Heterogeneous metal laser welding device based on swing beam and molten pool state online monitoring

Abstract

The application discloses a kind of based on swing light beam and molten pool state monitoring dissimilar metal laser welding device, to improve welding quality and joint stability.The device includes laser welding head with beam swing function, can be according to preset trajectory type, frequency and amplitude in welding area Nonlinear energy scanning is implemented;Acquisition module can high frame rate synchronous acquisition molten pool visual image, excitation spectrum and infrared thermal imaging signal, and through multimodal fusion extraction interface diffusion and metal mixing characteristics;Calculation module is according to the physical characteristic information and swing parameter and carries out dynamic characteristic modeling, quantifies the multidimensional state index of welding quality;Control module then is according to state index and implements laser power, welding speed and swing parameter Joint adjustment, constructs closed-loop feedback control, to realize to dissimilar metal welding process Real-time stable regulation and control and defect inhibition.

Description

Technical Field

[0001] This invention relates to the field of laser welding technology, and in particular to a dissimilar metal laser welding device based on oscillating beam and molten pool condition monitoring. Background Technology

[0002] In existing technologies, dissimilar metal laser welding has been widely used in automotive, aerospace, and electronics manufacturing due to its high energy density, good directionality, and automation adaptability. Conventional laser welding equipment typically applies heat using a fixed laser beam or a simple linear scanning method, controlling the welding process through static process parameter settings. Some systems integrate visual sensors or thermal imaging devices to achieve weld trajectory recognition and basic welding status monitoring.

[0003] Existing technologies still have many shortcomings in handling dissimilar metal welding. For example, large differences in material thermophysical properties lead to unstable molten pools, easy formation of brittle phases at the interface, insufficient mixing between metals, and a lack of in-depth modeling and real-time feedback adjustment capabilities for multi-source state data during welding, making it difficult to achieve highly reliable and consistent welding quality control. In addition, traditional condition monitoring methods are mostly single-channel sensing, lacking systematic modeling and response mechanisms for metal fusion behavior and interface diffusion characteristics, resulting in lagging control measures and limited adjustment accuracy.

[0004] In view of this, there is an urgent need to propose a technology path that integrates dynamic sensing and real-time control for dissimilar metal welding processes, so as to improve the stability of welding quality and the reliability of interface structure. Summary of the Invention

[0005] This application provides a dissimilar metal laser welding device based on oscillating beam and molten pool state monitoring to improve the stability of welding quality and the reliability of interface structure.

[0006] This application provides a dissimilar metal laser welding device based on oscillating beam and molten pool condition monitoring, comprising:

[0007] A laser welding head is used to output a laser beam and apply welding energy to dissimilar metal joints. The laser welding head has a beam oscillation function, which enables the laser beam to perform energy scanning along a nonlinear trajectory within the welding area based on preset oscillation trajectory parameters. The oscillation trajectory parameters include trajectory type, oscillation frequency, and oscillation amplitude.

[0008] The acquisition module is used to synchronously acquire multi-source data of the welding area at a high frame rate during laser welding, including visual images of the weld pool, plasma excitation spectra and infrared thermal imaging signals, and extract physical feature information characterizing the degree of metal mixing, interface diffusion behavior and dynamic fluctuations of the weld pool based on a multi-modal fusion method.

[0009] The calculation module is used to perform dynamic feature modeling based on the physical feature information and the swing trajectory parameters, and to calculate multi-dimensional state indicators characterizing welding quality using time series feature extraction and spatial energy distribution analysis methods, including molten pool stability index, instantaneous intermetallic fusion uniformity index and interface temperature difference fluctuation index.

[0010] The control module receives multi-dimensional status indicators and performs joint optimization adjustment of laser parameters and trajectory parameters based on the control weight matrix, including real-time adjustment of laser power, welding speed, oscillation frequency and trajectory shape.

[0011] This application has the following beneficial technical effects:

[0012] (1) By expanding the laser's range of action through nonlinear oscillation scanning of the laser beam, convection and diffusion within the molten pool are enhanced, effectively increasing the mixing degree between metals and suppressing the formation of brittle interfacial compounds. (2) By fusing multi-source data such as visual images, excitation spectra, and thermal imaging, the system can comprehensively extract key physical characteristics such as molten pool stability, temperature distribution, and interfacial diffusion, providing highly reliable state support for welding process control. (3) By calculating multi-dimensional welding quality indicators and implementing dynamic adjustments, the system can achieve joint optimization of laser power, welding speed, and oscillation trajectory, significantly improving the stability and consistency of the welding process. (4) Through dynamic modeling and parameter adjustment mechanisms, the device has good adaptive capabilities, can adapt to different material combinations and irregular joint paths, and improves the flexibility and reliability of the welding process. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of a dissimilar metal laser welding device based on oscillating beam and molten pool state monitoring provided in the first embodiment of this application. Detailed Implementation

[0014] Many specific details are set forth in the following description to provide a full understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this application; therefore, this application is not limited to the specific embodiments disclosed below.

[0015] The first embodiment of this application provides a dissimilar metal laser welding device based on oscillating beam and molten pool state monitoring. Please refer to... Figure 1 This figure is a schematic diagram of the first embodiment of this application. The following is in conjunction with... Figure 1 The first embodiment of this application provides a detailed description of a dissimilar metal laser welding device based on oscillating beam and molten pool state monitoring.

[0016] The dissimilar metal laser welding device based on oscillating beam and molten pool state monitoring includes a laser welding head 101, an acquisition module 102, a calculation module 103, and a control module 104.

[0017] The laser welding head 101 is used to output a laser beam and apply welding energy to the dissimilar metal joint. The laser welding head has a beam oscillation function, which makes the laser beam perform energy scanning along a nonlinear trajectory in the welding area based on preset oscillation trajectory parameters. The oscillation trajectory parameters include trajectory type, oscillation frequency and oscillation amplitude.

[0018] The laser welding head 101 is the core execution unit of this invention, used to accurately apply laser beam energy to the dissimilar metal bonding area. Its structure comprises a laser emitting assembly, a collimating and focusing assembly, a oscillation drive assembly, and a trajectory control interface, forming an integrated programmable execution device. The laser welding head 101 can be driven by an external control signal to oscillate its internal laser beam, causing the output laser beam to no longer irradiate linearly within the welding area, but rather to dynamically scan along a preset two-dimensional nonlinear trajectory. This expands the effective area of ​​the laser heat source and improves the heat input distribution at the interface between dissimilar metals.

[0019] Specifically, the laser welding head 101 is equipped with a beam oscillation actuator composed of a high-speed rotating mirror or a MEMS micro-mirror. This mechanism can precisely oscillate according to the oscillation trajectory parameters issued by the control module 104, including three dimensions: trajectory type, oscillation frequency, and oscillation amplitude. The trajectory type can be any programmable path such as circular, elliptical, infinite, or spiral. The oscillation frequency controls the time period for the laser beam to complete a complete trajectory path, which is usually adjustable in the range of 100Hz to 2kHz. The oscillation amplitude determines the maximum lateral displacement of the laser beam from the central axis, with a typical value between 0.1mm and 2mm, set according to the metal thermal conductivity, joint width, and expected weld width.

[0020] The laser welding head 101 maintains the thermal stability of its optical components through a built-in temperature-controlled cooling circuit to prevent focus drift or uneven output energy under high-power laser action. Simultaneously, to ensure beam focusing quality, the front end of the laser welding head 101 is equipped with a set of composite lenses or high-reflectivity laser windows with adjustable focal lengths to match combinations of dissimilar metal materials with different thicknesses and thermal diffusion characteristics. During beam oscillation, this focusing system ensures that the laser energy remains concentrated on the workpiece surface or slightly sinks to the center of the molten metal pool throughout the entire oscillation trajectory, resulting in consistent melt depth and a stable molten pool morphology.

[0021] The laser welding head 101 also includes a trajectory control interface for receiving trajectory parameter update commands from the control module 104 in real time and converting them into specific execution actions. This interface is connected to the main control system based on industrial standard communication protocols (such as EtherCAT or RS-485) and has interrupt response capability. It can quickly switch the beam motion mode according to changes in the molten pool state, thereby avoiding the expansion of weld defects caused by environmental disturbances or material inhomogeneity.

[0022] In use, the laser welding head 101 first receives the initially set oscillation trajectory parameters and focuses the laser output on the dissimilar metal bonding area. Through continuous, high-frequency beam oscillation, it achieves uniform energy distribution and multi-point excitation during the welding process, effectively promoting convection mixing and interfacial diffusion between metals. Simultaneously, through the programmable adjustment capability of the trajectory parameters, the laser welding head 101 can adapt to metal materials of different thicknesses and thermal conductivity, achieving dynamic balance compensation for thermal asymmetry during dissimilar metal welding, thereby avoiding the formation of hot cracks, porosity, and brittle phases between metals.

[0023] In summary, the laser welding head 101 not only performs the function of energy transfer, but also enhances the control of inter-metal reaction through dynamic oscillation, making it a fundamental physical execution component for achieving high-quality dissimilar metal connections.

[0024] The acquisition module 102 is used to synchronously acquire multi-source data of the welding area at a high frame rate during the laser welding process, including visual images of the weld pool, plasma excitation spectra and infrared thermal imaging signals, and extract physical feature information characterizing the degree of metal mixing, interface diffusion behavior and dynamic fluctuations of the weld pool based on a multi-modal fusion method.

[0025] The acquisition module 102 is a key component in this invention for realizing multi-dimensional perception and real-time data support of the molten pool state. Its core design is to enable high temporal and spatial resolution multi-source information acquisition of the molten pool and its surrounding area during the laser welding of dissimilar metals, thereby providing a physical basis for subsequent welding quality modeling and control adjustment.

