An array detection and positioning method combining multi-welding head battery welding

Abstract

This application provides an array detection and positioning method for multi-welding head battery welding, comprising: using infrared thermal imaging technology to monitor the temperature field distribution of the welding head in real time, and judging whether the temperature field of the welding head has shifted in combination with a preset temperature threshold. If the temperature of a certain welding head exceeds the threshold range, it is determined that the temperature field of the welding head has shifted. Based on the magnitude of the welding position shift, the motion control parameters of the digital welding system are adaptively adjusted, and the welding head shift error is compensated in real time through a closed-loop feedback control algorithm. Using online welding point quality monitoring technology, the welding point penetration depth and welding point diameter are extracted, and combined with a pre-established welding point quality evaluation model, the welding point quality is judged in real time to determine whether the welding point quality is qualified. If multiple welding points are unqualified, it is determined that the shift of the welding head temperature field affects the welding point quality.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to an array detection and positioning method combining multi-welding head battery welding. Background Technology

[0002] During prolonged operation of a multi-welding head array, the temperature field of the welding heads shifts due to thermal coupling effects between them. The spacing between the welding heads is a key factor affecting heat conduction; too small a spacing intensifies heat coupling and transfer between the heads, exacerbating the temperature field shift; too large a spacing weakens thermal coupling but increases the size of the welding array. This temperature field shift causes non-uniform thermal expansion deformation of the welding head base, leading to welding position displacement. As operating time increases, this displacement accumulates, significantly impacting the synchronization of the welding array. Inconsistent displacement between different welding heads results in variations in welding parameters and trajectories at each weld point, leading to differences in weld quality and ultimately affecting the overall welding quality of the workpiece. The thermal deformation of the welding head base is closely related to the welding head layout, welding process parameters, base structural design, and heat dissipation conditions. These factors must be considered holistically in the welding head layout design, and the welding process must be optimized to improve heat dissipation and reduce thermal deformation, thereby ensuring consistent weld quality. Summary of the Invention

[0003] This invention provides an array detection and positioning method combining multi-welding head battery welding, mainly including:

[0004] The arrangement spacing and material properties of the welding head array are obtained, and a mathematical model of heat conduction of the welding head array is established based on the arrangement spacing and material properties. The temperature field distribution of the welding head under different arrangement spacings is simulated and calculated by the finite element analysis method to determine the influence law of welding head spacing on heat conduction.

[0005] Infrared thermal imaging technology is used to monitor the temperature field distribution of the welding head in real time. Combined with a preset temperature threshold, it is determined whether the temperature field of the welding head has shifted. If the temperature of a certain welding head exceeds the threshold range, it is determined that the temperature field of the welding head has shifted.

[0006] After determining that the temperature field has shifted, the temperature field shift is calculated by the temperature difference before and after the temperature field shift of the welding head, the thermal expansion coefficient of the welding head base material is obtained, and the thermal expansion deformation of the welding head base under the temperature field shift is calculated by combining the temperature field shift. The displacement deformation of the welding head base is detected in real time by machine vision measurement technology.

[0007] The actual position of the welding head is obtained, and combined with the thermal expansion deformation and temperature field offset, and correlation analysis is performed. Through theoretical derivation and experimental verification, a ternary mathematical model of welding position offset, thermal expansion deformation and temperature field offset is constructed to calculate the welding position offset of each welding head caused by thermal effect in real time.

[0008] Based on the magnitude of the welding position offset, the motion control parameters of the digital welding system are adaptively adjusted, and the welding head offset error is compensated in real time through a closed-loop feedback control algorithm.

[0009] The online welding point quality monitoring technology is adopted to extract the welding point penetration and diameter. Combined with the pre-established welding point quality evaluation model, the welding point quality is judged in real time. If multiple welding points are unqualified, it is determined that the temperature field deviation of the welding head affects the welding point quality.

[0010] Machine learning algorithms are used to perform fusion calculations on the temperature field offset of the welding head, the welding position offset, and the weld quality. A multi-parameter coupled model of the welding process is constructed, and key parameters reflecting the temperature field offset law, welding position offset characteristics, and weld defect mode are extracted. Through data correlation analysis and mechanism deduction, the influence law of the welding head thermal field coupling effect on the welding quality is obtained, and the quantitative relationship between temperature field offset, welding position offset, and weld quality is established.

[0011] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0012] This invention discloses an array detection and positioning method for multi-head battery welding. By establishing a heat conduction model of the welding head array and using infrared thermal imaging to monitor the temperature field distribution of the welding heads, temperature field offset is determined. Based on the temperature field offset and the coefficient of thermal expansion, the thermal expansion deformation of the welding head base is calculated. Machine vision is used to measure the actual position of the welding heads, and a mathematical model of welding position offset and thermal effects is constructed. Welding system parameters are adaptively adjusted according to the offset to compensate for welding head errors in real time. Simultaneously, weld quality is monitored to determine the impact of temperature field offset on the weld. Machine learning algorithms are used to fuse and analyze the temperature field, position offset, and weld quality, establishing a multi-parameter coupled model to reveal the influence of thermal field coupling effects on welding quality. This invention achieves precise compensation for thermal effects during welding, effectively improving welding quality and accuracy. Attached Figure Description

[0013] Figure 1 This is a flowchart of an array detection and positioning method for battery welding with multiple welding heads according to the present invention.

[0014] Figure 2 This is a schematic diagram of an array detection and positioning method for battery welding with multiple welding heads according to the present invention.

[0015] Figure 3This is another schematic diagram of an array detection and positioning method for battery welding with multiple welding heads according to the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0017] like Figure 1 -3, This embodiment of the array detection and positioning method combined with multi-welding head battery welding may specifically include:

[0018] S101. Obtain the arrangement spacing and material properties of the welding head array, and establish a mathematical model of heat conduction of the welding head array based on the arrangement spacing and material properties. Simulate and calculate the temperature field distribution of the welding head under different arrangement spacings using the finite element analysis method, and determine the influence law of welding head spacing on heat conduction.

[0019] The horizontal and vertical spacing values ​​of the welding head array are obtained based on the arrangement of the welding head array. The initial temperature field values ​​of the welding head array surface are acquired using an infrared thermal imager to obtain a temperature distribution function. The thermal conductivity of the welding head material is extracted based on the initial temperature field values. A heat diffusion function is established using the thermal conductivity and thermal impedance, and the heat diffusion function includes the temperature diffusion coefficient. The heat conduction calculation region of the welding head array is divided into square grid cells. The fourth-order Runge-Kutta algorithm is used to solve the heat conduction differential equation within the grid cells to obtain the temperature time variation function. The temperature distribution function and the temperature time variation function are fitted together, and the linear least squares method is used to calculate the correspondence between the welding head spacing and the heat diffusion to obtain the spatial distribution law of the temperature field.

