcement learning and collect weld formation quality indicators during the execution of asymmetric helical oscillation motion.

[0128] Specifically, the trajectory optimization feedback mechanism based on reinforcement learning is the core framework for the system to achieve "self-learning and self-optimization." This mechanism must first be established during system initialization: a reward function is designed to convert weld formation quality indicators (such as weld height uniformity and undercut depth), heat input uniformity indicators (such as the standard deviation of the residence zone temperature), and edge fusion state indicators (such as the clarity of the fusion line) into quantifiable reward values. Higher quality, more uniform heat input, and more complete fusion result in higher reward values. Simultaneously, a policy network is constructed to generate control strategies (such as helical oscillation parameters) for micro-modulated trajectories. During and after each asymmetric helical oscillation motion, the system needs to collect weld formation quality indicators: during motion, a visual sensor captures the weld surface morphology in real time to initially obtain the waviness coefficient; after motion, a laser profile sensor scans the weld cross-section to measure weld height uniformity and undercut depth. If conditions permit, ultrasonic testing can also be used to obtain internal fusion state data. These quality indicators will serve as the core basis for subsequent calculation of the comprehensive reward value and updating the policy network, supporting the system's continuous optimization of the control strategy.

[0129] S212. When the comprehensive reward value of the current swing cycle is lower than the historical average, the control strategy of the micro-modulation trajectory is updated by the policy gradient algorithm and applied to the next swing cycle.

[0130] Among them, the comprehensive reward value refers to the value calculated by a preset reward function based on the weld formation quality index, heat input uniformity index, and edge fusion state index collected in step S211, which quantifies the control effect of the current oscillation cycle; the historical average value refers to the arithmetic mean of the comprehensive reward values ​​of multiple past oscillation cycles (such as the first 50 cycles) stored in the system, which is used to determine whether the current control strategy is better than the historical level; the policy gradient algorithm refers to the algorithm used in reinforcement learning to update the policy network parameters (such as the REINFORCE algorithm), which adjusts the parameters to improve the future reward value by calculating the gradient of the reward value with respect to the policy parameters; the control strategy of micro-modulation trajectory refers to the decision rule for generating micro-modulation trajectory parameters (such as the amplitude, frequency, and attenuation coefficient of the spiral oscillation), which determines the motion mode of the welding torch in the preset dwell area.

[0131] Specifically, after each oscillation cycle, the system calculates a comprehensive reward value based on the collected quality indicators. This value directly reflects the quality of the current control strategy. If the comprehensive reward value is higher than the historical average, it indicates that the current strategy is better than before and no update is needed. If it is lower than the historical average, it indicates that the current strategy is no longer suitable for the current welding conditions (such as increased workpiece heat accumulation or slight fluctuations in material properties), and the control strategy needs to be updated through the strategy gradient algorithm. The core logic of the strategy gradient algorithm is: with the goal of "improving the comprehensive reward value", it calculates the gradient of the current comprehensive reward value with respect to the strategy network parameters, and adjusts the network parameters along the gradient upward direction so that when the strategy network generates micro-modulation trajectory parameters next time, it is more inclined to output parameter combinations that can obtain higher reward values ​​(such as adjusting the amplitude and frequency of the spiral oscillation, and optimizing the matching degree between heat input and molten pool). The updated control strategy will be directly applied to the next oscillation cycle, realizing "cycle-level" real-time optimization and ensuring that the system can maintain high-quality welding results even when welding conditions change.

[0132] Optionally, in the first step, after each oscillation cycle, the system calls the reward function to calculate the current comprehensive reward value (e.g., R=85), and simultaneously reads the comprehensive reward values ​​of the past 50 cycles, calculates the historical average (e.g., 88), and finds that 85<88, triggering a policy update; in the second step, the REINFORCE policy gradient algorithm is used to input the current oscillation cycle's state (predicted value of molten pool state parameters), action (micro-modulation trajectory parameters), and comprehensive reward value R into the policy network, and calculates the gradient of the reward value with respect to network parameters (e.g., weights, biases); in the third step, the network parameters are updated along the gradient ascending direction (e.g., the learning rate is set to 0.001, and the parameters are adjusted according to the gradient value), generating an updated control policy; in the fourth step, before the start of the next oscillation cycle, the updated control policy is loaded into the motion control module. When the welding torch enters the preset dwell area, the system generates micro-modulation trajectory parameters based on the new policy and controls the welding torch to perform spiral oscillation motion.