[0026] In terms of hardware configuration, the acquisition module 102 includes at least three sub-devices: a high-speed industrial vision sensor, a spectral acquisition device, and an infrared thermal imager. These three correspond to data acquisition capabilities in the visible light domain, emission spectral domain, and infrared band, respectively. The high-speed industrial vision sensor preferably employs a global shutter type CMOS image sensor with a frame rate of no less than 100 frames per second and a resolution of no less than 1280×1024 pixels. It can stably capture visual images of the molten pool contour, spatter behavior, and weld formation during laser welding. Combined with light source assistance, it can achieve clear imaging even under strong background light interference in the welding area. The spectral acquisition device uses a highly sensitive linear spectrometer with a working wavelength range covering 350nm to 900nm. It is used to acquire the plasma spectral signals generated by metal excitation during welding. By analyzing the intensity and wavelength distribution of characteristic spectral lines, key information such as element type, concentration changes, and excitation temperature can be extracted, thereby indirectly reflecting metal mixing and reactivity. The infrared thermal imager uses a non-contact high-precision thermal infrared camera with a working wavelength of 8–14 μm, a thermal sensitivity better than 0.05°C, and a sampling frequency of no less than 50 frames / second. It can capture the temperature field distribution of the molten pool surface and the adjacent heat-affected zone in real time, and identify potential local overheating, cold crack sensitive areas, and uneven energy input problems.

[0027] The three types of sensors mentioned above achieve unified timestamp alignment and synchronous sampling through industrial-grade data acquisition cards or edge computing units. All data is preprocessed by the fusion processing unit within the acquisition module after acquisition, including image denoising, spectral baseline correction, and infrared thermal image grayscale temperature calibration. Based on a multimodal fusion method, spatial-temporal alignment and feature coupling are performed, ultimately forming a unified data packet format for output to the computing module 103. Regarding the fusion method, a feature-level fusion strategy is adopted, co-registering visual edge information with thermal imaging gradient changes, and superimposing spectral peak intensity as a weighting factor to improve the boundary recognition accuracy and dynamic response capability of the metal mixing region to state changes. Through this multimodal data structure, the following physical feature information can be extracted simultaneously: molten pool geometry, center temperature, interface temperature difference distribution, spectral peak shift trend, element excitation intensity, welding spatter frequency, and edge perturbation amplitude.

[0028] All sensors in the acquisition module 102 are mounted on a structural support coaxial with or on the same platform as the laser welding head 101. Their optical axis adjustment and focal length settings are calibrated to ensure that the acquisition area accurately covers the laser scanning area, avoiding blind spots or measurement errors. In addition, the acquisition module also has a built-in temperature control and filter protection system to ensure long-term stable operation under strong laser and high-temperature environments.

[0029] For example, the acquisition module 102 synchronously monitors the laser welding process of aluminum-steel dissimilar metals, acquiring three types of data: visual images of the welding area, plasma spectral signals, and infrared thermal imaging images. To perform multimodal fusion, these three types of data are first synchronized in time, ensuring that each frame of image, each set of spectral signals, and each frame of thermal image corresponds to the welding state at the same moment.

[0030] At one welding point, the visual image showed that the edge of the molten pool was slightly irregular, indicating that there may be local instability. The corresponding thermal image showed that the temperature at the edge of the area fluctuated greatly, while the temperature at the center was lower. At the same time, the emission intensity of aluminum in the spectral signal was abnormally high, while the emission signal of iron was weakened, indicating that the metal mixing ratio had changed and aluminum segregation may have occurred.

[0031] In this context, the three types of information are analyzed using a "feature overlay and fusion" method. Specifically, the boundary of the molten pool identified in the visual image is overlaid and compared with the temperature distribution in the thermal image to determine whether there is a corresponding temperature unevenness. Then, combined with the changes in the proportion of metal elements in the spectral signal, it is confirmed whether uneven metal diffusion or insufficient fusion has occurred in the region.

[0032] This fusion method allows for the extraction of three distinct physical characteristics: first, the degree of metal mixing is reflected by the change in the ratio of spectral intensities of aluminum and iron; second, interfacial diffusion behavior is determined by the uniformity of temperature distribution combined with changes in spectral peaks; and third, the dynamic fluctuations of the molten pool are comprehensively evaluated by changes in boundary shape and temperature jumps in thermal imaging. These characteristics are transmitted as structured information to the calculation module 103 for subsequent welding quality status analysis and parameter adjustment.

[0033] In summary, the acquisition module 102 not only realizes high-frequency, low-latency, and physically coupled data acquisition of the state of the laser welding pool of dissimilar metals from multiple dimensions, but also establishes a standardized physical feature set that can be used for subsequent calculation and analysis through fusion processing, enabling the system to have the basic ability to support dynamic modeling and real-time feedback control.

[0034] Furthermore, the acquisition module is specifically used for:

[0035] In the initial stage of welding, a weld area stability prediction model is established based on the oscillation trajectory parameters of the laser welding head and the preset metal combination characteristics, which is used to predict the potential welding quality fluctuation risk level at each trajectory position point throughout the entire oscillation cycle.

[0036] Based on the output of the stability prediction model, the swing trajectory is divided into multiple dynamic acquisition areas. For areas with high predicted risk levels, improved acquisition parameters are automatically configured, including image resolution improvement, thermal image refresh frequency increase, or spectral integration time extension.

[0037] In the actual welding process, the acquisition module performs local enhanced sampling on high-risk areas according to the configuration of the dynamic acquisition area. The spatial resolution or temporal density of the acquired visual images, plasma excitation spectra and infrared thermal imaging signals is higher than that of the standard area.

[0038] The data obtained from augmented sampling is labeled as high-confidence feature segments, and its corresponding trajectory location label and risk prediction level are attached, which are used as the priority input for keyframes in subsequent modeling.

[0039] The physical feature information extracted from the standard area and the high-risk area is uniformly structured and output as physical feature information to characterize the degree of metal mixing, interface diffusion behavior and dynamic fluctuation of the molten pool, and output to the calculation module for dynamic feature modeling based on the physical feature information.

[0040] In this implementation, the functional design of the acquisition module is not limited to conventional data acquisition, but integrates multi-level technologies such as predictive modeling, dynamic regional configuration, enhanced sampling strategies and structured data output, in order to achieve real-time identification and refined monitoring of potential fluctuation risks in welding quality during dissimilar metal laser welding, and ensure that the basic data for subsequent modeling and control have high confidence and high responsiveness.

[0041] In the initial stage of welding, the system first calls upon the built-in metal material database based on the task parameters to extract the combined characteristics of dissimilar metals involved in the current welding process. This set of characteristics includes, but is not limited to, the thermal conductivity, specific heat capacity, melting point, surface tension, laser reflectivity, and intermetallic reactivity of the two metals. These characteristics determine the impact of heat input on the interface region and the stability of the mixing behavior during welding. Simultaneously, the current oscillation trajectory parameters of the laser welding head are also read, including the oscillation trajectory type (e.g., infinity symbol, sawtooth, Lissajous curve, etc.), oscillation frequency (in Hertz), and oscillation amplitude (in millimeters). These parameters describe how the laser beam moves on the weld surface, thus affecting the local energy distribution.

[0042] Based on the above inputs, the system constructs a model for predicting the stability of the welded area. This model can employ structures such as a multilayer perceptron (MLP) or a lightweight random forest. During the training phase, it inputs quality fluctuation data from historical welding experiments under different oscillation trajectories and metal combinations, and outputs the risk level of welding quality fluctuation at each trajectory location. The risk level is a discrete label, typically categorized as low risk (level 0), low-to-medium risk (level 1), medium-to-high risk (level 2), and high risk (level 3). In actual welding tasks, this model is used to predict the risk level at each trajectory location throughout the entire oscillation cycle, forming a risk distribution map.

[0043] Based on the aforementioned risk distribution, the system then divides the oscillation trajectory into multiple dynamic acquisition areas, each corresponding to one or more trajectory location points. For trajectory areas predicted to be high-risk, the system automatically configures improved acquisition parameters to enhance the data acquisition capability of that area. The configuration of acquisition parameters includes three optimization strategies: First, increasing the resolution of the visual image, for example from 640×480 to 1280×960, to more precisely capture changes in the morphology of the molten pool boundary; second, increasing the infrared thermal imaging frame rate, for example from 30fps to 90fps, thereby enabling more frequent acquisition of temperature change processes and identification of short-term thermal flow disturbances; third, extending the integration time of the spectral sensor, allowing for more complete reception of signals from weak excitation light and improving the signal-to-noise ratio of alloy evaporation characteristics or intermetallic reaction spectra.

[0044] During the actual welding process, the acquisition module performs enhanced sampling in high-risk areas according to the aforementioned dynamic acquisition area division. Enhanced sampling is not limited to increasing the sampling frequency or resolution, but also includes changes in the sampling method, such as enabling multi-angle synchronous imaging in high-risk areas or increasing the spatial sampling point density of thermal imaging. In this way, the system can capture physical behavior changes within high-risk trajectory segments with higher temporal and spatial resolution, thereby providing sufficient data support for subsequent feature extraction.

[0045] All data obtained from augmented sampling will be labeled during the acquisition phase. First, a "high confidence feature segment" label will be added to indicate that the data has high reliability. At the same time, a "trajectory location label" will be added, which will be indexed by the normalized location identifier within the swing period (such as 0.15T, 0.60T, etc.). In addition, each data segment will also be labeled with a "risk prediction level" label, indicating the priority setting of the data in the modeling process.

[0046] These data are structured and organized within the acquisition module. The extracted physical features mainly include the frequency and amplitude characteristics of molten pool boundary vibrations in visual images, the relative intensity changes of key elements (such as Fe, Ni, and Cu) in spectral signals, and the temperature gradient and local temperature difference change patterns in infrared thermograms. All these data from high-risk and standard areas are uniformly formatted into three types of physical feature descriptions, corresponding to the three dimensions of metal mixing degree, interface diffusion behavior, and molten pool dynamic fluctuations, respectively.

[0047] For example, the degree of metal mixing can be reflected by the ratio of the intensities of different metal spectral lines in the spectrum and their distribution consistency along the welding path; the interface diffusion behavior can be measured by the width and gradient of the interface temperature transition region in a high-resolution infrared image; and the dynamic fluctuations of the molten pool can be represented by the frequency response changes of the molten pool boundary contour in an image sequence.

[0048] Finally, the acquisition module outputs all the aforementioned structured physical feature information to the calculation module, serving as the input data source for subsequent dynamic modeling and multi-dimensional state index calculation. The entire acquisition module's workflow is centered on the physical logic of the welding process. Through risk-aware data augmentation and tagging strategies, it significantly improves the system's monitoring accuracy and response capabilities in complex welding scenarios.

