Multi-station automatic welding cooperative control method for electric iron accessories

Abstract

The invention relates to the technical field of automatic welding control, and particularly discloses a multi-station automatic welding cooperative control method for electric iron accessories, which realizes full-closed-loop accurate control of a welding process through multi-sensor data fusion and an intelligent optimization algorithm. According to the method, an infrared thermal imager, an acoustic emission sensor and an ultrasonic thickness measuring device are adopted to collect molten pool temperature field, sound wave frequency spectrum and fusion depth data in real time, after feature extraction, the data are input into a multi-target optimization model, and a Pareto optimal solution set is generated with the fusion depth, the width of a heat affected zone and phase change latent heat as optimization targets; dynamically adjusting a segmented welding-back path and process parameters on the basis of an optimization result, and triggering model iterative correction through ultrasonic online monitoring; the central control unit adopts space voxelization monitoring and a three-level collision avoidance strategy to coordinate operation of the multiple mechanical arms, and collision emergency protection is achieved in combination with flexible pressure sensing.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automatic welding control, in particular to a multi-station automatic welding collaborative control method for power iron accessories. BACKGROUND

[0002] With the development of the power industry, the welding demand of power iron accessories (such as tower connectors, supports, etc.) is increasing. The traditional welding method mainly relies on manual operation or semi-automatic equipment, which has low efficiency and unstable quality. Although multi-station automatic welding technology can improve efficiency, it still faces many challenges in practical application: the welding process involves multiple quality indicators such as penetration, heat-affected zone, and residual stress, which need to be coordinated in real time; at the same time, multi-robot collaborative work is prone to interference or collision, affecting production safety. Existing automatic welding systems mostly use single sensors for monitoring and fixed parameter control, which cannot adapt to the diversified structural characteristics and strict quality requirements of power iron accessories, and there is an urgent need to develop more intelligent collaborative control methods.

[0003] The existing technology has the following shortcomings: Multi-objective collaborative optimization is insufficient: traditional control methods usually take penetration or weld appearance as a single optimization target, which is difficult to consider key indicators such as heat-affected zone width and latent heat of phase change, resulting in substandard mechanical properties of welded joints; Real-time regulation and control capability is limited: existing systems rely on preset process parameters and lack online adjustment mechanisms based on molten pool dynamic characteristics, which have poor adaptability when faced with dissimilar steel welding or complex working conditions; Multi-station collaborative safety is low: mechanical arm collision avoidance mostly uses static spatial partitioning or offline path planning, which cannot dynamically respond to sudden interference and is prone to shutdown or equipment damage due to trajectory conflicts; Sensory data fusion is insufficient: temperature, acoustic wave, and other sensory information are often processed independently, without establishing a cross-modal correlation model, resulting in delayed defect detection or misjudgment. These problems seriously hinder the quality and efficiency improvement of power iron accessory welding. SUMMARY

[0004] The purpose of the present application is to provide a multi-station automatic welding collaborative control method for power iron accessories to solve the problems in the background.

[0005] The purpose of the present application can be achieved by the following technical solutions: A multi-station automatic welding collaborative control method for power iron accessories, comprising the following steps: Real-time acquisition of thermal imaging data and arc acoustic emission signals during welding at each station, extraction of molten pool temperature distribution characteristics and acoustic frequency spectrum characteristics; The molten pool temperature distribution characteristics and the acoustic spectrum characteristics are input into a multi-objective optimization model, the model takes the penetration, the heat affected zone width and the material phase change latent heat as optimization targets, and generates a Pareto optimal solution set through a non-dominated sorting algorithm; According to the Pareto optimal solution set, the welding path of the corresponding station is dynamically adjusted, and the process parameters including current, voltage and welding speed are optimized through a segmented back welding strategy to optimize the heat input distribution; The actual penetration is monitored online through an ultrasonic thickness measuring device, when the deviation between the detected value and the target value exceeds the set threshold, the iterative calculation of the multi-objective optimization model is triggered, and the corrected welding parameters are output and synchronized to the corresponding station; The central control unit uniformly issues the optimized welding instructions of each station, and ensures that the multi-robot completes the collaborative welding operation under the anti-collision constraint condition.

[0006] As a further scheme of the present application, the heat imaging data acquisition and processing method specifically comprises: A short-wave infrared thermal imager is used to acquire the temperature field distribution of the molten pool area at a sampling frequency of not less than 1000 frames per second, and the molten pool area is divided into a circular monitoring area with a diameter of 20 mm with the welding arc as the center; The space-time domain filtering processing is performed on the continuous 10 frames of temperature field data, the random noise is eliminated through three-dimensional Gaussian filtering first, and then the time domain difference method is used to extract the temperature gradient change characteristics; A molten pool three-dimensional temperature field reconstruction model is established, the two-dimensional thermal image sequence is coupled with the welding speed parameter, the temperature decay curve along the welding seam advancing direction is calculated, and the temperature drop rate is extracted as the first characteristic parameter; The temperature drop rate is compared with the preset material thermal conductivity coefficient threshold value, and when the rate deviation exceeds 15%, the welding parameter correction flag is triggered.

[0007] As a further scheme of the present application, the acoustic spectrum characteristic extraction method specifically comprises: A wideband acoustic emission sensor is installed at the conductive nozzle of the welding torch to collect the original acoustic signals in the frequency band of 100 kHz to 500 kHz; The acoustic signal is decomposed by 16 layers through an improved wavelet packet decomposition algorithm, and the 7th to 9th layer detail coefficients are selected as the characteristic analysis frequency band; The standard deviation and kurtosis coefficient of the acoustic energy in the characteristic frequency band are calculated, when the standard deviation exceeds the preset threshold value and the kurtosis coefficient is less than 3, it is determined that the molten pool flow is in an unstable state; A nonlinear mapping relationship between the acoustic characteristics and the penetration is established, the acoustic energy distribution characteristics are converted into the penetration prediction value through a deep neural network model, and the prediction value is used to correct the welding speed parameter in real time.

[0008] As a further scheme of the present application: the construction method of the multi-objective optimization model specifically comprises: A molten depth prediction sub-model is established, the highest temperature value and temperature gradient change rate in the molten pool temperature distribution characteristics are nonlinearly mapped with the welding current parameter, and a molten depth prediction value is output through a deep neural network; A heat affected zone width evaluation sub-model is constructed, the main frequency band energy proportion in the acoustic wave frequency spectrum characteristics is associated with the welding speed parameter, and a support vector regression algorithm is used to calculate the heat affected zone width; A phase change latent heat calculation sub-model is designed, based on the phase change kinetics parameters in the material database, combined with the real-time collected cooling rate, the phase change latent heat value in the welding process is calculated; The outputs of the three sub-models are used as target vectors, and a multi-objective optimization is performed by using an improved non-dominated sorting algorithm, and the improved algorithm preferentially retains individuals with phase change latent heat values lower than a critical threshold in the population selection stage.