[0020] For example, the spacing values ​​of the welding head array in the horizontal and vertical directions are obtained according to the arrangement of the welding head array. Initial temperature field values ​​on the surface of the welding head array are collected using an infrared thermal imager at a sampling interval of 0.1 seconds. The temperature field gradient value is calculated based on these initial values, and a Taylor series expansion is performed on the temperature gradient value to obtain the temperature distribution function. The thermal conductivity of the welding head material is extracted from the initial temperature field values ​​collected by the thermal imager. The thermal diffusivity is calculated based on the density and specific heat capacity of the welding head material. A heat diffusion function Q = λA(T2-T1) / L, which includes the temperature diffusivity and thermal resistance, is established, where λ is the thermal conductivity, A is the cross-sectional area through which heat flows, T2-T1 is the temperature difference between two points, and L is the heat transfer distance. The heat conduction calculation region of the welding head array is divided into square grid cells with a side length of 0.5 mm. Temperature boundary conditions are set within the grid cells. The fourth-order Runge-Kutta algorithm is used to solve the heat conduction differential equation. The temperature time change function is obtained by iterative calculation of the grid cell node temperature at 0.1-second time intervals. Fitting the temperature distribution function and temperature-time variation function, the linear least squares method is used to calculate the correspondence between the welding head spacing and heat diffusion. From this correspondence, the spatial distribution law of the temperature field is obtained, and a quantitative characterization function of the welding head spacing on heat conduction is established. The welding head array includes lateral and longitudinal spacing in the process parameters. The lateral spacing represents the distance between adjacent welding heads in the horizontal direction, and the longitudinal spacing represents the distance between adjacent welding heads in the vertical direction. The welding head array spacing is usually in the range of 2-10 mm. For the temperature field characteristics of the welding head array, an infrared thermal imager obtains the temperature distribution by detecting the infrared radiation intensity on the welding head surface. When the sampling interval is set to 0.1 seconds, the transient temperature change process can be captured. The temperature field gradient value reflects the rate of temperature change in space. The thermal conductivity of the welding head material describes the strength of the material's thermal conductivity. For copper welding heads, the thermal conductivity is 380 W / m Kelvin at 20 degrees Celsius, and the thermal conductivity decreases with increasing temperature, dropping to 340 W / m Kelvin at 200 degrees Celsius. The thermal diffusivity characterizes the rate at which temperature propagates within a material and is related to the material's thermal conductivity, density, and specific heat capacity. The thermal diffusivity of a copper welding head is approximately 1.1 cm² / s. In the heat conduction calculation of the welding head array, the fineness of the mesh affects the calculation accuracy; a square mesh with a side length of 0.5 mm can better reflect the spatial distribution characteristics of the temperature field. Temperature boundary conditions include the isothermal boundary of the welding head heating surface and the convective heat transfer boundary of the outer surface. The temperature of the welding head heating surface is set to 350 degrees Celsius, and the heat transfer coefficient of the outer surface is taken as 20 W / m² Kelvin. The influence of the welding head spacing on the temperature field distribution is reflected in the thermal interaction between adjacent welding heads. When the welding head spacing is small, the heat transfer between the welding heads is more significant. At a spacing of 3 mm, the temperature difference between adjacent welding heads is less than 10 degrees Celsius, and the temperature field exhibits a relatively uniform distribution.As the spacing increases to 8 mm, the temperature difference between adjacent welding heads increases to 25 degrees Celsius, and the temperature field distribution exhibits obvious localized high-temperature regions. The spatial distribution law of the temperature field is quantitatively described by the correspondence between welding head spacing and temperature gradient; for every 1 mm increase in spacing, the temperature gradient increases by approximately 4 degrees Celsius per millimeter. During heat conduction, the thermal conductivity and thermal resistance of the welding head material jointly determine the efficiency of heat transfer. The higher the thermal conductivity, the faster the heat transfer, while the thermal resistance reflects the resistance during heat transfer. For copper welding heads, at an operating temperature of 350 degrees Celsius, the thermal resistance is approximately 0.15 Kelvin per watt. There is a difference in the heat conduction rate in the transverse and longitudinal directions, with the transverse conduction rate being 15% higher than the longitudinal conduction rate.

[0021] S102. Use infrared thermal imaging technology to monitor the temperature field distribution of the welding head in real time, and combine it with the preset temperature threshold to determine whether the temperature field of the welding head has shifted. If the temperature of a certain welding head exceeds the threshold range, it is determined that the temperature field of the welding head has shifted.

[0022] The surface temperature field image of the welding head array is acquired by an infrared thermal imager, and a Gaussian filter is applied to the temperature field image to obtain a filtered temperature field image. Temperature color values ​​of pixels in the welding head region are obtained from the filtered temperature field image, and temperature gradient values ​​are calculated based on these color values ​​to obtain temperature time-series data. The deviation between the temperature value in the welding head region and the reference temperature is calculated based on the temperature time-series data. When the temperature deviation exceeds a preset range, the welding head position coordinates and temperature deviation value are recorded. The temperature deviation values ​​at the welding head position are accumulated. If the accumulated temperature deviation value exceeds a preset temperature fluctuation threshold, it is determined that the temperature field of the welding head has an abnormal shift, and the coordinates of the welding head center point and the accumulated temperature deviation value are obtained as a temperature anomaly record.