[0049] Furthermore, the acquisition module is also used for:

[0050] When the risk prediction level at a certain swing trajectory location point reaches a set high-risk threshold, the visual image enhancement acquisition process is automatically started. The visual image enhancement acquisition process includes: performing multi-scale wavelet reconstruction processing on the visual image acquired at the high-risk trajectory location point, and combining edge-guided filtering and local contrast enhancement algorithms to enhance the texture saliency of the molten pool edge, interface disturbance and metal inter-metal detail structure contained in the visual image.

[0051] Based on the grayscale histogram distribution information of infrared thermal imaging signals collected at high-risk trajectory locations, the hot spot concentration index of the thermal imaging channel is dynamically calculated. When the hot spot concentration exceeds the adaptive saturation threshold, the slope of the response curve and the upper limit parameter of the saturation of the infrared thermal imaging signal are adjusted in real time, so that the infrared thermal imaging signal can improve the overall contrast of the low-temperature area while maintaining the details of the high-temperature area.

[0052] The enhanced visual image and the infrared thermal imaging signal with the adjusted response curve are aligned at the pixel level and geometrically registered between channels to form composite enhanced frame data with spatial consistency and modal complementarity. The composite enhanced frame data synchronously inherits its corresponding trajectory position label and risk prediction level information.

[0053] The composite enhanced frame data is structured and marked as high-confidence feature segments, and texture complexity index and hotspot adjustment parameters are added as auxiliary modeling factors during the structuring process to provide a basis for region refinement weights when outputting to the calculation module.

[0054] The high-confidence feature fragments are uniformly output as part of the physical feature information used to characterize the degree of metal mixing, interface diffusion behavior and dynamic fluctuations of the molten pool, and are fused with the standard area acquisition results and input into the calculation module to enhance the accuracy and response sensitivity of the modeling results at the high-risk trajectory location points.

[0055] In the dissimilar metal laser welding device based on oscillating beam and molten pool state monitoring proposed in this invention, when the acquisition module detects that the risk prediction level at a certain oscillation trajectory position point reaches a set high-risk threshold, it needs to perform a series of image enhancement processing operations to obtain higher-quality physical feature information. This provides the calculation module with input data that has stronger discriminative capabilities, thereby improving the perception and response accuracy of welding quality fluctuations. To achieve this goal, the design of the acquisition module integrates multi-scale signal processing, image enhancement, adaptive adjustment of infrared response curves, and multi-modal data registration techniques, forming a complete, feasible, and technically sound visual image enhancement acquisition process.

[0056] Specifically, when the risk prediction model determines, based on previous modeling results or real-time data analysis, that a certain oscillation trajectory location point has a significant risk of unstable welding quality, i.e., the prediction level reaches the set high-risk threshold (e.g., level 3, the full score), the system triggers the visual image enhancement acquisition process. The first step of this process is to perform multi-scale wavelet reconstruction processing on the visual image acquired at the trajectory location point. Wavelet transform is a tool that can simultaneously analyze the local frequency and temporal information of a signal at different scales, and is suitable for multi-resolution decomposition of welding images containing complex details such as boundary changes and molten pool disturbances. By performing two-dimensional discrete wavelet decomposition on the original image, the system can extract low-frequency background information and high-frequency edge information, and amplify the influence weight of high-frequency components in the reconstruction stage, making the boundary contours and texture changes clearer. In this device, the db4 wavelet is preferably used as the mother wavelet function, and the number of decomposition layers is set to 3 to balance edge detail preservation and computational efficiency.

[0057] After wavelet reconstruction, to further highlight the detailed intermetallic structures and molten pool boundary disturbances in the image, the system feeds the image into an edge-guided filter for denoising and detail preservation. Unlike traditional mean or Gaussian filtering, edge-guided filtering does not blur image edges while smoothing noise. Instead, it combines image gradient direction information and performs limited smoothing only in the vertical direction of the edges, thus preserving edge sharpness. This process significantly improves the local clarity of areas such as the edges of the laser molten pool and metal overlap lines. Furthermore, to enhance overall contrast, especially when the image grayscale distribution is concentrated in the low to mid-range, the system introduces a local contrast enhancement algorithm (such as CLAHE, contrast-limited adaptive histogram equalization), which dynamically stretches the grayscale range according to local image windows, restoring details in dark areas and ensuring the accuracy of subsequent texture complexity calculations.

[0058] Simultaneously, the acquisition module also processes the infrared thermal imaging signal at the high-risk trajectory location. Unlike visible light images, infrared images use temperature distribution as the signal source, and their key function is to reflect the spatial uniformity and local concentration of heat input in the welding area. To identify problems such as excessive heat input concentration and uneven energy distribution, the system first extracts the grayscale histogram distribution of the current thermal image frame and calculates the hot spot concentration index based on this. This index can be defined as the product of the proportion of pixels above the 80% grayscale threshold and its aggregation density function, used to measure whether the distribution of high-temperature areas in the image is concentrated in a small area. If the hot spot concentration exceeds the set adaptive saturation threshold (e.g., set to 0.35), it indicates that the welding heat is excessively concentrated, which may cause abnormal local penetration or metal burn-through.

[0059] When hotspot concentration is too high, the system performs real-time adjustment processing on the response curve of the infrared image to preserve detail information in high-temperature areas while simultaneously enhancing the visibility of temperature distribution in low-temperature areas. The adjustment process includes two parameters: the slope of the response curve and the upper limit of saturation. The slope controls the response sensitivity in the low-to-medium grayscale range; an increased slope means that the temperature difference response in low-temperature areas is amplified. The upper limit of saturation controls the upper limit of image brightness, preventing grayscale compression in high-temperature areas that could lead to information loss. This dynamic adjustment mechanism is based on a strategy combining Gamma correction and linear interval mapping, and adaptively sets local adjustment intervals based on the hotspot location distribution, achieving an optimal balance between visual appeal and quantitative analysis accuracy for the entire infrared image.

[0060] After image enhancement and infrared response adjustment, the system enters the multimodal data registration stage. Since the visible light and infrared images come from different types of sensors and their viewing angles, pixel densities, and focal lengths are not entirely consistent, direct channel overlay would lead to image mismatch and structural misalignment. Therefore, the system employs the SURF (Accelerated Robust Feature Extraction) keypoint extraction method to match feature points between the enhanced and infrared images, and uses the RANSAC algorithm to remove out-of-match points. Geometric registration is then completed through affine transformation. Subsequently, the two modal images are aligned at the pixel level, unifying their coordinate systems and image sizes to generate composite enhanced frame data. This data possesses spatial consistency and modal complementarity, containing not only texture, edge, and structural information from the visible light image but also thermal features such as temperature gradients and hotspot concentration from the infrared image.

[0061] While generating composite enhanced frame data, the system inherits the original trajectory location labels and risk prediction level information from the original data and uniformly marks them as high-confidence feature segments. "High confidence" means that, on the one hand, the data source comes from trajectory location points with high risk levels and a high probability of welding instability; on the other hand, after image enhancement and multimodal fusion processing, the information expressiveness is significantly better than ordinary frame data. During the marking process, the system also adds calculated texture complexity index and hotspot adjustment parameters as auxiliary factors in subsequent modeling. The texture complexity index can be calculated from the image gray-level co-occurrence matrix, reflecting the degree of change in the image's local entropy value; the hotspot adjustment parameters include the slope of the response curve, the Gamma value, and the saturation interval position, all used to characterize the enhancement processing features experienced by the thermal image in that frame.

[0062] In practice, obtaining the texture complexity index relies on the spatial structural feature analysis of image texture. Specifically, this can be achieved by constructing the gray-level co-occurrence matrix of the visual image and extracting image texture statistical features at multiple directions and pixel spacing scales, including indicators such as energy, contrast, correlation, and entropy. In particular, the distribution characteristics of local entropy values ​​serve as a primary reference parameter, characterizing the richness of texture changes at the microscale. In this invention, for each frame of visual image acquired during the welding process, the system constructs a local gray-level co-occurrence matrix in its edge regions and metal fusion detail regions, and statistically analyzes the maximum, average, and gradient of local entropy to comprehensively determine the texture complexity level of the image. If the gray-level values ​​of the metal fusion band in the image exhibit drastic fluctuations within a small spatial range, it can be considered to possess high complexity characteristics, and the system will assign a higher texture complexity index accordingly.

[0063] The acquisition of hotspot adjustment parameters is based on the parameter recording and adjustment process in the infrared thermal imaging signal enhancement workflow. At high-risk trajectory locations, the system analyzes the original grayscale histogram of the infrared image to determine if there are issues such as hotspot concentration, brightness saturation, or insufficient contrast in low-temperature areas. When these issues exist, the system activates the thermal image enhancement mechanism and dynamically adjusts the relevant parameters of the infrared response curve. The slope of the response curve is determined by the linear gain adjustment of the image brightness mapping function, the Gamma value originates from the setting of nonlinear brightness compression or expansion processing, and the saturation interval position reflects whether the dynamic range of the high-temperature region of the infrared image is compressed or widened before and after enhancement. The system records the response slope coefficient, the degree of nonlinear Gamma adjustment, and the set high-brightness saturation start and end intervals for each frame of infrared image in the image enhancement module. These enhancement process parameters together constitute the hotspot adjustment parameters for that frame of image, which are used as a reference for subsequent thermal image channel quality evaluation and weight adjustment.

[0064] Furthermore, in obtaining the texture complexity index, the system first extracts a 400×400 pixel area as the center of the weld region marked as high-risk in the visual image as an analysis window. This region typically covers the entire weld pool and the metal diffusion area at its edges. To measure texture complexity, the system employs the Gray-Level Co-occurrence Matrix (GLCM) method to extract the local entropy of the image as a characterization index of texture saliency.

[0065] Specifically, the system calculates co-occurrence matrices for this region at pixel distances of 1 and 2 in the 0°, 45°, 90°, and 135° directions, resulting in a total of 8 sets of matrices. For each set of co-occurrence matrices, the system calculates its entropy value. The entropy value is an indicator of the uncertainty of the gray-level distribution of image pixels and can express the texture complexity of the image in that direction. For example, in this embodiment, the statistical results of the 8 sets of entropy values ​​are as follows:

[0066] Average entropy: 3.92;

[0067] Maximum entropy: 5.12;

[0068] Minimum entropy: 2.87;

[0069] Standard deviation: 0.65;

[0070] The system sets a normalization rule, mapping the maximum entropy value to 1 and the minimum entropy value to 0. The texture complexity index is calculated using the standard Z-score normalization method.