[0009] As a further scheme of the present application: the generation method of the Pareto optimal solution set specifically comprises: The constraint conditions of the three targets of molten depth, heat affected zone width and phase change latent heat are set, wherein the lower limit of the molten depth is 40% of the plate thickness, the upper limit of the heat affected zone width is 5 mm, and the upper limit of the phase change latent heat is 80% of the critical value of the material; An adaptive grid method is used to divide the target space, and each grid cell records the current optimal solution set, and when a new solution is better than the existing solution in the grid, the new solution is replaced; An elite reservation strategy is introduced, and the top 10% of high-quality solutions are reserved in each iteration process, and the characteristic parameters thereof are used as the initialization conditions of the next generation population; When the improvement amplitude of the optimal solution set is less than 1% for three consecutive generations, the iteration is terminated, and the final Pareto front solution set is output as the welding parameter optimization scheme.

[0010] As a further scheme of the present application: the optimization method of the segmented rewelding strategy specifically comprises: According to the heat input distribution parameters in the Pareto optimal solution set, the continuous weld is divided into a plurality of sub-weld segments with a length of 30 to 50 mm; A reverse welding sequence planning is used, so that the welding directions of adjacent sub-weld segments are opposite, and a cooling interval of 5 to 8 seconds is set between adjacent weld segments; A current pulse of 1.5 times the normal value is preset at the starting position of each sub-weld segment to form a local remelted area to eliminate the intersegment bonding defects; The temperature field uniformity in the segmented welding process is monitored in real time, and when the temperature difference between adjacent regions exceeds 50 degrees Celsius, the length and welding direction of the subsequent weld segment are automatically adjusted.

[0011] As a further scheme of the present application, the dynamic adjustment method of the process parameters specifically comprises: A corresponding relationship library of welding current and molten pool oscillation frequency is established, when the frequency spectrum characteristic of the acoustic wave shows that the frequency deviation exceeds 5% of the reference value, the current output is adjusted by 0.5 ampere step; The voltage parameter is adjusted according to the real-time collected arc light intensity signal, and the fluctuation amplitude of the light intensity is kept within 10% of the average value; A variable step control algorithm is used to adjust the welding speed, and the welding is started at 80% of the standard speed in the initial stage, and the optimized speed is gradually increased after the detection of the qualified penetration depth; A parameter adjustment priority mechanism is set, when multiple parameters need to be adjusted at the same time, the parameter correction with the greatest influence on the latent heat of phase change is preferentially executed.

[0012] As a further scheme of the present application, the penetration depth monitoring method of the ultrasonic thickness measuring device specifically comprises: A double-chip focusing probe is used to emit ultrasonic waves to the molten pool area in a tilted incidence manner, the center frequency of the probe is set to 5 megahertz, and the incidence angle is controlled between 35 and 45 degrees; A feature extraction model of the time domain reflection signal is established, the time difference between the bottom echo and the fusion surface echo is extracted through wavelet transform, and the real-time penetration depth value is calculated in combination with the material sound velocity; A dynamic threshold adjustment mechanism is designed, when the fluctuation amplitude of the penetration depth in three consecutive detection periods is less than 0.1 millimeter, the set threshold is tightened from 0.5 millimeter to 0.3 millimeter; Three equidistant measurement points are set in the cross-sectional direction of the weld, and the range of the penetration depth values of the points is used as the welding stability criterion, and when the range exceeds 0.4 millimeter, the whole station parameter review is triggered.

[0013] As a further scheme of the present application, the iteration triggering mechanism of the multi-objective optimization model specifically comprises: An association matrix of the penetration depth deviation and the process parameters is constructed, when the deviation direction is contrary to the current parameter adjustment trend, the emergency iteration calculation is started; The width target of the heat affected zone is frozen in the iteration process, the two targets of the penetration depth and the latent heat of phase change are preferentially optimized, and the calculation time is controlled within 0.5 seconds; An incremental parameter updating strategy is used, the correction amount is limited within 15% of the original parameter value, and the optimal solution is gradually approached in three stages; A parameter adjustment effect prediction model is established, the penetration depth change trend after the correction is predicted before synchronization to the station, and when the prediction effect is not up to the standard, the iteration calculation is restarted.

[0014] As a further scheme of the present application, the collision avoidance control method of the multi-mechanical arm cooperative operation specifically comprises: A three-dimensional dynamic monitoring model based on spatial voxelization is established, each robot workspace is divided into cubic units with a side length of 5mm, and the occupied space units are marked in real time; A prospective trajectory prediction algorithm is used to calculate the spatial occupation change within 3 seconds in the future according to the current motion parameters of each robot, and a warning signal is generated when the risk of space unit overlap is detected; A hierarchical collision avoidance strategy is designed, the primary collision avoidance is realized by adjusting the welding speed to realize the trajectory fine adjustment, the intermediate collision avoidance adopts local path re-planning, and the high-level collision avoidance triggers the whole station pause and re-distributes the task priority; A flexible pressure sensor is installed at the end effector of the robot, and when the unexpected contact force detected is more than 2 Newton, the emergency retreat program is started immediately, and the retreat path is executed along the original motion trajectory in reverse.

[0015] The beneficial effects of the application are: (1) The application realizes a revolutionary breakthrough in the welding process by constructing a multi-modal perception-intelligent decision-precise execution closed-loop control system. The system innovatively deepens the fusion of three types of heterogeneous sensors, short-wave infrared thermal imaging (1000Hz sampling), wide-band acoustic emission sensing (100-500kHz) and ultrasonic thickness measurement (5MHz focused probe), extracts 23-dimensional feature vectors such as molten pool temperature gradient and acoustic frequency spectrum energy distribution through space-time domain filtering and wavelet packet decomposition; After these features are input into the multi-objective optimization model based on the improved NSGA-III algorithm, the molten depth (accuracy ±0.1mm), the heat affected zone width (controlled to be less than or equal to 5mm) and the phase change latent heat (error less than 5%) three key indicators can be optimized at the same time, and a Pareto optimal solution set is generated within 0.5 seconds; In the execution stage, the segmented back welding strategy (30-50mm welding segment length) and dynamic parameter adjustment (current step 0.5A, voltage adjustment 0.1V) are cooperatively controlled, especially when the ultrasonic detects that the molten depth deviation exceeds 0.3mm, the model is iterated online and generates a correction instruction, the whole process does not need human intervention. This "perception-decision-execution-feedback" closed-loop architecture makes the system have process self-adaptive ability similar to experienced welders, but breaks through the physiological limit of human in response speed (millisecond level), multi-parameter cooperation (more than 6 dimensions) and continuous stability, and improves the welding process from traditional "open-loop experience control" to "closed-loop intelligent optimization" to a new dimension.