[0023] For example, an infrared thermal imager acquires a surface temperature field image of the welding head array at a sampling interval of 0.1 seconds. The image resolution is set to 640x480 pixels. The temperature field image is then subjected to Gaussian filtering with a kernel size of 5x5 pixels and a standard deviation of 1.5 to obtain a filtered temperature field image. The temperature value and temperature field distribution data of each welding head region are extracted from the filtered temperature field image. The temperature color level values ​​of the pixels in the welding head array region are obtained from the filtered temperature field image. The temperature color level values ​​are divided into 256 gray levels within the range of 200 to 400 degrees Celsius. The temperature gradient value is calculated by scanning the eight neighboring pixels around each welding head region. A temperature time series data record table is constructed based on the temperature gradient values. The temperature fluctuation curve is obtained by statistically analyzing the numerical changes of the temperature time series data within a 0.1-second interval. The deviation between the temperature value of each welding head region and the 350°C reference temperature is calculated based on the temperature fluctuation curve. When the temperature deviation exceeds ±5°C, the coordinates of the welding head position and the temperature deviation value are recorded, and the temperature deviation value of the welding head position is accumulated in real time within a subsequent 0.5-second time window. The accumulated temperature deviation value of each welding head position is compared with a preset temperature fluctuation threshold. If the accumulated temperature deviation value of the welding head position exceeds ±20°C within the 0.5-second time window, it is determined that the temperature field of the welding head has an abnormal shift, and the coordinates of the welding head center point and the accumulated temperature deviation value are extracted as a temperature anomaly record. The infrared thermal imager obtains temperature distribution data by detecting the infrared radiation intensity on the surface of the welding head. The 640x480 pixel resolution can achieve a spatial resolution accuracy of 0.2 mm within a 10 cm x 8 cm welding head array area, and each welding head region contains approximately 400 effective pixels. A sampling interval of 0.1 seconds corresponds to a 10 Hz sampling frequency, meeting the monitoring requirements for rapid temperature changes in the welding head. During the heating phase of the welding head, it can capture a temperature change rate of 20 degrees Celsius per second. Temperature field image denoising uses a 5x5 pixel Gaussian filter kernel with a standard deviation of 1.5, effectively suppressing thermal noise while preserving temperature field edge features. For images at 350 degrees Celsius, the noise amplitude after filtering is reduced from ±2 degrees Celsius to ±0.5 degrees Celsius. Welding head region extraction is based on temperature gradient boundary detection. The temperature difference between the welding head center and the background temperature typically exceeds 100 degrees Celsius, forming a clear temperature boundary. Temperature color mapping uses 256 grayscale levels to represent the temperature range of 200 to 400 degrees Celsius, with a resolution of 0.8 degrees Celsius per grayscale level. The welding head temperature field gradient is calculated through the temperature difference of 8 neighboring pixels. Under normal operating conditions, the temperature gradient from the center to the edge of the welding head is within the range of 25 to 35 degrees Celsius per millimeter. The temperature fluctuation curve records the temperature change within a 0.1-second interval. During stable operation, the temperature fluctuation range should be kept within ±1 degree Celsius. The welding head temperature deviation value is calculated based on a reference temperature of 350 degrees Celsius, and the deviation threshold of ±5 degrees Celsius takes into account the temperature measurement accuracy and the allowable range of the process.The accumulated temperature deviation within a 0.5-second time window reflects the persistence of the temperature anomaly. Instantaneous deviations caused by normal fluctuations cancel each other out after integration, while persistent deviations lead to an increase in the accumulated value. When the accumulated temperature deviation exceeds ±20 degrees Celsius within 0.5 seconds, it indicates that the welding head temperature has been continuously deviating from the normal range. At this time, the recorded welding head coordinates and accumulated temperature deviation value can be used to locate the faulty welding head and determine the degree of temperature deviation. In practical applications, welding head temperature anomalies usually manifest as insufficient heating power leading to low temperatures, or heat accumulation leading to high temperatures. When the temperature is low, it will remain below 345 degrees Celsius within 0.5 seconds, with the accumulated deviation reaching -25 degrees Celsius. When the temperature is high, local heat dissipation is difficult, causing the temperature to remain above 355 degrees Celsius, with the accumulated deviation exceeding +25 degrees Celsius. By monitoring the accumulated temperature deviation value in real time, problematic welding heads can be detected and located in time before the temperature anomaly develops to the point of affecting process quality.

[0024] S103. After determining that the temperature field has shifted, the temperature field shift is calculated by the temperature difference before and after the temperature field shift of the welding head, the thermal expansion coefficient of the welding head base material is obtained, and the thermal expansion deformation of the welding head base under the temperature field shift is calculated by combining the temperature field shift. The displacement deformation of the welding head base is detected in real time by machine vision measurement technology.

[0025] Based on the temperature field values ​​of the welding head acquired by the infrared thermal imager, the temperature difference is calculated by the temperature values ​​of the central region of the welding head within a time window before and after the temperature field anomaly determination time, and the temperature field offset of the welding head base is obtained. Using the linear thermal expansion coefficient value in the copper material property database of the welding head base and the temperature field offset, the theoretical thermal expansion displacement is calculated according to the three-dimensional dimension parameters of the base. The corner feature marker coordinates are extracted from the base surface image acquired by the industrial camera, and the feature marker coordinates are tracked by the sub-pixel edge detection algorithm to obtain the actual deformation of the base surface. The theoretical thermal expansion displacement and the actual deformation are filtered by the Kalman filter algorithm to obtain the three-dimensional spatial deformation vector of the base. If the deformation vector exceeds the positive and negative threshold range, the thermal strain of the base is determined to be abnormal.

[0026] For example, based on the infrared thermal imager acquiring the temperature field values ​​of the welding head at 0.1-second intervals, the temperature values ​​of a 10x10 pixel range in the central region of the welding head are extracted within a 0.5-second time window before and after the temperature field anomaly determination time window. After time-series registration of the temperature values ​​before and after the temperature field anomaly, the temperature difference is calculated, and a three-dimensional temperature distribution function is constructed from the temperature difference.

[0027]

[0028] f(x, y, z) represents the temperature distribution function in three-dimensional space, A is the amplitude, (x0, y0, z0) is the center position of the temperature distribution, and σ is the standard deviation of the distribution, controlling the width of the distribution. The temperature field offset of each region of the welding head base is obtained through the three-dimensional temperature distribution function. Linear thermal expansion coefficient values ​​within the range of 20 to 400 degrees Celsius are obtained from the copper material property database of the welding head base. Based on the three-dimensional dimensions (length, width, height) of the welding head base and the temperature field offset, the theoretical thermal expansion displacement of the base in three-dimensional space is calculated using the linear thermal expansion formula, and this theoretical thermal expansion displacement is updated in real time at 0.1-second intervals. The coordinates of nine corner feature markers are extracted from a 1920x1080 pixel image of the base surface acquired by an industrial camera. A sub-pixel edge detection algorithm is used to track the feature marker coordinates in real time at 0.1-second intervals, calculating the displacement increment of the feature marker coordinates in three-dimensional space. The actual deformation of the base surface is obtained through this displacement increment. For the data sequence of theoretical displacement and actual deformation of the base within a 0.5-second time window, a Kalman filter algorithm is used to filter the displacement data in real time. The deformation vector of the base in three-dimensional space is extracted from the filtering result. If the deformation vector exceeds ±0.1 mm, the base is judged to have abnormal thermal strain. Anomalies in the welding head temperature field are determined by real-time monitoring of temperature differences. The temperature field images acquired by the infrared thermal imager have high spatiotemporal resolution. The 0.1-second sampling interval ensures the capture of transient temperature changes. The 10x10 pixel central area of ​​the welding head contains 100 temperature measurement points, providing a stable and reliable temperature data foundation. When an abnormal heating occurs in the welding head, the temperature field will show significant changes within a 0.5-second time window, with the temperature difference increasing from ±2 degrees Celsius under normal conditions to more than 10 degrees Celsius. The welding head base is made of high-purity copper, whose linear thermal expansion coefficient exhibits non-linear characteristics with temperature. It is 16.5 micrometers per meter per degree Celsius at 20 degrees Celsius, gradually increasing with temperature, reaching 18.5 micrometers per meter per degree Celsius at 400 degrees Celsius. The typical dimensions of the base are 50 mm long, 30 mm wide, and 20 mm high. When the temperature shifts by 50 degrees Celsius, the theoretical thermal expansion in the length direction reaches 41 micrometers, and in the width direction, it reaches 25 micrometers. In the base surface image captured by the industrial camera, nine feature markers are distributed in a 3x3 matrix with a marker spacing of 15 mm. A sub-pixel edge detection algorithm achieves a displacement measurement accuracy of 0.5 micrometers. When the base surface temperature rises, causing thermal expansion and deformation, the coordinates of the markers change, and the distance between adjacent markers increases, forming a non-uniform deformation field. In the initial stage of temperature field shift, the deformation field exhibits a characteristic of central expansion and peripheral contraction. As the temperature continues to rise, the deformation gradually transitions to overall expansion. The Kalman filter algorithm processes the displacement data in real time, with a filtering window set to 0.5 seconds and containing 5 consecutive sampling points. Under normal operating conditions, the thermal expansion deformation of the base remains stable, and the filtered displacement vector fluctuates within a range of ±0.02 mm.When an anomaly occurs in the temperature field, thermal expansion and deformation intensify, with the displacement vector increasing rapidly within 0.5 seconds. In cases of abnormal thermal expansion caused by excessively high welding head temperature, the top region of the base displaces upward by 0.15 mm, while the sides expand outward by 0.12 mm, resulting in uneven three-dimensional deformation. Conversely, thermal contraction caused by excessively low temperature leads to a reduction in the overall size of the base, with the top sinking by 0.08 mm and the sides contracting by 0.06 mm. Thermal deformation monitoring data shows that the base material accumulates strain during cyclic heating, manifested as an increase in the displacement of the marked points with each heating cycle. In the initial 10 heating cycles, the maximum displacement gradually increases from 0.1 mm to 0.13 mm, and the displacement direction changes from perpendicular to the surface to a 45-degree inclination, reflecting the evolution of stress distribution within the base. By tracking the magnitude and direction of the displacement vector in real time, early characteristics of abnormal thermal strain in the base can be identified.