[0071] Substitute the average entropy value of the current region into the Z-score normalization expression:

[0072] Index = (Current average entropy - Minimum entropy) ÷ (Maximum entropy - Minimum entropy);

[0073] Substituting the values, we get:

[0074] The index = (3.92 - 2.87) ÷ (5.12 - 2.87) ≈ 1.05 ÷ 2.25 ≈ 0.4667;

[0075] To further account for the impact of local texture inhomogeneity, the system multiplies the standard deviation by an empirical weighting factor (e.g., 1.2) as an enhancement term added to the exponent. The final calculation result is as follows:

[0076] The final texture complexity index = 0.4667 + (0.65 × 1.2 ÷ 10) = 0.4667 + 0.078 ≈ 0.5447;

[0077] In the above calculation, "10" is the normalization coefficient for the enhancement term, which is set based on the empirical distribution range of the product term of the local texture entropy value standard deviation and the enhancement factor in the training samples. Specifically, during the offline training phase, the system performs batch statistical analysis on image data of multiple high-risk trajectory locations in typical welding scenarios. The measured entropy standard deviation is generally in the range of 0.3 to 0.9, and the enhancement factor is set to an upper limit of 1.5. Therefore, the theoretical maximum value of the product term does not exceed 1.35. To avoid the enhancement term deviating from the basic entropy value by an order of magnitude, which would affect the numerical stability of the final texture complexity index, the system adopts a strategy of scaling the product term to the range of 0–0.1. After multiple rounds of participation tuning experiments, a fixed value of 10 was finally determined as the normalization denominator. This ensures that the product result, after being divided by 10, is controlled within a suitable weighting range, which can enhance the response sensitivity to complex texture structures without introducing false hotspot judgments due to excessive perturbation.

[0078] Since the system uses a joint scoring model of entropy normalization index and edge strength, the edge point density is extracted using Canny edge detection for the same region (the number of edge points detected in this region is 1783, and the total number of pixels is 160,000). The edge density is 1783 / 160000 ≈ 0.0111. The system normalizes the edge density (maximum density 0.02, minimum density 0.002), and the edge complexity score is (0.0111 - 0.002) ÷ (0.02 - 0.002) ≈ 0.505.

[0079] The system weights and fuses the texture entropy complexity and edge complexity scores, with the following fusion ratio:

[0080] The final texture complexity index = 0.5447 × 0.6 + 0.505 × 0.4 = 0.3268 + 0.202 = 0.5288;

[0081] To further emphasize the saliency of textures in high-risk areas, the system enhances the scores based on an enhancement factor of 1.3 corresponding to risk level 3.

[0082] The final texture complexity index = 0.5288 × 1.3 ≈ 0.6874;

[0083] For model consistency and empirical evaluation purposes, the system normalizes this value using a Sigmoid function (center 0.5, slope 10) before the final output, resulting in the standard exponent:

[0084] Exponent = 1 / (1 + exp(-10 × (0.6874 - 0.5))) ≈ 0.86;

[0085] Therefore, the final recorded texture complexity index is 0.86. This value is based on a clear image region, a clear image statistical feature extraction method, a dual fusion scoring and standardization process, and is processed by dynamic enhancement logic corresponding to the risk level.

[0086] For hotspot adjustment parameters, the system uses single-frame images acquired at high-risk trajectory points using infrared thermal imaging. First, its grayscale histogram is extracted, and the proportion of pixels in high-temperature regions within the image is calculated. High-temperature regions are defined as pixels with grayscale values ​​between 220 and 255 (approximately 1400°C to the upper limit for the application temperature). In this embodiment, the total image size is 640×512, totaling 327,680 pixels, of which 82,000 pixels fall within the 220–255 grayscale range, accounting for approximately 25%. The system uses the proportion of high-temperature pixel concentration as the hotspot concentration index, normalized to 0.25.

[0087] The system sets an adaptive saturation threshold of 0.20 for hot spot concentration. When the actual hot spot concentration exceeds this value, the system will trigger the response curve adjustment logic. In this example, since 0.25 > 0.20, the system enters adjustment mode.

[0088] To prevent oversaturation in high-temperature regions from causing loss of temperature structure information, the system increased the linear slope of the response curve from 1.0 to 1.2 to moderately stretch grayscale contrast. Furthermore, to preserve details in low-temperature regions, the system adjusted the Gamma value to 0.8 (less than 1 indicates highlight compression). Finally, because the proportion of pixels corresponding to the highest grayscale value of 255 was too high (approximately 3%), the system adjusted the maximum grayscale threshold to 230 to ensure that the image's dynamic range was not compressed. In summary, the hotspot adjustment parameters for this frame are: response curve slope 1.2, Gamma value 0.8, and saturation range set to [230–255].

[0089] These parameters are all automatically calculated by the system based on image statistical features, without relying on subjective settings. They maintain consistent calculation logic and technical implementation path under different welding materials, welding paths, or risk distributions, thus providing a stable, reliable, and sensitive key reference for state modeling and control adjustment strategies in subsequent calculation modules.

[0090] Finally, the acquisition module outputs the high-confidence feature fragments to the calculation module, serving as a core component for constructing physical characteristic information characterizing the degree of metal mixing, interface diffusion behavior, and dynamic fluctuations of the molten pool. To ensure the comprehensiveness and accuracy of the modeling process, the acquisition module also inputs the conventional data results collected from standard areas (i.e., trajectory location points that have not reached the high-risk threshold) into the calculation module, where the two are fused. The fusion process not only considers the consistency of spatial location label distribution but also assigns modeling weights based on risk level and confidence coefficient. For example, when constructing a molten pool stability model, the boundary perturbation amplitude and edge continuity features in high-confidence frames will receive higher time-series regression weights; when constructing an interface temperature difference fluctuation model, the temperature range and gradient direction changes provided by high-confidence frames will be preferentially adopted as the dominant indicators of the global fluctuation trend.

[0091] In this invention, after receiving the structured high-confidence feature fragments output by the acquisition module, the calculation module not only acquires the basic physical feature information contained therein (such as the edge position of the molten pool, the metal mixing texture, the temperature gradient field, etc.), but also uses the attached texture complexity index and hot spot adjustment parameters as auxiliary modeling factors to embed into the state index modeling process, thereby enhancing the ability to identify local welding abnormal behavior and improving the modeling accuracy and response sensitivity to high-risk areas.

[0092] Specifically, when constructing the molten pool stability index, the calculation module first extracts the boundary dynamic change feature vector based on the boundary information, optical flow perturbation features, and multi-frame temporal changes in the high-confidence feature segments. Subsequently, the system introduces the texture complexity index as a dynamic change weight correction factor to adjust the weight of each frame's boundary perturbation contribution during the temporal stacking process. For example, if the texture complexity index of a high-confidence image is 0.85 (ranging from 0 to 1), it indicates that the image contains rich metal fusion details and interface perturbation features. The system will assign a higher temporal stacking weight to the boundary change contribution of this frame, reflecting its high state discrimination value corresponding to its high structural information density. Conversely, if the texture complexity index of an image frame is only 0.3, its role in state modeling will be suppressed to avoid low-quality images causing a shift in the final state result.

[0093] For calculating the interface temperature difference fluctuation index, the temperature gradient map in high-confidence thermal images is mainly used, and hotspot adjustment parameters are introduced as a weighted reference for data reliability and regional sensitivity. The slope of the response curve in the hotspot adjustment parameters (e.g., 1.2, 1.5, 2.0) is used to determine whether the image frame has undergone enhancement processing. If the slope is high and the saturation range is set reasonably, the system will increase the weight of the frame in the temperature difference range statistics. Furthermore, if the acquisition module records that the thermal image frame has undergone a dual adjustment mechanism of local Gamma compression + regional brightness stretching, the system can set the hotspot adjustment level to "Level 2 processing" and input this parameter as an entropy value sensitive factor into the global thermal difference fluctuation model, so that the model can perform appropriate response amplification processing on the frame data with strong adjustment capabilities and prominent thermal distribution during the aggregation process.

[0094] Furthermore, during the modeling process, the texture complexity index and hotspot adjustment parameters are also used to assist in identifying abnormal trajectory segments. For example, during the spatial distribution analysis of state indicators, if the texture complexity index of multiple consecutive frames of data within a trajectory segment is greater than 0.9, and the hotspot adjustment parameters show that the image has reached the third-level saturation compensation state, the system will comprehensively consider that there is a potential unstable phenomenon of structural texture anomalies and concentrated heat input in this area, thus marking this segment as a potentially high-risk segment and using it as a target area for priority response by subsequent control modules.

[0095] The calculation module 103 is used to perform dynamic feature modeling based on the physical feature information and the swing trajectory parameters, and to calculate multi-dimensional state indicators characterizing welding quality using time series feature extraction and spatial energy distribution analysis methods, including the molten pool stability index, the instantaneous intermetallic fusion uniformity index, and the interface temperature difference fluctuation index.

[0096] The calculation module 103 is the core of analysis and modeling in the device of this invention. Its function is to jointly process the multi-source physical feature information provided by the acquisition module 102 and the beam oscillation parameters of the laser welding head 101 to construct a dynamic feature model, and extract key indicators of the welding process state for real-time adjustment by the subsequent control module 104. The operation of this module includes steps such as data reception, feature standardization, time series analysis, spatial energy distribution modeling, and multi-dimensional index calculation, ensuring that the extracted indicators have clear physical meaning, sufficient sensitivity, and good stability, and can accurately reflect the physical evolution state between metals during the welding process.

[0097] After receiving the visual images, spectral features, and thermal imaging data output by the acquisition module 102, the calculation module first synchronizes the time axes of the three types of data, removing dropped frames and outliers, and standardizes each type of feature data, such as unifying units, eliminating background interference, and adjusting spatial resolution. Subsequently, according to the time progression of the welding process, all data are segmented and organized into feature sequences with equal time intervals to capture continuous dynamic change trends.