[0133] The composite oscillating welding control system of this invention is applied to electronic equipment. Figure 4 A schematic diagram of the architecture of an electronic device suitable for implementing embodiments of the present invention is shown.

[0134] It should be noted that, Figure 4 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0135] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by instructions (computer programs), or by instructions (computer programs) controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor. The electronic device of this embodiment includes a storage medium and a processor, wherein the storage medium stores multiple instructions that can be loaded by the processor to execute any step of the method provided in the embodiments of the present invention.

[0136] Specifically, the storage medium and the processor are electrically connected directly or indirectly to enable data transmission or interaction. For example, these components can be electrically connected to each other via one or more signal lines. The storage medium stores computer-executable instructions that implement data access control methods, including at least one software functional module that can be stored in the storage medium in the form of software or firmware. The processor executes various functional applications and data processing by running the software program and module stored in the storage medium. The storage medium can be, but is not limited to, Random Access Memory (RAM), Read-Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The storage medium stores the program, and the processor executes the program after receiving the execution instructions.

[0137] Furthermore, the software programs and modules within the aforementioned storage medium may also include an operating system, which may include various software components and / or drivers for managing system tasks (e.g., memory management, storage device control, power management, etc.) and can communicate with various hardware or software components to provide an operating environment for other software components. The processor may be an integrated circuit chip with signal processing capabilities. The aforementioned processor may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc., which can implement or execute the methods, steps, and logic flowcharts disclosed in this embodiment. The general-purpose processor may be a microprocessor or any conventional processor.

[0138] Since the instructions stored in the storage medium can execute the steps in any of the methods provided in the embodiments of the present invention, the beneficial effects of any of the methods provided in the embodiments of the present invention can be achieved, as detailed in the preceding embodiments, and will not be repeated here.

[0139] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An adaptive variable parameter composite oscillation welding control method, applied to a composite oscillation welding control system, characterized in that, The method includes: Acquire real-time multiphysics sensing data, which includes information from at least two heterogeneous sensors used to characterize the geometry of the molten pool, the temperature field distribution, and the stability of the electric arc. The real-time multiphysics sensing data is input into a simulation model for predicting the behavior of the molten pool, and the predicted values ​​of the molten pool state parameters within a preset time period are output. The composite oscillation trajectory parameters corresponding to the predicted values ​​of the molten pool state parameters are matched from the preset database. The composite oscillation trajectory parameters include macroscopic reference trajectory parameters and microscopic modulation trajectory parameters used to define the composite motion trajectory of the welding torch tip. When the welding torch enters the preset dwelling area of ​​the weld swing edge according to the macro reference trajectory defined by the macro reference trajectory parameters, the tip of the welding torch is controlled to perform an asymmetric spiral oscillation motion in the preset dwelling area according to the micro modulation trajectory defined by the micro modulation trajectory parameters until it leaves the preset dwelling area. The trajectory shape of the asymmetric spiral oscillation motion is asymmetrically distributed with respect to the swing direction. The step of controlling the welding torch tip to perform asymmetric helical oscillation motion in the preset dwell area based on the micro-modulation trajectory defined by the micro-modulation trajectory parameters further includes: Real-time monitoring of the molten pool width variation within the preset dwell zone, the molten pool width variation reflecting the degree of heat accumulation in the edge region; When the change in the width of the molten pool exceeds a preset threshold, the amplitude attenuation coefficient of the spiral oscillation motion is adjusted according to a preset ratio. The amplitude attenuation coefficient is used to control the deceleration rate of the spiral motion radius. The dwell time of the welding torch tip in the preset dwell area is adjusted according to the adjusted amplitude attenuation coefficient.

2. The method according to claim 1, characterized in that, The steps for acquiring real-time multiphysics sensing data specifically include: The surface image data of the molten pool is acquired by a visual sensor, and the molten pool surface image data includes the molten pool outline, surface texture and liquid metal flow characteristics. Temperature field data of the welding area is acquired by an infrared thermal imaging sensor. The temperature field data of the welding area includes the center temperature of the molten pool, the temperature gradient distribution, and the range of the heat-affected zone. Welding electrical parameter data is monitored by current and voltage sensors, including instantaneous current value, voltage fluctuation amplitude, and arc length change rate. The molten pool surface image data, welding area temperature field data, and welding electrical parameter data are processed in time synchronization to generate time-aligned multiphysics sensing data.