[0029] S104. Obtain the actual position of the welding head, and combine it with the thermal expansion deformation and temperature field offset, and perform correlation analysis. Through theoretical derivation and experimental verification, construct a ternary mathematical model of welding position offset, thermal expansion deformation and temperature field offset, and calculate the welding position offset of each welding head caused by thermal effect in real time.

[0030] Based on the surface image of the welding head acquired by an industrial camera, the center coordinates of the calibration pattern are extracted from the surface image. The camera parameters are then calibrated using these center coordinates to obtain the welding head position deviation vector. Temperature field data of the welding head area is acquired using an infrared thermal imager. The temperature deviation value is calculated based on this data, and the thermal expansion displacement vector is obtained by combining the temperature deviation value with the coefficient of thermal expansion. A Bayesian network model is constructed based on the welding head position deviation vector and the thermal expansion displacement vector. A position compensation vector is extracted from the Bayesian network model. A recursive filtering operation is performed on the position compensation vector to obtain the temperature compensation value. A three-dimensional spatial compensation function is then constructed based on the temperature compensation value.

[0031] For example, based on a 1920x1080 pixel image of the welding head surface acquired by an industrial camera at 0.1-second intervals, the center coordinates of a 9-point calibration pattern on the welding head surface are extracted. The camera's internal and external parameters are calibrated using the Zhang Zhengyou calibration algorithm. The 3D spatial positions of the feature points on the welding head surface in the world coordinate system are calculated through 3D reconstruction. The 3D spatial positions are compared with the coordinates of the machining reference points to obtain the welding head position deviation vector. Temperature field distribution data within a 10x10 pixel range of the welding head area are obtained from an infrared thermal imager. The temperature deviation value of each pixel is calculated for a reference temperature of 350 degrees Celsius. A temperature field spatial offset function is constructed. Combined with the thermal expansion coefficient variation curve of copper material within the range of 20 to 400 degrees Celsius, the 3D thermal expansion displacement vector of the welding head base is calculated, establishing a real-time mapping relationship between the temperature deviation value and the thermal expansion displacement vector. Based on the changes in the welding head position deviation vector and thermal expansion displacement vector within a 0.5-second time window, a coupled model of temperature field offset and spatial displacement is trained using a fifth-order Bayesian network algorithm. The Bayesian network includes position nodes, temperature nodes, and thermal expansion nodes. The position compensation vector is extracted from the coupled model and updated in three-dimensional space at 0.1-second intervals. A three-dimensional spatial recursive filter is used for real-time calculation of the position compensation vector, with a 0.5-second sliding time window. Forward recursive calculation of the position compensation vector is performed within the sliding window, and the temperature compensation value of the actual welding head position is obtained from the recursive calculation results. A three-dimensional spatial compensation function incorporating temperature field changes and thermal expansion deformation is constructed. When the industrial camera monitors the welding head position online, a 9-point calibration pattern is used. The calibration pattern consists of a 3x3 matrix of circular markers with a diameter of 2 mm and a spacing of 15 mm. Sub-pixel-level center coordinates are extracted through these circular markers. Camera calibration employed Zhang Zhengyou's algorithm. The calibration board measured 100x100 mm, and calibration images were acquired from five different viewpoints. The calculated camera focal length was 8 mm, principal point coordinate offset was less than 0.1 mm, and radial distortion coefficient was within 0.01. The surface temperature distribution of the welding head exhibited a characteristic of high temperature at the center and low temperature at the edges. A 10x10 pixel area covered a 100 square millimeter area in the center of the welding head, with each pixel corresponding to an actual size of 1 square millimeter. Using 350 degrees Celsius as the reference temperature, under normal operating conditions, the temperature fluctuation range in the central area was within ±2 degrees Celsius, while the temperature gradient at the edges reached 15 degrees Celsius per millimeter. When the welding head experienced abnormal heating, the central temperature rose to 380 degrees Celsius, and the edge temperature subsequently rose to 365 degrees Celsius, with the temperature gradient decreasing to 8 degrees Celsius per millimeter. The coefficient of thermal expansion of copper exhibits a significant temperature dependence, being 16.5 micrometers per meter per degree Celsius at 20 degrees Celsius, increasing to 17.5 micrometers per meter per degree Celsius at 200 degrees Celsius, and reaching 18.5 micrometers per meter per degree Celsius at 400 degrees Celsius. The thermal expansion displacement of the welding head base in three-dimensional space is anisotropic, with the largest longitudinal expansion at 45 micrometers, followed by the transverse expansion at 30 micrometers, and the smallest vertical expansion at 20 micrometers.The mapping relationship between temperature field shift and thermal expansion displacement shows that for every 10 degrees Celsius increase in temperature, the longitudinal displacement increases by 9 micrometers and the lateral displacement increases by 6 micrometers. The Bayesian network-constructed coupled model of temperature field shift and spatial displacement comprises a three-layer network structure: the input layer receives temperature field data, the middle layer processes thermal expansion calculations, and the output layer generates position compensation. The model was trained using 1000 sets of temperature field and displacement data, achieving a prediction accuracy of ±2 micrometers after convergence. The position compensation vector is updated 50 times within a 0.5-second sliding window, achieving a compensation frequency of 10 Hz. Three-dimensional spatial recursive filtering uses a 5-point filter kernel to smooth the compensation vector, controlling the position fluctuation within ±0.5 micrometers. Under conditions of rapid temperature changes in the welding head, the temperature field increased from 350 degrees Celsius to 380 degrees Celsius in 0.3 seconds, the thermal expansion displacement response lagged by 0.1 seconds, and the position compensation was updated within 0.5 seconds. During the compensation process, the maximum positional deviation reached 55 micrometers, which was reduced to 35 micrometers after recursive filtering. The deviation from the theoretically calculated thermal expansion displacement was less than 5 micrometers. The compensation function maintained a follow-up delay of less than 0.2 seconds to temperature field changes, meeting the real-time compensation requirements for the welding head position.