[0016] (2) The multi-station collaborative system constructed by the application realizes intelligent collaborative operation in a complex welding scene through innovative spatial digital modeling and hierarchical safety protection mechanism. The system uses a 5mm-precision three-dimensional voxelized spatial modeling technology to divide the entire workspace into millions of voxel units, and tracks the motion trajectory of each robot arm in real time through an octree data structure; combined with a multi-dimensional dynamic prediction algorithm (considering 6 parameters such as load inertia and acceleration limit), it can predict collision risks 3 seconds in advance and generate a three-level response strategy: velocity adjustment (±30% dynamic adjustment) to solve long-term risks; local path re-planning (3 candidate path optimization options) to deal with medium-term conflicts; and full-system emergency braking (response time less than 100ms) to handle sudden situations. At the hardware level, a 16-point ring-shaped pressure sensing array (0.1N resolution) is integrated at the end of each robot arm, which triggers a reverse retreat program (50% speed along the original trajectory for 10-50mm) when detecting abnormal contact force of 2N or more. This system realizes precise synchronization of multiple stations through the distributed decision-making architecture of the central controller (20ms refresh cycle), enabling 4-6 welding robots to safely collaborate in a 0.5m² dense space, reducing the footprint of traditional welding units by 40% while reducing equipment collision accidents to zero, and truly achieving safe and efficient production in a high-density layout. BRIEF DESCRIPTION OF DRAWINGS

[0017] The application will be further described below with reference to the drawings.

[0018] Figure 1 is a flow chart of the multi-station automatic welding collaborative control method for power iron accessories of the application. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the application.

[0020] Embodiment 1, please refer to Figure 1 The application is a multi-station automatic welding collaborative control method for power iron accessories, comprising the following steps: Real-time acquisition of thermal imaging data and arc acoustic emission signals during welding at each station, extraction of molten pool temperature distribution characteristics and acoustic spectrum characteristics; Input the molten pool temperature distribution characteristics and acoustic spectrum characteristics into a multi-objective optimization model, the model takes the penetration depth, heat-affected zone width and material latent heat of phase change as the optimization objectives, and generates a Pareto optimal solution set through a non-dominated sorting algorithm; According to the Pareto optimal solution set, the welding path of the corresponding station is dynamically adjusted, and the welding path is optimized by a segmented back-welding strategy to optimize the heat input distribution, and the process parameters including current, voltage and welding speed; The actual penetration depth is monitored online by an ultrasonic thickness measuring device. When the deviation between the detected value and the target value exceeds the set threshold, the iterative calculation of the multi-objective optimization model is triggered, and the corrected welding parameters are output and synchronized to the corresponding station. The central control unit uniformly issues optimized welding instructions for each station to ensure that the multi-robot completes collaborative welding operations under collision avoidance constraints.

[0021] In the thermal imaging data acquisition stage, the system configures a short-wave infrared thermal imager as the core temperature measurement device. The thermal imager continuously acquires the temperature field distribution of the molten pool area at a sampling frequency of 1000 frames per second. The monitoring area is set as a circular range with a diameter of 20 millimeters centered on the welding arc. This high-frequency sampling can completely capture the dynamic change process of the molten pool, providing high spatiotemporal resolution raw data for subsequent analysis. The installation position of the thermal imager is precisely calibrated to ensure that its optical axis forms a 45-degree angle with the welding torch axis, avoiding direct arc interference and fully covering the active area of the molten pool.

[0022] The collected continuous temperature field data needs to be strictly filtered. The system implements three-dimensional Gaussian filtering on the continuous 10 frames of thermal image data. First, two-dimensional Gaussian smoothing is performed in the spatial dimension to eliminate random noise in single-frame images. Then, one-dimensional Gaussian filtering is performed in the time dimension to suppress frame-to-frame jitter noise. After filtering, the time-domain difference algorithm is used to calculate the temperature change rate of adjacent frame pixel points, extracting the temperature gradient change characteristics reflecting the dynamic characteristics of the molten pool. This processing process effectively improves the signal-to-noise ratio, making the small temperature change characteristics prominent.

[0023] Based on the pre-processed temperature field data, the system constructs a molten pool three-dimensional temperature field reconstruction model. This model couples the two-dimensional thermal image sequence with real-time welding speed parameters in space and time, calculates the temperature drop rate representing heat conduction efficiency by analyzing the temperature decay characteristics along the welding seam direction. In specific implementation, the system selects equidistant sampling points in the welding seam length direction, records the temperature change curve of each point over time, and obtains the temperature drop rate value through curve fitting. This parameter can sensitively reflect the balance between welding heat input and material heat dissipation.

[0024] The temperature drop rate parameter will be compared in real time with the thermal conductivity threshold value in the material database. When the rate deviation exceeds 15%, the system automatically triggers the welding parameter correction flag. This threshold is pre-set according to the process test data of different material combinations, for example, the reference rate range of Q235B steel is 35-50°C / s. After the correction flag is triggered, the system will start the subsequent multi-objective optimization process to adjust the current welding parameters.

[0025] In terms of acoustic signal acquisition, the system integrates a wideband acoustic emission sensor at the welding gun electrode nozzle. The sensor's working frequency band is set to 100-500 kHz, which can capture the characteristic acoustic signals during the welding process. The sensor uses a high-temperature alloy shell for protection and is equipped with an adaptive gain control circuit inside to ensure stable signal amplitude under different welding currents. During signal acquisition, the system monitors the signal-to-noise ratio in real time and automatically starts the band-pass filtering function when environmental interference is too large.

[0026] The collected original acoustic signals need to be processed by feature extraction. The system uses an improved wavelet packet decomposition algorithm to decompose the signal into 16 levels. The 7th to 9th layer detail coefficients are selected as the feature analysis frequency band, which contains the most characteristic information reflecting the change of the molten pool state. During the decomposition process, an adaptive threshold denoising technique is used to maximize the elimination of interference while preserving effective features. After decomposition, the system calculates the energy standard deviation and kurtosis coefficient of each feature band, which can quantify the stability and peak characteristics of the acoustic signal.

[0027] Based on the acoustic feature parameters, the system implements molten pool state judgment. When the energy standard deviation of a certain frequency band exceeds the pre-set threshold (such as 20% of the reference value) and the kurtosis coefficient is less than 3, it is determined that the molten pool is in a flow instability state. This judgment standard is established through a large number of process tests, and its accuracy has been verified to be above 92%. The judgment result will be fed back to the control system in real time as an important basis for process adjustment.

[0028] To realize the accurate mapping of acoustic features to penetration depth, the system establishes a deep neural network model. The network uses a five-layer hidden layer structure, with the input layer receiving the energy distribution characteristics of each frequency band and the output layer generating the penetration depth prediction value. During network training, the adaptive momentum optimization algorithm is used, and more than 100,000 sets of labeled data are used to complete model training. In actual application, the system updates the penetration depth prediction value every 50 milliseconds, and when the prediction value deviates from the target value by more than 0.3 mm, a welding speed adjustment instruction is automatically generated. This acoustic feature-based penetration depth monitoring method has obvious advantages over traditional contact-type measurement, enabling truly non-contact real-time monitoring.