[0098] During the time series analysis phase, the calculation module employs a sliding window technique to extract local fluctuations, average trends, and peak characteristics. Simultaneously, it combines these with beam oscillation trajectory parameters (such as trajectory type, frequency, and amplitude) for paired modeling, thereby enabling causal correlation analysis of the process state. For example, if a visual image shows drastic fluctuations at the molten pool boundary within a certain period, and the thermal image also exhibits a sharp temperature jump accompanied by abrupt changes in the intensity of a certain element in the spectrum, the system marks this region as an abnormal state segment and initiates the index calculation process.

[0099] Regarding the definition and calculation of state indicators, the molten pool stability index is an indicator that measures the magnitude and frequency of changes in the shape of the molten pool boundary during welding. Specifically, it is calculated by extracting the molten pool edge contour from consecutive frames of a visual image, calculating the standard deviation and rate of change of the boundary contour. The greater the fluctuation in the boundary contour, the higher the index value, indicating molten pool instability. For example, when the boundary position change exceeds a preset threshold (e.g., 0.3 mm) in five consecutive frames and there is an oscillation frequency higher than 10 Hz, the stability index will increase significantly, suggesting a risk of welding spatter or local collapse.

[0100] The instantaneous intermetallic fusion uniformity index is used to assess the uniformity of mixing of dissimilar metal elements in the molten pool within a given time period. The index is calculated based on the emission intensities representing different metal elements in the spectral signal. After ensuring baseline calibration of the spectral signal, the system analyzes the spectral intensity ratios of representative metal elements such as aluminum and iron, and calculates their dispersion across different sampling points (or regions). If the aluminum / iron intensity ratios differ little across multiple sampling regions at the same time point, it indicates relatively uniform mixing, and the index value is low. Conversely, if local deviations or abrupt changes occur, the index increases, indicating uneven metal fusion.

[0101] The interface temperature fluctuation index measures the uniformity of heat input and the stability of heat diffusion by analyzing the lateral temperature distribution along the joint interface on the weld cross-section. It is calculated by extracting temperature values ​​from the weld centerline and several sampling points on both sides of the weld from a thermal image, and then calculating the rate of change and periodic oscillation amplitude of the temperature difference. If frequent temperature changes are found in a certain area, or if there is an alternation between high-temperature concentrations and low-temperature boundaries, the index value will increase, indicating a risk of heat input fluctuations in that area, which may lead to cracks or discontinuous interface structure.

[0102] The entire calculation process is executed once within each laser trajectory cycle, ensuring that the system can update the state evaluation results at a high frequency. The calculation module also organizes all indicators into a structured data format and binds them with timestamps and trajectory parameters to form a complete process state vector, which is then output to the control module 104.

[0103] To facilitate understanding of the module's workflow, an example is provided below. For instance, in a practical scenario of aluminum-steel dissimilar metal lap welding, to enhance elemental mixing in the weld zone and suppress the formation of brittle intermetallic compounds, the system is configured to have the laser beam output from the laser welding head oscillate along an elliptical trajectory. The oscillation frequency is set to 100 Hz, the oscillation amplitude to 1.5 mm, and the trajectory type to a counter-clockwise closed ellipse. The welding speed is 300 mm per minute, and the laser power is set to 2.2 kW.

[0104] During the welding process, the acquisition module simultaneously acquires three types of data at a rate of 30 frames per second: visible light images of the welding area, plasma excitation spectra, and infrared thermal imaging images. Every second is a processing cycle, and the calculation module 103 binds and analyzes these 30 frames of multi-source data with the trajectory parameters of the beam oscillation to complete one dynamic feature modeling and multi-dimensional state index extraction.

[0105] To complete the dynamic feature modeling, the calculation module first calls the swing trajectory parameters recorded in the control system, including trajectory type, frequency, and amplitude, and calculates the precise position of the laser beam on the weld joint for each frame based on the trajectory model in the system. For example, the laser beam in the first frame corresponds to the lower left corner of the elliptical trajectory, and in the 16th frame it is located in the upper right corner. Each frame of data corresponds precisely to the position of its laser irradiation point in the weld coordinate system, marked as positions P0 to P29.

[0106] Based on this frame's location information, the calculation module processes the acquired data. First, for visual images, a fixed threshold (e.g., pixel grayscale value greater than 180) is used to extract the molten pool region in each frame. Then, a contour analysis algorithm (such as contour fitting and edge expansion) is used to calculate the maximum horizontal width of the molten pool in each frame. The resulting 30 width values ​​are as follows (in millimeters):

[0107] [2.1, 2.2, 2.3, 2.0, 2.2, 2.1, 2.5, 2.4, 2.2, 2.3, ..., 2.0];

[0108] The calculation module performs standard deviation calculation on the molten pool width sequence to obtain the molten pool stability index, which is 0.18 mm in this case. In addition, the calculation module groups each frame of image according to the trajectory position and finds that the images corresponding to frame numbers P7–P10 have large fluctuations, which means that the molten pool is unstable when the laser beam is in the upper part of the trajectory, which is related to the possibility of energy concentration or reflection at that position.

[0109] Next, the spectral data corresponding to each frame is processed. The emission intensities of aluminum (396.15 nm) and iron (404.58 nm) are extracted from each frame's spectral signal, and the aluminum / iron ratio is calculated, resulting in the following sequence:

[0110] [1.10, 1.12, 1.09, 1.08, 1.15, 1.14, 1.07, 1.05, 1.20, ..., 1.06];

[0111] The calculation module calculates the standard deviation of this ratio sequence to obtain the instantaneous intermetallic fusion uniformity index, which is 0.17 in this case. Combined with the analysis of the spot trajectory position, it was found that the proportion of aluminum element continuously increases in frames P14–P18, indicating that the fusion in this region is insufficient and the aluminum segregation phenomenon is more serious.

[0112] Subsequently, the infrared thermal imaging data is processed. In each frame of the thermal image, the system selects 5 equidistant pixels on each side of the weld centerline and calculates the average temperature difference between the left and right sides (left temperature - right temperature), forming the temperature difference time series as follows (unit: degrees Celsius):

[0113] [1.9, 2.1, 2.0, 2.2, 1.8, 2.3, 2.1, ..., 1.9];

[0114] The calculation module takes the difference between the maximum and minimum values ​​of the sequence to obtain the interface temperature fluctuation index, which is 0.5℃ in this case. Through trajectory matching analysis, it was found that the maximum temperature fluctuation occurs in the region where the beam swings to the upper right corner of the ellipse (frames P20–P24), suggesting that the heat input in this region is slightly higher, and there may be heat accumulation.

[0115] Finally, the three status indicators for this processing cycle are output as follows:

[0116] Molten pool stability index: 0.18 mm;

[0117] Instantaneous intermetallic fusion uniformity index: 0.17;

[0118] Interface temperature fluctuation index: 0.5℃;

[0119] In addition, each indicator is accompanied by a spatial location marker, indicating the trajectory segment where the potentially defective welding area is located. This is used to guide the control module 104 to perform parameter adjustments, such as reducing laser power, reducing the swing amplitude, or fine-tuning the swing trajectory, so that the welding behavior in that area tends to be stable.

[0120] Furthermore, the computing module is specifically used for:

[0121] The system receives physical feature information, including risk prediction level and trajectory location label, output by the acquisition module, and divides the physical feature information into multiple spatial partition data segments according to the trajectory location, wherein each data segment corresponds to a sampling area under a swing trajectory.

[0122] Based on the risk prediction level, weighting coefficients are assigned to different spatial partition data segments. These weighting coefficients are used to increase the contribution weight of data corresponding to high-risk areas in the state indicator modeling.

[0123] When calculating the molten pool stability index, dynamic information fragments from the boundary with higher risk level are used first, and the boundary profile change frequency is weighted and averaged by time-series superposition to form a stability state result that reflects the dominant behavior of the high-fluctuation region.

[0124] When calculating the instantaneous inter-metal fusion uniformity index, key frame data from the spectral feature information are selected, and the index is adjusted based on the standard deviation of the spectral ratio of different sampling points within the same swing trajectory region, combined with the risk weight, to highlight the dominant contribution of abnormal fusion behavior.

[0125] When calculating the interface temperature fluctuation index, the temperature gradient information of each spatial partition is used to calculate the temperature difference range within the region, and all partition results are aggregated into a global temperature fluctuation index by risk weighting, so as to achieve a sensitive response to the uneven heat input at the interface of dissimilar metals.

[0126] In this implementation, the computing module is primarily responsible for structuring and dynamically modeling the physical feature information output by the acquisition module to generate multi-dimensional state indicators characterizing welding quality, providing a quantitative basis for subsequent control and adjustment. This process revolves around three core indicators: the molten pool stability index, the instantaneous intermetallic fusion uniformity index, and the interface temperature difference fluctuation index. The entire workflow is centered on risk perception, integrating trajectory location tags to achieve spatially differentiated feature modeling and a weighted response mechanism.

[0127] First, the calculation module receives output data from the acquisition module. This data includes not only the fusion results of visual images, plasma excitation spectra, and infrared thermal imaging signals, but also trajectory location labels and risk prediction levels for each data segment. These labels indicate the specific location of the data segment within the welding trajectory, as well as the potential probability level of quality fluctuations in that area during the welding process. This information allows the calculation module to spatially partition the data according to the actual oscillation trajectory of the laser beam.

[0128] The calculation module divides the data of the entire welding process according to the oscillation trajectory, discretizing the entire welding path into multiple spatial partition data segments. Each data segment corresponds to the data set under a certain time period or trajectory interval in the welding path. This partitioning method allows the system to model and analyze on a region-by-region basis, enhancing the model's ability to perceive spatial location-related features. For example, when performing a welding task with an "∞" shaped oscillation trajectory, the system can divide the trajectory into five partitions: upper left ring, upper right ring, intersection center, lower left ring, and lower right ring. The data set within each partition includes the aforementioned multi-source physical features.

[0129] After obtaining the partitions, the system reads the risk prediction level corresponding to each partition and assigns a weighting coefficient based on this. The risk prediction level is typically a discrete value from 0 to 3, corresponding to low to high risk. According to the set risk-weight mapping function, the system assigns higher weighting coefficients to high-risk areas. This weighting coefficient plays a role in the subsequent calculation of state indicators, aiming to increase the influence of high-risk area data on the overall state assessment. For example, if a partition is predicted to be level 3 (high risk), its weighting coefficient might be set to 1.5; while the weighting coefficient for level 0 areas would be 0.7. This mechanism effectively enhances the model's responsiveness to welding anomalies in key areas, facilitating early detection of anomalies.