[0032] S105. Based on the magnitude of the welding position offset, the motion control parameters of the digital welding system are adaptively adjusted, and the welding head offset error is compensated in real time through a closed-loop feedback control algorithm.

[0033] The compensation gain coefficient is calculated by comparing the welding head position offset with a preset compensation threshold using a particle swarm optimization algorithm. A three-dimensional spatial compensation vector is constructed by multiplying the compensation gain coefficient with the position offset, and a compensation response curve containing position, velocity, and acceleration components is generated from the compensation vector. The compensation amount is extracted from the compensation response curve to construct a third-order state-space equation, and a Kalman state observer is used to iteratively calculate the state-space equation to obtain the compensation control amount. The position loop gain parameter, velocity loop gain parameter, and current loop gain parameter are updated according to the compensation control amount. The actual compensation displacement is obtained by adjusting the gain parameters, and the compensation tracking error is obtained by comparing the actual compensation displacement with the theoretical compensation amount. A proportional-integral-derivative (PID) controller is used to perform closed-loop adjustment of the compensation tracking error.

[0034] For example, the compensation gain coefficient is calculated using a particle swarm optimization algorithm based on a comparison between the welding head position offset and a preset compensation threshold of 0.1 mm. The particle swarm size is set to 50, and the number of iterations is 100. A three-dimensional spatial compensation vector is constructed by multiplying the compensation gain coefficient and the position offset. A compensation response curve containing position, velocity, and acceleration components is generated from the compensation vector. The position compensation, velocity compensation, and acceleration compensation are extracted from the compensation response curve. A third-order state-space equation is constructed using the compensation quantities. A Kalman state observer is used to iterate the state-space equation at 0.001-second intervals. The compensation control quantity is obtained from the iteration results and converted into a servo control command. The position loop gain, velocity loop gain, and current loop gain parameters are updated in real time according to the servo control command. The position loop gain ranges from 0.5 to 2, the velocity loop gain ranges from 2 to 8, and the current loop gain ranges from 5 to 20. The compensation control quantity is dynamically tracked within a 0.001-second sampling interval by adjusting the gain parameters to obtain the actual compensation displacement. Numerical comparisons were performed between the actual and theoretical compensation displacements to calculate the compensation tracking error. A proportional-integral-derivative (PID) controller was used for closed-loop adjustment of the compensation tracking error, with a proportional coefficient of 4, an integral time of 0.02 seconds, and a derivative time of 0.005 seconds. The adjustment output was fed back to the compensation gain coefficient calculation unit to achieve iterative optimization of the weld head position compensation. The weld head position compensation control involves a multi-layer feedback adjustment mechanism. A preset compensation threshold of 0.1 mm was used based on the welding process requirements for joint position accuracy; compensation control was activated when the value exceeded this threshold. Particle swarm optimization selected 50 particles, each containing weighted coefficients for position compensation, velocity compensation, and acceleration compensation. The optimal compensation gain combination was found in 100 iterations. When the position offset was 0.15 mm, the optimized compensation gain coefficient was 1.2 in the horizontal direction and 0.8 in the vertical direction, ensuring a linear correspondence between the compensation amount and the offset. The state-space equation describes the dynamic change process of the compensation amount, with position compensation, velocity compensation, and acceleration compensation constituting the third-order state variables. The Kalman observer updates the frequency estimate of the state variables every 0.001 seconds, with the initial value of the prediction covariance matrix set to 0.01 and the measurement noise covariance set to 0.001, enabling rapid tracking of the compensation state. In the case of a sudden 0.2 mm change in the welding head position, the observer converges to the steady-state estimate within 0.005 seconds, with a position estimation error of less than 0.01 mm. Dynamic adjustment of servo control parameters ensures precise execution of compensation commands. The position loop gain starts at 0.5 and gradually increases to 2 as the compensation amount increases, enhancing the position tracking bandwidth. The velocity loop gain is adjusted from 2 to 8, increasing the gain value for larger compensation amounts to accelerate the response speed. The current loop gain varies from 5 to 20 to ensure smooth torque output. In 0.2 mm step displacement compensation, the position loop gain is adjusted to 1.5, the velocity loop gain to 6, and the current loop gain to 15, achieving a settling time of 0.015 seconds.A proportional-integral-derivative (PID) controller forms the outermost compensation control loop. A proportional gain of 4 provides the basic compensation gain, a 0.02-second integral time eliminates steady-state error, and a 0.005-second derivative time improves dynamic response characteristics. The compensation tracking error is obtained by comparing the actual displacement with the theoretical compensation amount. During a positive 0.1 mm displacement compensation process, the overshoot is controlled within 15%, and the steady-state error is less than 0.005 mm. The controller output and particle swarm optimization form a dual compensation loop. During the optimization iteration process, the fitness function of the particle swarm includes three evaluation indicators: overshoot, settling time, and steady-state error. In position offset compensation caused by welding head temperature changes, a 0.05 mm displacement deviation occurs for every 10 degrees Celsius increase in temperature. The compensation response is completed within 0.05 seconds, and the position compensation error is controlled within 0.008 mm. Servo parameters transition smoothly during compensation, avoiding mechanical vibration. The maximum rate of change of the position loop gain is limited to 5 rpm, and the rate of change of the velocity loop gain is limited to 20 rpm, ensuring the dynamic stability of the compensation process.

[0035] S106. Using online welding point quality monitoring technology, the welding point penetration depth and welding point diameter are extracted. Combined with the pre-established welding point quality evaluation model, the welding point quality is judged in real time to determine whether the welding point quality is qualified. If multiple welding points are unqualified, it is determined that the temperature field deviation of the welding head affects the welding point quality.

[0036] The system receives surface images of weld joints captured by an industrial camera, obtains grayscale images through grayscale quantization, extracts the weld joint edge contours using the Sobel operator based on the grayscale images, and obtains sub-pixel coordinates of the weld joint contours using Gaussian curve fitting, resulting in a feature vector containing weld joint diameter, penetration depth, grayscale mean, and contour roundness. Based on the feature vectors, a deep convolutional neural network (DCNN) is used to identify and calculate the weld joint quality score. The DCNN consists of three convolutional layers and two fully connected layers to obtain a standardized quality score. For the standardized quality score, a sliding time window is used to scan the score sequence. If the weld joint quality score is lower than a preset threshold within the time window, it is determined that the temperature field of the weld head corresponding to that time window has an abnormal shift. Based on the feature vectors and quality score data within the temperature field shift time window, a recursive least squares algorithm is used to establish a mapping function from the feature vectors to the quality scores, and the convergence trend of the mapping function is used to determine the temperature field shift pattern.