[0029] The above parallel processing flow of thermal imaging and acoustic wave signals realizes the organic integration of feature parameters through a data fusion module. The fusion module uses a weighted decision mechanism to dynamically adjust the weight proportion of feature parameters according to the confidence index of each sensor. For example, when the environmental electromagnetic interference is large, the weight of the acoustic wave feature is appropriately reduced; when there is spatter shielding in the molten pool area, the weight of the thermal imaging feature is increased. This adaptive fusion strategy ensures that the system can obtain reliable feature parameters under various working conditions.

[0030] The system also designs a perfect data verification mechanism. For thermal imaging data, the data validity is verified by comparing the temperature correlation of adjacent areas; for acoustic wave signals, the rationality of the energy distribution of each frequency band is checked. When abnormal data is detected, the system can automatically start the data repair program or switch to the standby sensor. These measures significantly improve the reliability of the feature acquisition process.

[0031] In terms of hardware implementation, the system uses a distributed processing architecture. Each welding station is equipped with an independent feature extraction unit responsible for data preprocessing of the station; the central processing unit is responsible for centralized execution of advanced feature analysis and decision generation. This architecture not only guarantees real-time requirements, but also facilitates system expansion. The clock signals of all processing units are strictly synchronized to ensure the time consistency of data from each station.

[0032] In the establishment stage of the molten depth prediction sub-model, the system first constructs a deep neural network architecture. The network uses a three-layer hidden layer design, and the input layer receives three key parameters: the maximum temperature value of the molten pool area, the temperature gradient change rate along the welding direction, and the real-time welding current value. The maximum temperature value is extracted from the thermal imaging data, taking the maximum temperature of 5 consecutive frames in a 20mm diameter monitoring area; the temperature gradient change rate is obtained by calculating the temperature drop per second; the welding current value is obtained from the real-time feedback signal of the power module. The network output layer generates a molten depth prediction value with an accuracy of 0.1mm. During training, more than 100,000 sets of labeled data under different process conditions are used, and an adaptive learning rate algorithm is used to optimize the network weights. In practical applications, the sub-model updates the prediction results every 200ms, providing real-time data support for subsequent optimization.

[0033] The implementation of the heat-affected zone width evaluation sub-model is based on the support vector regression algorithm. The system takes the main frequency band energy proportion in the acoustic wave spectrum feature as the core input parameter, which is obtained by calculating the ratio of the energy in the 100-300 kHz frequency band to the total energy. At the same time, the model introduces the welding speed parameter as an auxiliary input, and the speed data is obtained from the encoder feedback of the servo motor. The support vector regression algorithm uses a radial basis function kernel, and the kernel parameter is determined by cross-validation method. The model training stage collects five thousand sets of experimental data covering different material thickness combinations to ensure the universality of the evaluation results. During the operation of the sub-model, the input parameters are first standardized, and then the support vector machine is used to calculate the heat-affected zone width prediction value, with an output accuracy of 0.2 mm.

[0034] The phase change latent heat calculation sub-model is based on the material database. The system pre-stores the phase change kinetics parameters of various steels, including austenitizing temperature, critical cooling rate and other key indicators. In the real-time calculation stage, the model combines the cooling rate data collected by the infrared thermal imager to determine the phase change latent heat value under the current working condition through interpolation method. The measurement method of cooling rate is to set a monitoring point 20 mm behind the weld, record the time required for the temperature to drop from 800°C to 500°C, and then convert it to degrees Celsius per second. The model has a built-in material identification module that can automatically match the corresponding material data according to the welding process parameters. For the case of dissimilar steel welding, the system uses a weighted average method to handle the phase change characteristics of two materials, and the weight coefficient is dynamically adjusted according to the fusion ratio.

[0035] The integrated operation of the multi-objective optimization model follows a specific process. The system normalizes the output values of the three sub-models to form a target vector, including the penetration prediction value, the heat-affected zone width evaluation value, and the phase change latent heat calculation value. The optimization process uses an improved non-dominated sorting algorithm, which has three key improvements based on the traditional NSGA-II framework: first, in the population initialization stage, the initial solution set is generated according to the historical process data to improve the convergence efficiency; second, in the crossover and mutation operation, uniform crossover is used for welding current parameters and Gaussian mutation is used for welding speed; finally, the phase change latent heat constraint condition is introduced in the selection operator, which preferentially retains individuals with latent heat values below 80% of the critical value of the material. The new solution set generated in each iteration is strictly checked for feasibility to ensure that the parameter combination is within the allowable range of the equipment.

[0036] The generation process of the Pareto optimal solution set adopts an adaptive grid method for management. The system divides the three-dimensional target space into several grid cells, and the size of each cell is dynamically adjusted according to the current solution set distribution. The grid division follows the following principles: the depth dimension is divided by 0.5 millimeters, the heat-affected zone width dimension is divided by 0.2 millimeters, and the latent heat of phase change dimension is divided by 5 joules per square millimeter. Each grid cell maintains an elite solution set, and when a new solution is better than the existing solution in the cell, a replacement operation is performed. The comparison standard uses the Pareto dominance relationship, while considering the degree of satisfaction of the constraint conditions. To maintain population diversity, the system limits each grid cell to retain a maximum of three non-dominated solutions.

[0037] The implementation of the elite retention strategy ensures the inheritance of high-quality genes. At the end of each generation iteration, the system performs a comprehensive evaluation and ranking of all individuals, selecting the top 10% of high-quality solutions as the elite population. The evaluation criteria consider the achievement of the three objectives: the base score is obtained when the penetration depth reaches 40% of the plate thickness, and the score increases by 0.1 millimeter for each additional 0.1 millimeter; the base score is obtained when the heat-affected zone width is less than 5 millimeters, and the score increases by 0.1 millimeter for each 0.1 millimeter decrease; the base score is obtained when the latent heat of phase change is less than 80% of the critical value, and the score increases by 5% for each 5% decrease. The characteristic parameters of the elite solutions will be used as the core parents of the next generation population, generating 80% of the new individuals through crossover operations, and the remaining 20% through random generation to maintain diversity.

[0038] The termination condition of iteration considers both efficiency and quality. The system monitors the improvement of the optimal solution set in three generations in real time, and calculates the average progress of the three objective functions. When the improvement amplitude of the last three generations is less than 1%, the algorithm is considered to have converged. The specific calculation method is to compare the change rate of the Hypervolume index of the solution set in the current generation and the last three generations, which reflects the coverage volume of the solution set in the target space. At the same time, the maximum number of iterations is set to 50 generations to prevent infinite loops. After the iteration is terminated, the system performs post-processing on the final Pareto front solution set, removes solutions that are obviously deviated from the engineering reality, and retains 15-20 groups of the most representative optimization schemes.