3. The method according to claim 1, characterized in that, The step of inputting the real-time multiphysics sensing data into a simulation model for predicting molten pool behavior and outputting predicted values ​​of molten pool state parameters for a future preset time period specifically includes: The real-time multiphysics sensing data is subjected to noise reduction preprocessing to generate standardized sensing data, which is normalized data after removing outliers and noise interference. Extract the time-series feature vector from the standardized sensor data. The time-series feature vector contains trend information and periodic patterns of dynamic changes in the molten pool. The time-series feature vector is input into a deep learning-based molten pool behavior prediction model, which is a neural network model trained on historical welding data. The molten pool behavior prediction model calculates the geometric parameters, temperature distribution parameters, and flow state parameters of the molten pool within a preset time period in the future, and generates predicted values ​​of the molten pool state parameters including confidence scores.

4. The method according to claim 1, characterized in that, Before the step of matching the composite oscillation trajectory parameters corresponding to the predicted values ​​of the molten pool state parameters from a preset database, the method further includes: A mapping relationship library between welding process parameters and oscillation trajectory parameters is constructed, and the mapping relationship library contains the optimal trajectory parameter combinations under different materials, plate thicknesses and welding positions; The trajectory parameter search space is determined based on the mapping relationship library corresponding to the material properties and structural features of the current welding task. The trajectory parameter search space is the range of parameter values ​​that satisfy the process constraints. A fuzzy matching algorithm is used to find historical cases within the trajectory parameter search space that are closest to the predicted values ​​of the molten pool state parameters. These historical cases include verified successful welding parameter combinations. Based on the historical cases, a set of candidate composite swing trajectory parameters is generated, and the composite swing trajectory parameters are determined from the set of candidate composite swing trajectory parameters according to preset selection rules.

5. The method according to claim 1, characterized in that, After the step of controlling the welding torch tip to perform asymmetric helical oscillation motion in the preset dwell area according to the micro-modulation trajectory defined by the micro-modulation trajectory parameters, the method further includes: A trajectory optimization feedback mechanism based on reinforcement learning is established, which includes reward function design and policy network update. The weld formation quality indicators during the execution of asymmetric helical oscillation motion are collected. The weld formation quality indicators include weld reinforcement uniformity, undercut depth and surface ripple coefficient. Calculate the comprehensive reward value for the current oscillation cycle, which is composed of a weighted average of weld formation quality index, heat input uniformity index, and edge fusion state index; When the comprehensive reward value is lower than the historical average, the control strategy of the micro-modulation trajectory is updated by the strategy gradient algorithm to generate optimized spiral oscillation parameters. The optimized spiral oscillation parameters are applied in real time to the next oscillation cycle.

6. The method according to claim 1, characterized in that, Before the step of the welding torch entering the preset dwell zone of the weld oscillation edge according to the macro reference trajectory defined by the macro reference trajectory parameters, the method further includes: The spatial boundary of the preset residence area is determined based on the geometric features of the weld and the heat accumulation distribution characteristics. The spatial boundary defines the key area where spiral oscillation control needs to be performed. By analyzing the defect distribution patterns in historical welding data, high-risk locations in the oscillation trajectory are identified. These high-risk locations are areas prone to undercut and incomplete fusion defects. Based on the current welding speed, oscillation amplitude, and material thermophysical properties, calculate the critical heat accumulation radius of the oscillation edge, which is the distance threshold at which overheating symptoms begin to appear at the edge of the molten pool; Using the high-risk location as the center and the critical heat accumulation radius as the benchmark, a dynamic residence zone boundary function is established. The dynamic residence zone boundary function outputs the residence zone range that is adjusted in real time with the welding process. The dwell area is divided into a core dwell area and a transition buffer zone. The core dwell area performs a complete spiral oscillation motion, while the transition buffer zone performs a smooth transition motion with gradually changing amplitude.

7. A composite oscillating welding control system, characterized in that, The system includes: one or more processors and memory; The memory is coupled to the one or more processors, the memory being used to store computer program code, the computer program code including computer instructions, the one or more processors invoking the computer instructions to cause the system to perform the method as described in any one of claims 1-6.

8. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is executed on the composite oscillating welding control system, the system performs the method as described in any one of claims 1-6.

9. A computer program product, characterized in that, When the computer program product is run on the composite oscillating welding control system, the system performs the method as described in any one of claims 1-6.

Images