[0037] For example, an industrial camera captures 1920x1080 pixel images of the solder joint surface at 0.1-second intervals. An 8-bit grayscale quantization process generates a 256-level grayscale image. The Sobel edge detection operator is used to extract the solder joint edge contour. Gaussian curve fitting is used to obtain the sub-pixel coordinates of the solder joint contour, and the solder joint diameter is calculated. The penetration depth feature is extracted from the grayscale value variation curve, constructing a four-dimensional feature vector including the solder joint diameter, penetration depth, grayscale mean, and contour roundness. A solder joint quality score is calculated based on the four-dimensional feature vector. The solder joint diameter threshold range is set between 0.8 and 1.2 mm, the penetration depth threshold range is set between 0.3 and 0.6 mm, and the contour roundness deviation is limited to within 0.05. A deep convolutional neural network is used to identify the feature vector. The neural network contains 3 convolutional layers and 2 fully connected layers, outputting a standardized quality score from 0 to 100. Solder joint quality scores are recorded into a scoring sequence at 0.1-second intervals. A sliding time window is used to scan the scoring sequence in real time, with the window length set to 0.5 seconds and containing 5 consecutive solder joints. If the solder joint quality scores are all below 80 points within the time window, the temperature field of the solder head corresponding to that time window is considered to have an abnormal shift. Based on the solder joint feature vectors and quality score data within the temperature field shift time window, a recursive least squares algorithm with a forgetting factor of 0.95 is used to establish a mapping function from feature vectors to quality scores. The function parameters are updated every 0.1 seconds, and the convergence trend of the mapping function is used to determine the quantitative effect of temperature field shift on solder joint quality. A high-resolution industrial camera is used for solder joint appearance inspection. A 1920x1080 resolution provides 10-micron image accuracy within a 10mm x 8mm field of view, and the 0.1-second sampling interval meets the real-time monitoring requirements of the solder joint formation process. Eight-bit grayscale quantization converts the original image into 256 grayscale levels. The solder joint area typically occupies 180 to 220 grayscale levels, while the background area ranges from 50 to 80 grayscale levels, creating a clear grayscale contrast. Sobel edge detection calculates grayscale gradients in both the horizontal and vertical directions. The gradient value at the solder joint edge reaches 50 grayscale per pixel, significantly higher than the 5 grayscale per pixel of the background noise. Gaussian curve fitting performs sub-pixel interpolation on the edge contour, achieving an edge localization accuracy of 0.5 micrometers. The diameter of a normal solder joint is approximately 1 millimeter. The weld depth feature is extracted from the half-width at half-maximum of the grayscale profile curve, with a standard weld depth of approximately 0.4 millimeters and a contour roundness deviation of less than 0.02. The deep convolutional neural network is trained using 1000 sets of standard solder joint samples. The first convolutional layer uses 16 3x3 kernels to extract edge features, the second convolutional layer uses 32 3x3 kernels to extract texture features, and the third convolutional layer uses 64 3x3 kernels to extract shape features. The fully connected layer converts the feature map into a quality score. Normal solder joints score above 90, minor defects score between 80 and 90, and severe defects score below 80. A sliding time window monitors the solder joint quality trend in real time. The 0.5-second window length covers 5 consecutive solder joints, reflecting the stability of the welding process.When the temperature field shifts, the solder joint diameter changes systematically, shifting from the standard 1 mm to 0.85 mm, and the penetration depth decreases from 0.4 mm to 0.25 mm, causing the quality score to drop below 75 points. When the average score of five consecutive solder joints drops to 70 points, the temperature field anomaly is considered to persist. The recursive least squares algorithm continuously updates the mapping relationship between the temperature field shift and solder joint quality, with a forgetting factor of 0.95 ensuring the algorithm is more sensitive to the latest data. In the initial stage of the temperature field shift, for every 0.1 mm decrease in solder joint diameter, the quality score drops by 15 points, and for every 0.1 mm decrease in penetration depth, the quality score drops by 20 points. As the shift continues, the rate of change in the quality score gradually slows down, demonstrating the nonlinear response characteristics of solder joint quality to the temperature field shift. The solder joint quality evaluation model exhibits differentiated response characteristics to different types of defects. Overmelting caused by excessively high temperatures manifests as a solder joint diameter increasing to 1.3 mm, a penetration depth exceeding 0.7 mm, and a quality score dropping to 65 points. Low temperatures leading to poor solder joints manifest as solder joint diameters less than 0.7 mm and penetration depths less than 0.2 mm, resulting in a quality score dropping to 50 points. This differentiated response helps identify the specific direction and extent of temperature field shifts.

[0038] S107. Machine learning algorithms are used to perform fusion calculations on the temperature field offset of the welding head, the welding position offset, and the weld quality. A multi-parameter coupled model of the welding process is constructed. Key parameters reflecting the temperature field offset law, the welding position offset characteristics, and the weld defect mode are extracted. Through data correlation analysis and mechanism deduction, the influence law of the welding head thermal field coupling effect on the welding quality is obtained, and the quantitative relationship between temperature field offset, welding position offset, and weld quality is established.

[0039] A feature vector sequence is constructed based on the temperature field offset, weld head displacement, and weld quality score. This feature vector sequence is then used to obtain a coupled response function via a bidirectional long short-term memory network. Principal component dimensionality reduction is performed on the output data of the coupled response function. Temperature gradient, displacement increment, and acceleration components are obtained using a feature contribution rate threshold. These components constitute a displacement response curve under thermo-coupling. The displacement response curve is used to perform temporal registration of the temperature field offset and weld head displacement. A three-dimensional quality response surface is constructed using the temporal registration result and the weld quality score. This three-dimensional quality response surface is then used to obtain a mapping function between welding parameters and quality characteristics via cubic spline interpolation. A state transition matrix is ​​constructed based on this mapping function. This state transition matrix is ​​then identified online using a recursive least squares method. The online identification results reveal the coupled evolution law of temperature field offset, weld head displacement, and weld quality.