[0039] The output and application of the optimization results adopt a hierarchical strategy. The system ranks the Pareto solution set by comprehensive score and recommends the top three as the preferred scheme. Each scheme includes complete process parameter combinations: the welding current value is accurate to 1 ampere, the voltage value is accurate to 0.1 volt, and the welding speed is accurate to 0.5 centimeters per minute. It also provides target prediction values for each scheme, including the penetration depth guarantee range, the heat-affected zone width estimate, and the latent heat of phase change calculation. The operator can select a scheme according to actual needs, or authorize the system to automatically select the scheme with the highest comprehensive score for execution. The selected scheme is sent to each station controller through industrial Ethernet to ensure synchronous updating of multi-station parameters.

[0040] In the implementation phase of the segmented rewelding strategy, the system first analyzes the heat input distribution parameters in the Pareto optimal solution set. According to the parameter combination recommended by the solution set, the continuous weld is intelligently divided into several sub-weld segments, with the length dynamically adjusted within the range of 30 to 50 millimeters. The specific division principle is: for plates with a thickness greater than 12 millimeters, shorter segment lengths are used, and for thin plates, longer segment lengths are used; at the weld corner, the segment length is automatically shortened to improve accuracy. The division process takes into account the material heat capacity characteristics to ensure that the heat accumulation of each sub-weld segment is controlled within the allowed range. The system establishes a weld segment topology relationship diagram to record the spatial position and process parameters of each segment.

[0041] The reverse welding sequence planning adopts a unique spatial coding technique. The system assigns each sub-weld segment a unique spatial coordinate code, and determines the optimal welding sequence by analyzing the coordinate relationship. The welding direction of adjacent sub-weld segments is strictly opposite, i.e., the previous segment uses left-to-right welding, and the next segment changes to right-to-left. This alternating direction design can effectively balance the welding stress distribution. An intelligent cooling interval is set at the intersection of adjacent weld segments, and the system dynamically adjusts the interval time based on real-time infrared temperature measurement data, with 5 seconds for thin plate areas and 8 seconds for thick plate areas. During cooling, auxiliary air cooling devices are started to ensure that the interlayer temperature is controlled within the process requirements.

[0042] The sub-weld segment starting treatment uses high-voltage pulse technology. At the moment of arc starting in each sub-weld segment, the system automatically applies a current pulse of 1.5 times the normal value, with a duration precisely controlled at 0.3 seconds. This pulse parameter is automatically adjusted according to the plate thickness and material, with a maximum of 200% of the rated value. The pulse period is synchronized with the increase in wire feed speed to ensure stable droplet transfer. This treatment forms a local remelted area at the beginning of the segment, with a depth of up to 120% of the normal penetration, effectively eliminating inter-segment bonding defects. The system monitors the molten pool shape through high-speed photography and optimizes the pulse parameters in real time.

[0043] Temperature field uniformity monitoring uses a distributed sensing network. Eight infrared temperature measurement points are symmetrically arranged on both sides of the weld, with a spacing of 10 millimeters. The system scans the temperature distribution every 0.5 seconds, and calculates the temperature difference between adjacent measurement points. When a temperature difference of more than 50 degrees Celsius is detected, the adaptive adjustment mechanism is triggered: for the side with higher temperature, the cooling interval time of the next weld segment is automatically extended; for the side with lower temperature, the welding current is appropriately increased to compensate for the heat input. At the same time, the length of the subsequent weld segment is dynamically adjusted, with a decrease of 5 millimeters for every 10 degrees Celsius of temperature difference, with a maximum adjustment amplitude of 30% of the initial value.

[0044] The closed-loop regulation of welding current is based on acoustic feature analysis. The system establishes a complete database of molten pool oscillation frequency, including benchmark frequency values under different materials and process parameters. During real-time monitoring, the main frequency component of the acoustic signal is analyzed using fast Fourier transform. When the frequency deviation exceeds 5% of the benchmark value, the current adjustment program is started. The adjustment process uses a gradual strategy, with a step size of 0.5 ampere for fine tuning, and the interval between two adjustments is not less than 3 seconds. At the same time, the frequency change trend is monitored. If the frequency continues to deviate after three consecutive adjustments, a process abnormality alarm is triggered.

[0045] The intelligent adjustment of voltage parameters relies on the arc light intensity sensing system. A high dynamic range photoelectric sensor is installed on the side of the welding torch, with a sampling frequency of 2000 Hz. The system establishes a corresponding model of arc light intensity and voltage, and maintains the light intensity fluctuation within 10% of the average value during normal welding. When abnormal light intensity is detected, the voltage is adjusted as follows: when the light intensity suddenly increases, reduce the voltage by 1-2 volts; when the light intensity continues to weaken, increase the voltage by 0.5-1 volt. During the adjustment process, the arc morphology is monitored simultaneously, and the adjustment amplitude is optimized through fuzzy logic algorithm to ensure the stability of the arc.

[0046] The variable step size control of welding speed uses a three-stage adjustment strategy. During the starting stage, the welding speed is uniform at 80% of the standard speed, while the laser range finder tracks the development of the penetration depth in real time. When the penetration depth reaches 90% of the target value, the speed is increased to 95% of the standard value in the medium speed stage. After reaching the target completely, the high speed stage is entered, and the speed is gradually increased to the optimized value. Each stage transition is strictly verified: the speed is only allowed to increase when the penetration depth meets the standard for three consecutive detection periods, and the speed increase amplitude is not more than 5%. The deceleration program is automatically implemented at the end of the weld to prevent the terminal arc pit from being too deep.

[0047] The multi-parameter collaborative adjustment uses a priority decision tree. The system establishes a parameter influence coefficient matrix to quantify the sensitivity of each parameter to the latent heat of phase change. When multiple parameters need to be adjusted simultaneously, the following priority is followed: first, adjust the parameters with an influence coefficient greater than 0.8 on the latent heat of phase change; second, handle the parameters with an influence coefficient between 0.5 and 0.8; finally, fine-tune the parameters with an influence coefficient less than 0.5. After each adjustment, the system state is re-evaluated to ensure that no new unstable factors are introduced. For conflicting adjustment requirements, the expert system is started for arbitration.

[0048] In the configuration of the ultrasonic thickness measurement device, the system selects a double-crystal focused probe as the core detection device. The probe adopts a special acoustic lens design, with a center frequency accurately calibrated at 5 MHz and a bandwidth covering the 3-7 MHz range. The probe is installed on a special bracket 20 mm behind the welding torch, maintaining an optimal incident angle of 35-45 degrees with the workpiece surface. This angle range has been verified through extensive experiments to ensure that the ultrasonic waves have sufficient penetration depth and echo signal strength in the molten pool area. The probe integrates a temperature compensation module, which can automatically correct the speed drift caused by environmental temperature changes. The system sets the probe to emit ultrasonic pulses at a frequency of 100 times per second, with each pulse width controlled at 0.2 microseconds, ensuring energy concentration and good axial resolution.