[0130] In calculating the molten pool stability index, the system focuses on analyzing the trajectory of the molten pool boundary over time in visual image data. Specifically, the system extracts the boundary contour of the molten pool in each frame of the image and calculates the frequency and amplitude of boundary changes between consecutive frames. For data segments with higher risk levels, the system adopts their boundary contour change features with higher weight, forming a sample set that dominates the judgment of molten pool stability. Subsequently, the system performs time-overlay analysis on these samples, calculates the weighted average of their oscillation frequencies and the dispersion of their amplitude changes, and obtains a quantitative index describing the molten pool boundary fluctuation characteristics during the welding process. The smaller the index value, the more stable the molten pool boundary; the greater the fluctuation, the more uneven the laser input energy or the more severe the interface disturbance. For example, when welding dissimilar metals such as copper and steel, due to the large difference in melting points between the two materials, the molten pool at the copper end fluctuates significantly. In the initial stage of heat input, the molten pool boundary amplitude reaches a maximum of 1.5 mm, and the oscillation frequency fluctuates around 20 Hz, while the steel end is relatively stable. At this time, after extracting and weighting the dynamic features of the copper end through high-frequency image sequences, the overall molten pool stability index is significantly lowered, indicating that the control module needs to increase the energy input balance adjustment in this area.

[0131] When calculating the instantaneous intermetallic fusion uniformity index, the system first identifies keyframes in multi-frame spectral signals, i.e., sampling moments with significant changes in spectral line intensity and obvious alloy element characteristics. The system selects several representative sampling points in each trajectory partition, extracts the spectral line intensities at specific wavelengths (e.g., Fe, Ni, Cu), and calculates the standard deviation of these spectral line ratios. A larger standard deviation indicates uneven element distribution in the fusion region, suggesting the presence of intermetallic diffusion barriers or uneven energy distribution at that welding instant. A risk weighting mechanism is also applied here: the standard deviation in high-risk regions is given a higher index adjustment influence, making it a dominant contributor to the output fusion uniformity index. For example, in the middle intersection section of a Lissajous trajectory, due to reduced welding speed and high energy density, the Cu-Fe spectral line intensity ratio in this region changes significantly (e.g., from 1.2 to 2.5), indicating drastic changes in intermetallic mixing and poor fusion uniformity. The system, considering the risk level of 3 for this region, increases its contribution weight, making the data from this region decisive for the overall fusion uniformity index result.

[0132] For calculating the interface temperature fluctuation index, the system focuses on analyzing the spatial variation of the temperature gradient in the infrared thermal imaging signal. Each trajectory partition corresponds to an image sampling area. The system first extracts the temperature values ​​of several sampling points within this area, calculates the range between the highest and lowest temperatures within the area, and the gradient rate of change of the temperature distribution. These indicators reflect whether the heat input is uniformly distributed at the interface during the welding process. Then, the system weights and aggregates the temperature range indicators from all spatial partitions, with the weights still set based on the risk level, ultimately forming the global interface temperature fluctuation index. The more severe the temperature fluctuation, the more significant the unevenness of heat input at different metal interfaces during the welding process, which can easily lead to problems such as stress concentration, lack of fusion, or cracks. Taking a specific welding operation as an example, in the lower right corner of the oscillation trajectory, due to the faster movement speed of the laser beam and insufficient heat accumulation, the temperature difference in the infrared image of this segment reached as high as 80℃, while the temperature difference in the upper left corner was only 40℃. After assessing the risk level, the system assigned a higher weighting coefficient to the lower right corner segment, resulting in a higher interface temperature fluctuation index in the final output. This indicated that there was an energy distribution offset problem in this segment, and the welding speed and oscillation frequency needed to be adjusted through the control module.

[0133] For example, in a dissimilar metal laser welding process, assuming the current welding task is a butt weld between copper and stainless steel components, the weld area is covered by an elliptical-straight line combination laser oscillation trajectory. After the welding device is started, the acquisition module continuously acquires visual images, plasma excitation spectra, and infrared thermal imaging signals of the welding area at a high frame rate. Simultaneously, each frame of data is labeled with a trajectory position and the risk prediction level is marked according to the initial stage stability prediction model. Suppose that segment C in the welding path (closer to the copper substrate) is identified as having high thermal conductivity and low initial energy absorption, and is marked as risk level 3. The visual image of the corresponding data segment shows a discontinuous molten pool boundary, drastic changes in the intensity of the Cu element spectral lines in the spectral information, and the infrared image reflects a temperature difference of up to 95°C in this area.

[0134] After receiving the raw data, the calculation module first divides the entire welding process into six trajectory partitions, A to F, based on trajectory position labels. Each partition's spatial data set stores the image, spectral, and thermal imaging data, as well as the risk level within that area. Subsequently, the system assigns a weight value to each partition according to a preset risk-weight mapping function. For example, risk level 0 corresponds to a weight of 0.8, level 1 to 1.0, level 2 to 1.2, and level 3 to 1.5. This ensures that data from high-risk areas have a greater impact in subsequent state modeling.

[0135] In calculating the molten pool stability index, the system focuses on processing the boundary contour features in the visual image data. For each frame, the system uses an image edge detection algorithm (such as Canny edge detection) to extract the outer contour of the molten pool and marks its boundary coordinates. By tracking the offset of the same boundary point in consecutive frames, the frequency and amplitude of boundary changes can be obtained. The system performs statistics for each partition separately, and then combines the results by weighting the partitions. Taking partition C as an example, its average boundary amplitude is 1.8 mm and the change frequency is 22 Hz, while most other partitions are less than 0.6 mm and 12 Hz. Therefore, partition C becomes the dominant index component, and its weighted result significantly affects the overall index. The final molten pool stability index output value is 0.63 (normalized value, range 0–1), indicating that the current welding process of the system has high-frequency unstable boundary fluctuations.

[0136] The calculation of the instantaneous intermetallic fusion uniformity index focuses on excitation spectral data. In each frame of the spectrum, the system extracts the spectral line intensities at key wavelengths such as Cu, Fe, and Cr, and calculates the standard deviation of the element ratios for different sampling points within the partition. In partition C, the Cu / Fe spectral line intensity ratios at multiple sampling points differ significantly; for example, P1 is 1.1, P2 is 1.9, P3 is 2.3, and P4 is 1.3, with a standard deviation of 0.52. In contrast, low-risk partitions, such as partition A, have a standard deviation of only 0.08. Based on the high risk level and high ratio standard deviation of partition C, the system assigns an index enhancement factor to it, making it dominate the fusion uniformity result. The final index output value is 0.47 (normalized range 0–1), indicating significantly uneven material mixing distribution, potentially indicating incomplete diffusion or interface inclusions.

[0137] The interface temperature fluctuation index is calculated using temperature data from infrared thermal imaging images. Within each trajectory partition, the system sets several fixed sampling points (e.g., a 5x5 grid distribution), extracts their temperature values, and calculates the difference between the highest and lowest temperatures (range) and the average gradient. Assuming the temperature range in partition C is 85°C and the average gradient is 22°C / mm, while the average values ​​for other partitions are only 35°C and 7°C / mm respectively, the system aggregates the range and gradient data from all partitions according to risk weights to form an overall temperature fluctuation index. The output result is 0.68, significantly higher than the empirical threshold of 0.4 for this type of welding process, further confirming the presence of uneven heat input in this segment.

[0138] Based on the calculation results of the above three status indicators, the control module can further locate high-risk areas in the welding process and execute compensatory control strategies such as increasing laser power, refining oscillation frequency, and adjusting trajectory amplitude to achieve the goal of balanced welding quality and minimizing defects. This entire set of status indicators constitutes the central data source of the closed-loop control chain. Its construction process comprehensively integrates spatial information, temporal dynamics, and risk assessment, forming a crucial foundation for intelligent sensing and control of dissimilar metal welding.

[0139] The entire calculation process features three technological breakthroughs. First, the sampling area is explicitly divided by the trajectory location and linked to the predicted risk level, providing a clear spatial reference and response basis for the condition assessment. Second, a weighting mechanism is introduced, which does not average all data but instead enhances the data contribution of key areas through risk perception, thereby strengthening the model's sensitivity to welding defects. Third, the condition indicators are all quantifiable and calculated based on direct sensor data, rather than subjective evaluations, thus possessing good repeatability and industrial application value.

[0140] In summary, the computation module described in this embodiment, through spatial partitioning, weight adjustment, temporal analysis, and data fusion, fully leverages the value of the physical characteristic information provided by the acquisition module, achieving accurate modeling of the multidimensional states of welding quality and significantly improving the intelligent perception and control capabilities of the welding process. Through the molten pool stability index, instantaneous intermetallic fusion uniformity index, and interface temperature difference fluctuation index output by this module, the control module can acquire the multidimensional state changes of the welding process in real time, forming a feedback path highly coupled with the physical mechanism, ensuring a stable, efficient, and defect-free dissimilar metal welding process.

[0141] The control module 104 is used to receive multi-dimensional status indicators and perform joint optimization adjustment of laser parameters and trajectory parameters based on the control weight matrix, including real-time adjustment of laser power, welding speed, oscillation frequency and trajectory shape.

[0142] The control module 104 is the core component of the present invention to realize the closed-loop adaptive control function. Its main function is to dynamically adjust the key parameters such as laser power, welding speed, oscillation frequency and oscillation trajectory shape of the laser welding head according to the multi-dimensional state indicators output in real time by the calculation module 103. This effectively addresses the quality fluctuations caused by material differences, uneven heat diffusion or energy input offset during the welding of dissimilar metals, ensuring the stability of the welding process and the reliability of the final joint.