[0040] For example, based on synchronous data collected by an infrared thermal imager, an industrial camera, and a weld joint detection device, a 12-dimensional feature vector is constructed, including temperature field offset, weld head displacement, and weld joint quality score. The feature vector sequence is recorded at a 0.1-second sampling interval. A bidirectional long short-term memory network is used to extract the temporal correlation of the feature sequence. The network contains two hidden layers, each with 128 neurons, and the training sample length is set to 50 time steps. The coupled response function of temperature field offset and displacement change is obtained from the training results. Principal component dimensionality reduction is performed on the output data of the coupled response function, and a feature contribution rate threshold of 0.95 is set. Key variables of temperature field offset and displacement change, including temperature gradient, displacement increment, and acceleration components, are extracted. A displacement response curve under thermo-mechanical coupling is constructed. A quantitative correspondence between temperature field offset and weld head displacement is established through the displacement response curve, and thermo-mechanical coupling characteristic parameters are obtained from the quantitative correspondence. The thermo-coupling characteristic parameters and weld quality scores were time-series registered to construct a three-dimensional quality response surface. The coordinate axes of the quality response surface corresponded to the temperature field offset, weld head displacement, and weld quality score, respectively. Cubic spline interpolation was used to fit the quality response surface, and the quality change trend caused by the temperature field offset was extracted from the fitted surface to establish a mapping function between welding parameters and quality characteristics. A fourth-order state transition matrix was constructed using the mapping function, with state variables including temperature field offset, displacement change, quality score, and parameter change rate. A recursive least squares method with a forgetting factor of 0.95 was used to identify the state transition matrix online, with an identification period of 0.1 seconds. The coupling evolution law of temperature field offset, weld head displacement, and weld quality was obtained from the identification results. The multi-parameter coupling characteristics of the welding process are reflected in the composition of a 12-dimensional feature vector. The temperature field offset includes four components: center temperature, temperature gradient, temperature change rate, and temperature distribution uniformity. The weld head displacement includes four components: three-dimensional coordinate displacement, displacement velocity, and attitude angle. The weld quality score includes four components: penetration depth, diameter, strength, and appearance. The feature vectors are recorded at 0.1-second intervals to form a temporal data stream. A bidirectional long short-term memory network extracts temporal features through forward and backward hidden layers, with 128 neurons in each layer providing sufficient feature representation capabilities. The coupled response of temperature field offset and displacement change exhibits significant nonlinear characteristics. Principal component analysis shows that the cumulative contribution rate of temperature gradient, displacement increment, and acceleration components reaches 96.8%. When the temperature field offset undergoes a 20-degree Celsius step change, the weld head displacement produces a 45-micrometer response displacement within 0.2 seconds. The displacement curve exhibits obvious overshoot characteristics, with a maximum overshoot of 12 micrometers and a settling time of 0.8 seconds. The displacement response of the welding head in the horizontal plane is direction-dependent, with a displacement sensitivity of 2.5 micrometers per degree Celsius in the X direction and 1.8 micrometers per degree Celsius in the Y direction.The quality response surface reflects the combined impact of temperature field shift and displacement change on weld quality. The three-dimensional surface shows a significant decrease in quality under conditions of a 20°C temperature shift and a 45-micrometer displacement. The weld penetration depth decreases from the standard value of 0.4 mm to 0.32 mm, the diameter decreases from 1.0 mm to 0.85 mm, and the quality score decreases from 95 to 78. The cubic spline interpolation-fitted quality response surface indicates that the influence weight of temperature field shift on quality is 0.65, and the influence weight of displacement change is 0.35. The fourth-order state transition matrix describes the dynamic evolution of temperature field shift, displacement change, and quality score, with matrix elements updated in real time using a recursive least squares method. Observational data show a phase delay of approximately 0.15 seconds between temperature field shift and displacement change, and a response delay of 0.25 seconds for the quality score. The eigenvalue distribution of the state transition matrix reflects the system's stability, with the dominant eigenvalue having a magnitude of 0.92, indicating good convergence characteristics. A forgetting factor of 0.95 ensures the parameter identification's ability to quickly track changes in operating conditions, with the root mean square error of the identification results controlled within 3%. The coupling evolution law reveals the transmission mechanism by which temperature field shifts cause weld head displacement through thermo-coupling effects, leading to weld quality degradation. During the steady-state phase of the welding process, temperature field fluctuations are within ±5 degrees Celsius, corresponding to displacement fluctuations of less than 10 micrometers, and the quality score remains above 90. When a sustained temperature field shift occurs, the system enters an unstable state, with the displacement exhibiting an exponential growth trend, while the quality score shows an approximately linear decline.

[0041] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for array detection and positioning of multi-welding-head battery welding, characterized in that, The method includes: acquiring the arrangement spacing and material properties of the welding head array; establishing a mathematical model of heat conduction of the welding head array based on the arrangement spacing and material properties; simulating and calculating the temperature field distribution of the welding heads under different arrangement spacings using finite element analysis to determine the influence of the welding head spacing on heat conduction; using infrared thermal imaging technology to monitor the temperature field distribution of the welding heads in real time; determining whether the temperature field of the welding heads has shifted based on a preset temperature threshold; if the temperature of a certain welding head exceeds the threshold range, it is determined that the temperature field of that welding head has shifted; after determining that the temperature field has shifted, calculating the temperature field shift amount based on the temperature difference before and after the temperature field shift; acquiring the thermal expansion coefficient of the welding head base material; calculating the thermal expansion deformation amount of the welding head base under the temperature field shift condition based on the temperature field shift amount; and detecting the displacement deformation of the welding head base in real time using machine vision measurement technology; acquiring the actual position of the welding head; and performing correlation analysis based on the thermal expansion deformation amount and the temperature field shift amount; and verifying the results through theoretical derivation and experimental verification. A ternary mathematical model is constructed to calculate the welding position offset, thermal expansion deformation, and temperature field offset of each welding head in real time due to thermal effects. Based on the magnitude of the welding position offset, the motion control parameters of the digital welding system are adaptively adjusted, and a closed-loop feedback control algorithm is used to compensate for welding head offset errors in real time. Online weld quality monitoring technology is employed to extract weld penetration depth and diameter, and combined with a pre-established weld quality evaluation model, to determine in real time whether the weld quality is acceptable. If multiple consecutive welds fail to meet quality standards, it is determined that the welding head temperature field offset affects the weld quality. Machine learning algorithms are used to fuse the welding head temperature field offset, welding position offset, and weld quality, constructing a multi-parameter coupled model of the welding process. Key parameters reflecting the temperature field offset pattern, welding position offset characteristics, and weld defect patterns are extracted. Through data correlation analysis and mechanism deduction, the influence of the welding head thermal field coupling effect on welding quality is obtained, and a quantitative relationship between temperature field offset, welding position offset, and weld quality is established.

2. The method according to claim 1, characterized in that, The process involves obtaining the spacing and material properties of the welding head array, establishing a mathematical model for heat conduction of the welding head array based on these properties, and simulating the temperature field distribution of the welding heads under different spacings using finite element analysis to determine the influence of welding head spacing on heat conduction. This includes: obtaining the horizontal and vertical spacing values ​​of the welding head array according to its arrangement; acquiring initial values ​​of the surface temperature field of the welding head array using an infrared thermal imager to obtain a temperature distribution function; extracting the thermal conductivity of the welding head material from the initial temperature field values; establishing a heat diffusion function using the thermal conductivity and thermal impedance, where the heat diffusion function includes the temperature diffusion coefficient; dividing the heat conduction calculation region of the welding head array into square grid cells; solving the differential equation of heat conduction within the grid cells using a fourth-order Runge-Kutta algorithm to obtain a temperature-time variation function; fitting the temperature distribution function to the temperature-time variation function; and using the linear least squares method to calculate the relationship between the welding head spacing and heat diffusion to obtain the spatial distribution law of the temperature field.