[0049] The time-domain reflection signal processing uses an advanced wavelet transform algorithm. The original echo signal received by the system is first filtered by a band-pass filter to remove low-frequency mechanical noise and high-frequency electromagnetic interference. Then, the Daubechies wavelet basis is applied for five-layer decomposition, and the feature points of the bottom echo and fusion face echo are accurately extracted from the third layer detail coefficients. The time difference measurement uses a cross-correlation algorithm to slide and compare the two echo signals in the time domain, and when the correlation coefficient reaches 0.95 or above, it is determined as an effective match. Combined with the sound velocity parameters in the material database (such as the longitudinal wave speed of Q345 steel being 5900 m / s), the system calculates the real-time penetration value, with an accuracy of 0.05 mm. To improve reliability, the average value of 10 consecutive waveforms is taken as the final result each time.

[0050] The dynamic threshold adjustment mechanism realizes intelligent fault-tolerant control. The system defaults to an initial threshold of 0.5 mm, and when it detects that the penetration fluctuation amplitude of three consecutive periods (30 ms) is less than 0.1 mm, it determines that the welding process has entered a stable state, and the threshold is automatically tightened to 0.3 mm at this time. This adjustment is realized through a fuzzy logic controller, considering three factors: the standard deviation of the last 10 measurements, the current welding stage (start welding / stable / arc recovery), and the material type. When environmental interference increases or the welding position changes, the system will automatically restore a more relaxed threshold to avoid false alarms. All threshold adjustment records are stored in the process database for subsequent quality traceability.

[0051] Multi-point penetration monitoring adopts a spatial sampling strategy. In the cross-section direction of the weld, three measurement positions are arranged at equal intervals, with the interval automatically adjusted according to the plate thickness (usually 1.5 times the plate thickness). Each measurement point is equipped with an independent ultrasonic channel, and data is collected synchronously. The system calculates the penetration range of the three points in real time, and when the value exceeds 0.4 mm, it is determined that the welding stability is abnormal. At this time, not only is the current station parameter reviewed, but also preventive checks are coordinated by the central controller in other stations. The review process includes recalibrating the ultrasonic probe, verifying the material sound velocity parameters, and checking the welding gun alignment state, to ensure the reliability of the measurement results.

[0052] Penetration deviation analysis uses correlation matrix technology. The system maintains a three-dimensional correlation matrix that records typical penetration variation patterns under different process parameter combinations. The x-axis of the matrix represents the current change direction (increase / decrease), the y-axis represents the voltage change direction, and the z-axis represents the speed change direction. Each unit stores the corresponding penetration deviation trend (increase / decrease / unchanged). When the measured penetration deviation direction conflicts with the current parameter adjustment trend (e.g., current increases but penetration decreases), the system immediately marks it as an abnormal condition and triggers an emergency iteration calculation. This intelligent judgment based on prior knowledge can effectively identify special cases where conventional control strategies fail.

[0053] The optimization model iteration process implements target priority management. In the case of emergency iteration, the system temporarily freezes the optimization of the heat-affected zone width, and concentrates computing resources on the calculation of the two key indicators of penetration and latent heat of phase change. The iteration calculation uses a simplified version of the non-dominated sorting algorithm, with the population size reduced to 30% of the regular size and the generation limit set to within 5 generations. During the calculation process, the weight coefficient of the penetration target is increased to 0.7, and the weight of the latent heat of phase change is set to 0.3, ensuring rapid convergence to a solution that meets the penetration requirements. The entire iteration process is strictly controlled within 0.5 seconds, and real-time response is achieved through multi-thread parallel calculation.

[0054] Parameter updating adopts a gradual adjustment strategy. The system limits each correction to within 15% of the original parameter value and implements it in three stages: the first stage performs a 50% correction, lasting 2 seconds to monitor the response effect; the second stage adds a 30% correction based on feedback; and the final stage performs fine-tuning. This step-by-step approach not only avoids oscillation caused by over-adjustment but also ensures adjustment efficiency. For key parameters such as welding current, a hardware-based ramp-up and ramp-down function is also set to ensure smooth parameter transition and not affect the stability of the power supply.

[0055] The adjustment effect prediction is based on digital twinning technology. The system establishes a simplified simulation model of the welding process, which is virtually verified before the parameter correction scheme is issued. The model inputs the current working condition data and the parameters to be adjusted, and predicts the change curve of the penetration depth in the next 5 seconds within 0.1 seconds. The prediction result is matched with the target value, and when the degree of agreement is less than 85%, it is determined that the prediction is not up to standard and needs to be iterated. The prediction model uses an online learning mechanism to continuously absorb actual welding data to optimize its accuracy, ensuring that the prediction reliability continuously improves as the usage time grows.

[0056] In the construction phase of the spatial voxelized model, the central control unit first establishes a high-precision three-dimensional workspace coordinate system. The coordinate system takes the welding platform reference surface as the XY plane and the vertical direction as the Z axis, with a measurement accuracy of 0.1 millimeters. The system divides the entire workspace into cubic units with a side length of 5 millimeters, each unit assigned a unique spatial code. The voxel division uses an octree data structure for management, automatically subdividing in dense areas and appropriately merging in sparse areas, ensuring accuracy and optimizing computing resources. Each voxel unit records the occupancy state in real time, including the number of occupied robots, the expected occupancy time period, and the safety level. The system updates the voxel occupancy map every 50 milliseconds, using multi-thread parallel computing to ensure real-time performance.

[0057] The prospective trajectory prediction algorithm uses multi-dimensional motion parameter fusion. The system collects the current pose data of each robot (including joint angles, end position, and velocity vector), and calculates the future 3-second motion trajectory through a dynamics model. The prediction process considers five key factors: acceleration limits of each joint servo motor, load inertia of the end effector, geometric constraints of the preset path, uniform speed segment maintenance required by the welding process, and possible abnormal vibration. The prediction result is converted into a voxel occupancy sequence, with two types of dangerous voxels specially marked: one is the conflict voxel occupied by multiple robots at the same time, and the other is the interference voxel that may collide with the workpiece / clamp. The prediction algorithm runs at a cycle of 100 milliseconds, synchronized with the voxel map update.