[0143] The control module is directly connected to the computing module in the system, receiving three main status indicators from the computing module: molten pool stability index, instantaneous intermetallic fusion uniformity index, and interface temperature difference fluctuation index. The molten pool stability index measures the degree of shape fluctuation of the molten pool boundary over a period of time, typically represented by the standard deviation of the maximum lateral width of the molten pool in multiple consecutive frames of images; a higher value indicates a more unstable molten pool. The instantaneous intermetallic fusion uniformity index reflects the uniformity of mixing dissimilar metals in the weld molten pool, generally quantified by the standard deviation of the distribution of the emission intensity ratio of representative metal elements in the spectrum; a lower value indicates more uniform mixing. The interface temperature difference fluctuation index represents the range of temperature difference fluctuation on both sides of the weld's central axis per unit time, calculated from infrared thermal imaging data as the difference between the maximum and minimum temperature differences; a higher value indicates stronger heat input non-uniformity, suggesting potential thermal stress accumulation or cracking.

[0144] The control module has a pre-set set of control weight matrices to determine the adjustment priority based on the importance of these three indicators and the degree of abnormality of their current values. For example, when the molten pool stability index is close to the set threshold (e.g., 0.3 mm) while the other two indicators are still within the normal range, the system prioritizes reducing the laser power or slightly slowing down the welding speed to reduce molten pool disturbance. When the metal fusion uniformity index exceeds the threshold (e.g., 0.2), the system will prioritize changing the shape of the oscillation trajectory, for example, from an ellipse to an "∞" shape, to prolong the residence time difference of the laser on the two metals at the interface and improve the lateral mixing efficiency.

[0145] At the control strategy execution level, this module rapidly assesses the status indicators within each second cycle and incrementally optimizes welding parameters to prevent drastic changes from causing new instabilities. For example, if the interface temperature fluctuation index rises to 0.8°C in a certain cycle, the control module can lower the oscillation frequency from 100 Hz to 95 Hz and slightly increase the oscillation amplitude to ensure sufficient heat transfer time for the laser in the boundary region, thereby balancing the interface temperature distribution. After each parameter adjustment, the control module feeds the adjustment result back to the system and waits for the data input and indicator updates for the next cycle to determine whether further adjustments are needed.

[0146] To prevent parameter oscillations or over-control, the control module is equipped with a suppression threshold. When the status indicators fluctuate within an acceptable range (e.g., the change is less than 5%), no adjustment will be performed, and the current welding parameters will remain unchanged. When multiple indicators deviate from the set value at the same time, the control module will adjust them sequentially according to the preset priority to prevent mutual interference caused by adjusting multiple parameters at the same time.

[0147] The entire control module is implemented based on a programmable logic controller (PLC) or an embedded control board, and is equipped with a real-time scheduling mechanism for executing laser control commands, motion control commands (such as the frequency of the swing motor), and communication interrupt response. Combining status indicators and a control weight matrix, the module can complete a full calculation-judgment-execution cycle every second, ensuring timely system response and stable adjustment.

[0148] To more clearly illustrate the actual operating mechanism of the control module 104, the following uses a specific welding task as an example to demonstrate its entire process from receiving indicators, referencing the weight matrix, judging the status, executing adjustments, to forming a feedback closed loop.

[0149] In a lap welding operation of dissimilar aluminum and steel, the system periodically receives three status indicators from the calculation module 103 every second. Within a certain cycle, the molten pool stability index is 0.25 mm, the instantaneous intermetallic fusion uniformity index is 0.22, and the interface temperature fluctuation index is 0.6 degrees Celsius. The preset control weight matrix in the system is shown below:

[0150] ;

[0151] This matrix was obtained by engineers through multiple rounds of experiments. Specifically, in welding experiments, the changing trends of the three state indicators were observed after adjusting a single parameter individually. Principal component analysis and regression analysis were used to deduce the relative influence of each parameter on each state indicator, followed by normalization to form the proportional matrix described above. Each number in the matrix represents the priority weight of the control parameter in adjusting the corresponding state indicator; a higher weight indicates that the parameter is more sensitive to improving the control effect on that indicator.

[0152] During this cycle, the molten pool stability index reached 0.25, close to the system's set threshold of 0.30, indicating slight instability. The control module, based on the index's weight in the matrix, weightedly adjusted the laser power (0.6), welding speed (0.3), and oscillation frequency (0.1). The system employs a joint optimization strategy, determining the adjustment range allocation according to the weighted coefficient ratio. For example, with a maximum acceptable total adjustment range of 10% set for this cycle, the laser power would be adjusted by 6%, the welding speed by 3%, and the oscillation frequency by 1%. During execution, the laser power decreased from 2.2 kW to approximately 2.07 kW, the welding speed decreased from 300 mm / min to 291 mm / min, and the oscillation frequency slightly decreased from 100 Hz to 99 Hz.

[0153] Meanwhile, the fusion uniformity index was 0.22, exceeding the threshold of 0.2. The system identified this index as abnormal and triggered further joint adjustments. In the weight matrix corresponding to this index, the oscillation frequency (0.4) and oscillation trajectory shape (0.3) had high proportions. Therefore, the system further reduced the oscillation frequency by 3%, from 99 Hz to approximately 96 Hz, and simultaneously switched the trajectory from the original ellipse to an "∞" shape (lemniscate) to enhance the scanning span of the laser beam in the lateral direction and improve the lateral metal mixing uniformity.

[0154] When the interface temperature difference fluctuation index is 0.6, although it does not exceed the set threshold (0.8), it is in a critical state. Based on the control matrix of the temperature difference index, the system only slightly increases the swing amplitude (corresponding to the energy coverage range of the adjustment trajectory), from 1.5 mm to 1.6 mm, in order to enhance the heat diffusion coverage area, but does not perform large-scale movements.

[0155] After all parameters are adjusted, the control module writes the new parameter set to the laser head and motion control unit, generates an adjustment execution report, returns it to the system master control node, and continues to collect and verify whether the adjustment effect has achieved the expected results in the next cycle. If the adjusted indicators show a clear downward trend, the system will maintain the parameters and enter the observation state; if the indicators continue to deteriorate or other indicators show abnormal linkage, the system will restart the joint adjustment process according to the new state.

[0156] As can be seen from the above specific process, the role of the control module 104 in this invention is not passive response, but rather based on a clear, adjustable, and quantifiable weight system to achieve real-time and refined linkage control for multiple indicators and parameters, thereby providing stable and efficient quality control support for the laser welding process of dissimilar metals.

[0157] Furthermore, the control module is specifically used for:

[0158] The system receives a multi-dimensional state index output by the calculation module, which includes trajectory location labels and risk prediction levels, and divides the oscillating trajectory into multiple trajectory segments, each trajectory segment corresponding to a subset of state indices.

[0159] Based on the risk prediction level of the trajectory segment, a trajectory segment-level control weight sub-matrix is ​​constructed to dynamically allocate the adjustment priorities of the laser power, welding speed, oscillation frequency and trajectory shape among the trajectory segments.

[0160] Based on the spatial distribution differences of the state indicators in each trajectory segment, trajectory segments with local significant anomaly characteristics are identified and assigned adjustment priority weights, so that the control response first acts on the region with significant state fluctuations.

[0161] Based on the risk prediction level corresponding to each trajectory segment, the upper limit of its control adjustment range is set so that the parameters corresponding to high-risk segments allow for a wider range of laser power changes, swing amplitude adjustments, or trajectory shape switching.

[0162] The parameter adjustment strategy throughout the entire oscillation cycle is jointly scheduled within and between trajectory segments to maintain the overall continuity of the welding process while achieving enhanced control and compensation for local quality fluctuation areas.

[0163] The adjusted laser parameters and trajectory parameters are synchronously sent to the laser welding head to implement segmented response control to optimize the welding quality of the dissimilar metal bonding area.

[0164] In the laser welding of dissimilar metals, the welding quality is affected by a variety of complex factors, especially in the high-temperature, high-energy-density welding region. The dynamic changes in the molten pool, the fusion uniformity between metals, and the thermal diffusion characteristics of the interface can all fluctuate to varying degrees as the laser beam's oscillation trajectory changes. To achieve precise control of this highly dynamic process, the control module must possess the ability to make fine-grained response adjustments based on state perception results. The control module proposed in this invention is based on this core idea, built upon an understanding of the coupling relationship between the oscillation trajectory and physical state indicators, thereby achieving dynamic, segmented optimization of laser parameters and trajectory parameters.

[0165] Specifically, the control module first receives multi-dimensional state indicators output from the calculation module. These state indicators are derived from the fusion, modeling, and analysis of multi-source data collected by the acquisition module during the welding process. They contain information closely related to welding quality, such as the molten pool stability index, the instantaneous intermetallic fusion uniformity index, and the interface temperature difference fluctuation index. Simultaneously, the state indicators also include a specific location label for each sampling point on the oscillation trajectory, as well as the corresponding risk prediction level, thus providing a precise mapping of the state in both spatial and risk dimensions. Based on this, the control module divides the entire oscillation trajectory into several trajectory segments according to temporal and spatial order. Each segment corresponds to a continuous trajectory path and is associated with an independent subset of state indicators.

[0166] For example, imagine a laser beam oscillating along a sinusoidal or lissajous curve, with multiple rotational and intersecting regions within a single oscillation cycle. In actual welding, due to the longer residence time of the laser beam in specific trajectory segments or localized energy density concentrations, these areas are highly likely to experience enhanced molten pool disturbance, uneven metal fusion, or significant temperature fluctuations. The control module can identify trajectory segments with obvious quality fluctuation characteristics by analyzing subsets of state indicators from these segments. For instance, within a given segment, if the molten pool stability index exhibits drastic fluctuations, fusion uniformity decreases, and temperature differences widen, that segment can be classified as a high-risk segment.

[0167] After identifying high-risk sections, the control module constructs a set of trajectory section-level control weight sub-matrices for different trajectory sections. These weight sub-matrices are used to prioritize the adjustment of laser and trajectory parameters across different trajectory sections. The adjustment priority of each control parameter, such as laser power, welding speed, oscillation frequency, and trajectory shape, is adjusted according to the risk level of that section. The control module prioritizes responses to high-risk sections by assigning them higher weights within the sub-matrices. For example, in a typical sub-matrice, a section with a "high" risk level might have a laser power adjustment weight of 0.8 and an oscillation frequency adjustment weight of 0.7, while a section with a "low" risk level might have weights of only 0.2 and 0.1, respectively. This dynamic weighting mechanism allows the control system resources to focus primarily on addressing areas most likely to result in quality defects.