3. The method according to claim 1, characterized in that, The method employs infrared thermal imaging technology to monitor the temperature field distribution of the welding head in real time, and determines whether the temperature field of the welding head has shifted in combination with a preset temperature threshold. If the temperature of a certain welding head exceeds the threshold range, it is determined that the temperature field of the welding head has shifted. This includes: acquiring a temperature field image of the surface of the welding head array using an infrared thermal imager, and processing the temperature field image using a Gaussian filter to obtain a filtered temperature field image; obtaining the temperature color level value of the pixels in the welding head area from the filtered temperature field image, and calculating the temperature gradient value based on the temperature color level value to obtain temperature time series data; calculating the deviation amplitude between the temperature value in the welding head area and the reference temperature based on the temperature time series data, and recording the welding head position coordinates and temperature deviation value when the temperature deviation amplitude exceeds a preset range; performing an accumulation calculation on the temperature deviation value of the welding head position, and if the accumulated temperature deviation value exceeds a preset temperature fluctuation threshold, it is determined that the temperature field of the welding head has shifted abnormally, and obtaining the coordinates of the welding head center point and the accumulated temperature deviation value as a temperature anomaly record.

4. The method according to claim 1, characterized in that, After determining that a temperature field shift has occurred, the temperature field shift is calculated by the temperature difference before and after the shift, the thermal expansion coefficient of the welding head base material is obtained, and the thermal expansion deformation of the welding head base under the temperature field shift is calculated by combining the temperature field shift. The displacement deformation of the welding head base is detected in real time by machine vision measurement technology, including: collecting the welding head temperature field value by an infrared thermal imager, calculating the temperature difference by the temperature value of the welding head center area within the time window before and after the temperature field anomaly determination time, and obtaining the temperature field shift of the welding head base; using the linear thermal expansion coefficient value in the copper material property database of the welding head base and the temperature field shift, the theoretical thermal expansion displacement is calculated based on the three-dimensional dimension parameters of the base; corner feature marker coordinates are extracted from the base surface image collected by an industrial camera, and the feature marker coordinates are tracked by a sub-pixel edge detection algorithm to obtain the actual deformation of the base surface; the theoretical thermal expansion displacement and the actual deformation are filtered by a Kalman filter algorithm to obtain the three-dimensional spatial deformation vector of the base. If the deformation vector exceeds the positive and negative threshold range, the thermal strain of the base is determined to be abnormal.

5. The method according to claim 1, characterized in that, The process involves acquiring the actual position of the welding head, combining it with thermal expansion deformation and temperature field offset, and performing correlation analysis. Through theoretical derivation and experimental verification, a ternary mathematical model is constructed to determine the welding position offset caused by thermal effects for each welding head. This model includes: extracting the center coordinates of a calibration pattern from an image of the welding head surface acquired by an industrial camera; using these center coordinates to calibrate camera parameters to obtain a welding head position deviation vector; acquiring temperature field data of the welding head area using an infrared thermal imager; calculating a temperature deviation value based on the temperature field data; using the temperature deviation value combined with the coefficient of thermal expansion to obtain a thermal expansion displacement vector; constructing a Bayesian network model based on the welding head position deviation vector and the thermal expansion displacement vector; extracting a position compensation vector from the Bayesian network model; performing a recursive filtering operation on the position compensation vector; obtaining a temperature compensation value through the recursive filtering operation; and constructing a three-dimensional spatial compensation function based on the temperature compensation value.

6. The method according to claim 1, characterized in that, The process of adaptively adjusting the motion control parameters of the digital welding system based on the magnitude of the welding position offset, and compensating for the welding head offset error in real time through a closed-loop feedback control algorithm, includes: comparing the welding head position offset with a preset compensation threshold numerically, and calculating the compensation gain coefficient using a particle swarm optimization algorithm; constructing a three-dimensional spatial compensation vector by multiplying the compensation gain coefficient with the position offset, and generating a compensation response curve containing position, velocity, and acceleration components from the compensation vector; extracting the compensation amount from the compensation response curve to construct a third-order state-space equation, and iteratively calculating the state-space equation using a Kalman state observer to obtain the compensation control amount; updating the position loop gain parameters, velocity loop gain parameters, and current loop gain parameters based on the compensation control amount, obtaining the actual compensation displacement by adjusting the position loop gain parameters, velocity loop gain parameters, and current gain parameters, obtaining the compensation tracking error by numerically comparing the actual compensation displacement with the theoretical compensation amount, and using a proportional-integral-derivative controller to perform closed-loop adjustment of the compensation tracking error.

7. The method according to claim 1, characterized in that, The aforementioned online weld joint quality monitoring technology extracts the weld joint penetration depth and diameter, and, combined with a pre-established weld joint quality evaluation model, determines in real time whether the weld joint quality is up to standard. If multiple consecutive weld joints fail to meet quality standards, it is determined that the temperature field shift of the welding head affects the weld joint quality. This includes: receiving weld joint surface images captured by an industrial camera; obtaining a grayscale image through grayscale quantitative processing; extracting the weld joint edge contour using the Sobel operator based on the grayscale image; obtaining sub-pixel coordinates of the weld joint contour using Gaussian curve fitting; and obtaining a feature vector containing weld joint diameter, penetration depth, grayscale mean, and contour roundness. Based on the feature vector, [the following steps are taken]... A deep convolutional neural network (DCNN) is used to identify and calculate the solder joint quality score. The DCNN consists of three convolutional layers and two fully connected layers to obtain a standardized quality score. For the standardized quality score, a sliding time window is used to scan the score sequence. If the solder joint quality score is lower than a preset threshold within the time window, it is determined that the temperature field of the solder head corresponding to the time window has an abnormal shift. Based on the feature vector and quality score data within the temperature field shift time window, a recursive least squares algorithm is used to establish a mapping function from the feature vector to the quality score. The convergence trend of the mapping function is used to determine the temperature field shift pattern.

8. The method according to claim 1, characterized in that, The method employs machine learning algorithms to fuse and calculate the welding head temperature field offset, welding position offset, and weld quality, constructing a multi-parameter coupled model of the welding process. Key parameters reflecting the temperature field offset pattern, welding position offset characteristics, and weld defect patterns are extracted. Through data correlation analysis and mechanism deduction, the influence of the welding head thermal field coupling effect on welding quality is obtained, establishing a quantitative relationship between temperature field offset, welding position offset, and weld quality. This includes: constructing a feature vector sequence based on the temperature field offset, welding head displacement, and weld quality score; obtaining a coupled response function from the feature vector sequence through a bidirectional long short-term memory network; and performing principal component analysis on the output data of the coupled response function. Dimensionality reduction is performed by obtaining the temperature gradient, displacement increment, and acceleration components through a feature contribution rate threshold. These components constitute a displacement response curve under thermo-mechanical coupling. The displacement response curve is used to perform time-series registration of the temperature field offset and the weld head displacement. A three-dimensional quality response surface is constructed using the time-series registration result and the weld quality score. The mapping function between welding parameters and quality characteristics is obtained from the three-dimensional quality response surface through cubic spline interpolation. A state transition matrix is ​​constructed based on the mapping function. The state transition matrix is ​​identified online using the recursive least squares method. The coupling evolution law of temperature field offset, weld head displacement, and weld quality is obtained through the online identification result.

Images