[0058] The hierarchical collision avoidance strategy implements dynamic risk prevention and control. When the primary warning is triggered (the conflict occurs after 1 second), the system adjusts the welding speed first: under the premise of maintaining welding quality, the speed of the conflicting robot arm is reduced by 10-30%, and the passing time is staggered. The amount of speed adjustment is calculated by an optimization algorithm to ensure that the overall production rhythm is not affected. When the medium-level warning is triggered (the conflict occurs within 0.5 seconds), local path re-planning is started: three candidate paths are generated on both sides of the original path, and the optimal solution is selected based on the three indicators of shortest distance, minimum energy consumption, and process stability. The advanced warning (the conflict is imminent) triggers a full-station pause: the central controller issues an emergency stop command within 0.1 seconds, and all robot arms remain in their current poses and are locked, and then restart according to the new task sequence after manual confirmation of safety.

[0059] The flexible pressure sensing system implements the last line of defense protection. A ring-shaped pressure sensor array is installed at the connection between the robot end effector and the welding torch, with a total of 16 sensing units evenly distributed. The sensor uses a piezoresistive principle, with a range of 0-20 Newton and a resolution of 0.1 Newton. The system monitors the readings of each sensing unit in real time, and when any unit detects an abnormal contact force exceeding 2 Newton, it triggers a three-level response: first, the welding power is cut off to prevent arc damage; second, the reverse retreat program is started, and the robot arm accurately retreats along the original motion trajectory at a speed of 50% of the normal speed; at the same time, a collision alarm is sent to the central controller, including the collision position, force size, and direction vector. The retreat distance is dynamically calculated based on the collision force, usually 10-50 mm, to ensure complete separation from contact.

[0060] The central control unit uses a distributed decision-making architecture. The main controller is responsible for overall coordination, and each station has a sub-controller to handle local real-time control. The command issuance uses a dual-channel redundant design: industrial Ethernet transmits process parameters and path data, and real-time bus transmits synchronization signals and emergency commands. All control commands are attached with time stamps and check codes to ensure strict synchronization of multiple stations. The system maintains a global state table, recording the working mode (welding / collision avoidance / pause) of each robot arm, the current position, and the task progress, which is refreshed every 20 milliseconds. During parameter synchronization, an incremental transmission mechanism is used, only sending the changed part to reduce network load.

[0061] The abnormal handling mechanism enables system self-recovery. When the collision avoidance operation causes a process interruption, the system automatically records the precise position of the interruption point and the process state. When resuming welding, it first reignites 5 mm ahead of the interruption point, uses 90% of the normal parameters for transition, and restores the full parameters after reaching the interruption point. For weld joints caused by collision avoidance, the system automatically marks the position for focused inspection in subsequent quality inspection. All collision avoidance events generate detailed logs, including the time of occurrence, collision avoidance type, handling measures, and result evaluation, which are used for process optimization and system improvement.

[0062] Working principle of the present application: The present application realizes the precise control of the welding process through multi-modal sensor data fusion and intelligent optimization. Firstly, the short-wave infrared thermal imager and wide-band acoustic emission sensor are used to collect the molten pool temperature field distribution and arc sound wave signals in real time during the welding process, and the temperature gradient change rate, sound wave energy distribution and other characteristic parameters are extracted through three-dimensional Gaussian filtering and wavelet packet decomposition; Then these features are input into the multi-objective optimization model, which integrates three sub-models of penetration prediction, heat affected zone evaluation and latent heat calculation, and generates a set of Pareto optimal solutions that meet the constraints of penetration, heat affected zone width and latent heat based on the improved non-dominated sorting algorithm; Based on the optimization results, the system dynamically implements the segmented back-off strategy, controls the heat input distribution through reverse welding order, pulse arc and intelligent cooling, and adjusts the current, voltage and welding speed in real time according to the acoustic frequency spectrum characteristics and arc light intensity signals; During the process, the double-chip ultrasonic probe is used to monitor the penetration depth online, and the model iteration optimization is triggered when the deviation exceeds the limit; Finally, the central control unit coordinates the operation of multiple mechanical arms, ensures the safety of cooperation through spatial voxelization monitoring, three-level collision avoidance strategy and flexible pressure sensing. The scheme realizes the organic integration of multi-objective optimization of welding quality, adaptive adjustment of process parameters and multi-station collaborative control, and significantly improves the efficiency and quality stability of power accessory welding.

[0063] The above describes one embodiment of the present application in detail, but the content described is only the preferred embodiment of the present application, and cannot be considered as limiting the scope of the present application. Any equivalent changes and improvements made within the scope of the present application shall still belong to the scope of the present application.

Claims

1. A collaborative control method for multi-station automatic welding of electric iron accessories, characterized in that: The following steps are involved: Real-time collection of thermal imaging data and arc acoustic emission signals during the welding process of each workstation, and extraction of molten pool temperature distribution characteristics and acoustic wave spectrum characteristics; The temperature distribution characteristics of the molten pool and the acoustic spectrum characteristics are input into a multi-objective optimization model. The model uses the depth of penetration, the width of the heat-affected zone, and the latent heat of the material phase change as optimization targets, and generates a Pareto optimal solution set through a non-dominated sorting algorithm. According to the Pareto optimal solution set, the welding path and process parameters of the corresponding workstation are dynamically adjusted, wherein the welding path optimizes the heat input distribution through a segmented desoldering strategy, and the process parameters include current, voltage, and welding speed; The actual penetration depth is monitored online by an ultrasonic thickness measuring device. When the deviation between the detected value and the target value exceeds a set threshold, the iterative calculation of the multi-objective optimization model is triggered, and the corrected welding parameters are output and synchronized to the corresponding workstation; The central control unit uniformly issues optimized welding instructions to each workstation to ensure that multiple robotic arms can complete collaborative welding operations under collision avoidance constraints.

2. The method for controlling the multi-station automatic welding of electric iron accessories according to claim 1 is characterized in that: The thermal imaging data collection and processing method specifically includes: A short-wave infrared thermal imager is used to obtain the temperature field distribution of the molten pool area at a sampling frequency of not less than 1000 frames per second. The molten pool area is demarcated as a circular monitoring area with a diameter of 20 mm with the welding arc as the center; The temperature field data of 10 consecutive frames were filtered in the spatiotemporal domain. The random noise was first eliminated by three-dimensional Gaussian filtering, and then the temperature gradient change characteristics were extracted by time domain difference method. A 3D temperature field reconstruction model for the molten pool was established. The 2D thermal image sequence was coupled with the welding speed parameter to calculate the temperature decay curve along the weld advance direction, and the temperature drop rate was extracted as the first characteristic parameter. The temperature drop rate is compared with a preset material thermal conductivity threshold, and a welding parameter correction flag is triggered when the rate deviation exceeds 15%.

3. The multi-station automatic welding collaborative control method for electric iron accessories according to claim 1 is characterized in that: The method for extracting the acoustic wave spectrum features specifically includes: A broadband acoustic emission sensor is installed at the welding gun contact tip to collect the original acoustic wave signal in the frequency range of 100 kHz to 500 kHz; The improved wavelet packet decomposition algorithm is used to decompose the acoustic signal into 16 layers, and the detail coefficients from the 7th to the 9th layer are selected as the characteristic analysis frequency band; The standard deviation and kurtosis coefficient of the acoustic wave energy within the characteristic frequency band are calculated. When the standard deviation exceeds the preset threshold and the kurtosis coefficient is less than 3, it is determined that the molten pool flow is unstable. A nonlinear mapping relationship between acoustic wave characteristics and penetration depth is established, and the acoustic wave energy distribution characteristics are converted into penetration depth prediction values ​​through a deep neural network model. The prediction values ​​are used to correct the welding speed parameters in real time.

4. The method for controlling the multi-station automatic welding of electric iron accessories according to claim 1, characterized in that: The method for constructing the multi-objective optimization model specifically includes: A penetration prediction sub-model is established to perform nonlinear mapping between the maximum temperature value and temperature gradient change rate in the molten pool temperature distribution characteristics and the welding current parameters, and the predicted penetration value is output through a deep neural network. A sub-model for evaluating the width of the heat-affected zone (HAZ) was constructed, which correlated the energy proportion of the main frequency band in the acoustic spectrum characteristics with the welding speed parameter, and the support vector regression algorithm was used to calculate the width of the HAZ. Design a phase change latent heat calculation sub-model, based on the phase change kinetic parameters in the material database and the real-time collected cooling rate, to calculate the phase change latent heat value during the welding process; The outputs of the three sub-models are used as target vectors, and an improved non-dominated sorting algorithm is used for multi-objective optimization. The improved algorithm gives priority to retaining individuals with latent heat values ​​below a critical threshold during the population selection stage.

5. The method for controlling the multi-station automatic welding of electric iron accessories according to claim 1, characterized in that: The method for generating the Pareto optimal solution set specifically includes: Set constraints for three targets: penetration depth, heat-affected zone width, and phase change latent heat. The lower limit of penetration depth is 40% of the plate thickness, the upper limit of heat-affected zone width is 5 mm, and the upper limit of phase change latent heat is 80% of the material critical value. Adaptive grid method is used to divide the target space. Each grid cell records the current optimal solution set and replaces the existing solution when the new solution is better than the existing solution in the grid. An elite retention strategy is introduced to retain the top 10% of high-quality solutions in each iteration, and its characteristic parameters are used as the initialization conditions for the next generation population; When the improvement of the optimal solution set of three consecutive generations is less than one percent, the iteration is terminated and the final Pareto front solution set is output as the welding parameter optimization solution.

6. The method for controlling the multi-station automatic welding of electric iron accessories according to claim 1, characterized in that: The optimization method of the segmented desoldering strategy specifically includes: According to the heat input distribution parameters in the Pareto optimal solution set, the continuous weld is divided into several sub-weld segments with lengths ranging from 30 to 50 mm; Use reverse welding sequence planning so that the welding directions of adjacent sub-segments are opposite, and set a cooling interval of 5 to 8 seconds between adjacent weld segments; A current pulse of 1.5 times the normal value is preset at the starting position of each sub-welding segment to form a local remelting zone to eliminate the bonding defects between segments; Real-time monitoring of the temperature field uniformity during segmented welding. When the temperature difference between adjacent areas exceeds 50 degrees Celsius, the length and welding direction of subsequent welding segments are automatically adjusted.

7. The method for controlling the multi-station automatic welding of electric iron accessories according to claim 1, characterized in that: The dynamic adjustment method of the process parameters specifically includes: Establish a corresponding relationship library between welding current and molten pool oscillation frequency. When the acoustic spectrum characteristics show that the frequency deviation exceeds the reference value by 5%, adjust the current output in steps of 0.5 amperes. Adjust the voltage parameters according to the real-time collected arc light intensity signal to keep the light intensity fluctuation range no more than 10% of the average value; The variable step length control algorithm is used to adjust the welding speed. Initially, welding is started at 80% of the standard speed. After the penetration depth reaches the standard, the speed is gradually increased to the optimized speed. Set up a parameter adjustment priority mechanism. When multiple parameters need to be adjusted at the same time, the parameter that has the greatest impact on the phase change latent heat is corrected first.

8. The method for controlling the multi-station automatic welding of electric iron accessories according to claim 1, characterized in that: The penetration monitoring method of the ultrasonic thickness measuring device specifically includes: A dual-chip focusing probe is used to transmit ultrasonic waves to the molten pool area in an oblique incidence mode. The probe center frequency is set to 5 MHz, and the incident angle is controlled between 35 and 45 degrees. A feature extraction model for time-domain reflection signals is established. The time difference between the bottom echo and the fusion surface echo is extracted through wavelet transform, and the real-time penetration value is calculated based on the material sound velocity. Design a dynamic threshold adjustment mechanism to tighten the threshold from 0.5 mm to 0.3 mm when the fluctuation amplitude of the penetration depth is less than 0.1 mm for three consecutive inspection cycles; Three equidistant measuring points are set in the direction of the weld cross section, and the range of the penetration value at each point is used as the welding stability criterion. When the range exceeds 0.4 mm, a full-station parameter review is triggered.

9. The method for controlling the multi-station automatic welding of electric iron accessories according to claim 1, characterized in that: The iterative triggering mechanism of the multi-objective optimization model specifically includes: Construct a correlation matrix between penetration deviation and process parameters. When the deviation direction contradicts the current parameter adjustment trend, initiate emergency iterative calculation. During the iteration process, the heat-affected zone width target is frozen, and the two targets of penetration depth and phase change latent heat are optimized first, with the calculation time controlled within 0.5 seconds; An incremental parameter update strategy is adopted to limit the correction amount to 15% of the original parameter value, and the optimal solution is gradually approached in three stages; A parameter adjustment effect prediction model is established to predict the trend of the corrected melting depth before synchronizing to the workstation. If the prediction effect does not meet the standard, the iterative calculation is restarted.

10. The method for controlling multi-station automatic welding of electric iron accessories according to claim 1, characterized in that: The collision avoidance control method for the multi-manipulator collaborative operation specifically includes: A three-dimensional dynamic monitoring model based on spatial voxelization was established, which divided the workspace of each robotic arm into cubic units with a side length of 5 mm and marked the occupied spatial units in real time. A forward-looking trajectory prediction algorithm is used to calculate the spatial occupancy changes within the next 3 seconds based on the current motion parameters of each robotic arm, and a warning signal is generated when the risk of spatial unit overlap is detected; Design a hierarchical collision avoidance strategy. Primary collision avoidance fine-tunes the trajectory by adjusting the welding speed. Intermediate collision avoidance uses local path replanning. Advanced collision avoidance triggers a pause for all workstations and reallocates task priorities. A flexible pressure sensor is installed at the end effector of the robotic arm. When it detects that the unexpected contact force exceeds 2 Newtons, the emergency retraction program is immediately initiated, and the retraction path is executed in the opposite direction of the original motion trajectory.

Images