[0168] Furthermore, the control module not only needs to identify the risk level but also analyze the spatial distribution characteristics of the state indicators within the trajectory segment. The key to this step is extracting regions with significant local anomalies and assigning them additional adjustment priorities. For example, within a trajectory segment, although most areas may be stable, several locations near the trajectory boundary may exhibit severe vibrations at the edge of the melt pool. These local anomalies will be identified through spatial clustering analysis or boundary dynamic detection methods and given higher local response weights in the control matrix. The resulting adjustment strategy will no longer be a uniform adjustment of the entire segment but will achieve more refined spatial control, concentrating the control response on the areas with the most prominent problems.

[0169] While assigning adjustment priorities, the control module also sets an upper limit for the control adjustment amplitude of each trajectory segment. Specifically, for trajectory segments determined to be high-risk, the allowable range of laser power variation can be expanded, for example, from ±5% to ±15%; the adjustment range of the oscillation amplitude can also be increased, for example, from ±0.3 mm to ±1 mm; and even the trajectory shape can be switched from the original sinusoidal oscillation to an elliptical or lissajous mode to achieve a more uniform energy distribution. This strategy can provide greater adjustment space for high-risk areas after the control weights are determined, thereby improving the system's response flexibility and correction capability.

[0170] To ensure the continuity and stability of the overall welding process, the control module also needs to coordinate control strategies between different trajectory segments. Therefore, based on control within each trajectory segment, the module further performs joint scheduling between trajectory segments. This process not only prioritizes the adjustment of high-risk segments but also avoids abrupt parameter jumps between different segments, which could lead to new molten pool disturbances. For example, during the transition from a low-risk segment to a high-risk segment, the laser power should not be instantly increased from 80% to 100%, but rather a smooth transition should be achieved with the support of the control algorithm, realizing seamless connection of the continuous control trajectory. The joint scheduling mechanism ensures that the overall process parameters remain stable even when faced with local disturbances, thereby improving the overall weld quality consistency.

[0171] After the control strategy is determined, the control module synchronously sends the calculated final laser parameters and trajectory parameters to the laser welding head, which then executes the adjustment commands at the physical level. It is worth noting that there is a time delay between parameter sending and laser head response. Therefore, the control module needs to introduce a predictive mechanism, combining the welding head's response characteristics, oscillation period, and welding speed to pre-schedule control commands, ensuring timely and accurate response. Furthermore, to prevent frequent laser welding head adjustments from causing mechanical fatigue or control system instability, the control module sets a minimum adjustment interval and a minimum stabilization period based on the rate of change between segments, thereby balancing the trade-off between control response rate and equipment lifespan.

[0172] To illustrate with a complete example, when welding aluminum-titanium dissimilar metal joints, the significant difference in thermal conductivity and melting point between the two metals makes them highly susceptible to localized overheating or poor fusion at the interface. The initial oscillation trajectory is set to an ellipse, with an oscillation frequency of 50 Hz and a power of 1.2 kW. The acquisition module identifies that at both ends of the trajectory's major axis, due to a slightly longer laser beam dwell time, the molten pool stability index decreases rapidly, and the interface temperature fluctuation index increases. The calculation module identifies this area as a high-risk trajectory segment. Accordingly, the control module sets the power control weight to 0.9 and the oscillation frequency adjustment weight to 0.8 in the corresponding segment, allowing a maximum power increase of ±20%. Simultaneously, to avoid new disturbances caused by sudden power changes, the control module introduces a time-series smoothing mechanism for power changes, causing it to linearly change to the new set value within 5 oscillation cycles. The oscillation trajectory then smoothly switches from an ellipse to a figure-eight pattern within 7 cycles to expand the lateral coverage and improve the uniformity of heat distribution. After this section, the control module gradually calls back the parameters to restore the standard trajectory configuration, thereby achieving dynamic compensation for local poor quality trends.

[0173] In summary, the control module of this invention provides a complete laser welding control strategy with spatial positioning and risk response capabilities. It can perform segmented enhanced control of local abnormal areas while ensuring overall welding stability. Its core lies in constructing a multi-dimensional control weight matrix based on state-aware data and realizing spatial segmentation of parameter adjustment, risk priority allocation, and temporal joint scheduling, thereby significantly improving the quality consistency, stability, and process robustness of dissimilar metal joint welding processes.

[0174] Although this application discloses preferred embodiments as described above, it is not intended to limit this application. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of this application. Therefore, the scope of protection of this application should be determined by the scope defined in the claims of this application.

Claims

1. A laser welding device for dissimilar metals based on oscillating beam and molten pool condition monitoring, characterized in that, include: A laser welding head is used to output a laser beam and apply welding energy to dissimilar metal joints. The laser welding head has a beam oscillation function, which enables the laser beam to perform energy scanning along a nonlinear trajectory within the welding area based on preset oscillation trajectory parameters. The oscillation trajectory parameters include trajectory type, oscillation frequency, and oscillation amplitude. The acquisition module is used to synchronously acquire multi-source data of the welding area at a high frame rate during laser welding, including visual images of the weld pool, plasma excitation spectra and infrared thermal imaging signals, and extract physical feature information characterizing the degree of metal mixing, interface diffusion behavior and dynamic fluctuations of the weld pool based on a multi-modal fusion method. The calculation module is used to perform dynamic feature modeling based on the physical feature information and the swing trajectory parameters, and to calculate multi-dimensional state indicators characterizing welding quality using time series feature extraction and spatial energy distribution analysis methods, including molten pool stability index, instantaneous intermetallic fusion uniformity index and interface temperature difference fluctuation index. The control module is used to receive multi-dimensional status indicators and perform joint optimization adjustment of laser parameters and trajectory parameters based on the control weight matrix, including real-time adjustment of laser power, welding speed, oscillation frequency and trajectory shape; Specifically, the acquisition module is used for: In the initial stage of welding, a weld area stability prediction model is established based on the oscillation trajectory parameters of the laser welding head and the preset metal combination characteristics, which is used to predict the potential welding quality fluctuation risk level at each trajectory position point throughout the entire oscillation cycle. Based on the output of the stability prediction model, the swing trajectory is divided into multiple dynamic acquisition areas. For areas with high predicted risk levels, improved acquisition parameters are automatically configured, including image resolution improvement, thermal image refresh frequency increase, or spectral integration time extension. In the actual welding process, the acquisition module performs local enhanced sampling on high-risk areas according to the configuration of the dynamic acquisition area. The spatial resolution or temporal density of the acquired visual images, plasma excitation spectra and infrared thermal imaging signals is higher than that of the standard area. The data obtained from augmented sampling is labeled as high-confidence feature segments, and its corresponding trajectory location label and risk prediction level are attached, which are used as the priority input for keyframes in subsequent modeling. The physical feature information extracted from the standard area and the high-risk area is uniformly structured and output as physical feature information to characterize the degree of metal mixing, interface diffusion behavior and dynamic fluctuation of the molten pool, and output to the calculation module for dynamic feature modeling based on the physical feature information.

2. The dissimilar metal laser welding device based on oscillating beam and molten pool state monitoring according to claim 1, characterized in that, The calculation module is specifically used for: The system receives physical feature information, including risk prediction level and trajectory location label, output by the acquisition module, and divides the physical feature information into multiple spatial partition data segments according to the trajectory location, wherein each data segment corresponds to a sampling area under a swing trajectory. Based on the risk prediction level, weighting coefficients are assigned to different spatial partition data segments. These weighting coefficients are used to increase the contribution weight of data corresponding to high-risk areas in the state indicator modeling. When calculating the molten pool stability index, dynamic information fragments from the boundary with higher risk level are used first, and the boundary profile change frequency is weighted and averaged by time-series superposition to form a stability state result that reflects the dominant behavior of the high-fluctuation region. When calculating the instantaneous intermetallic fusion uniformity index, key frame data from the spectral feature information are selected, and the index is adjusted based on the standard deviation of the spectral ratio of different sampling points within the same swing trajectory region, combined with risk weights, to highlight the dominant contribution of abnormal fusion behavior. When calculating the interface temperature fluctuation index, the temperature gradient information of each spatial partition is used to calculate the temperature difference range within the region, and all partition results are aggregated into a global temperature fluctuation index by risk weighting, so as to achieve a sensitive response to the uneven heat input at the interface of dissimilar metals.

3. The dissimilar metal laser welding device based on oscillating beam and molten pool state monitoring according to claim 2, characterized in that, The control module is specifically used for: The system receives a multi-dimensional state index output by the calculation module, which includes trajectory location labels and risk prediction levels, and divides the oscillating trajectory into multiple trajectory segments, each trajectory segment corresponding to a subset of state indices. Based on the risk prediction level of the trajectory segment, a trajectory segment-level control weight sub-matrix is ​​constructed to dynamically allocate the adjustment priorities of the laser power, welding speed, oscillation frequency and trajectory shape among the trajectory segments. Based on the spatial distribution differences of the state indicators in each trajectory segment, trajectory segments with local significant anomaly characteristics are identified and assigned adjustment priority weights, so that the control response first acts on the region with significant state fluctuations. Based on the risk prediction level corresponding to each trajectory segment, the upper limit of its control adjustment range is set so that the parameters corresponding to high-risk segments allow for a wider range of laser power changes, swing amplitude adjustments, or trajectory shape switching. The parameter adjustment strategy throughout the entire oscillation cycle is jointly scheduled within and between trajectory segments to maintain the overall continuity of the welding process while achieving enhanced control and compensation for local quality fluctuation areas. The adjusted laser parameters and trajectory parameters are synchronously sent to the laser welding head to implement segmented response control to optimize the welding quality of the dissimilar metal bonding area.

4. The dissimilar metal laser welding apparatus based on oscillating beam and molten pool state monitoring according to claim 1, characterized in that, The acquisition module is also used for: When the risk prediction level at a certain swing trajectory location point reaches a set high-risk threshold, the visual image enhancement acquisition process is automatically started. The visual image enhancement acquisition process includes: performing multi-scale wavelet reconstruction processing on the visual image acquired at the high-risk trajectory location point, and combining edge-guided filtering and local contrast enhancement algorithms to enhance the texture saliency of the molten pool edge, interface disturbance and metal inter-metal detail structure contained in the visual image. Based on the grayscale histogram distribution information of infrared thermal imaging signals collected at high-risk trajectory locations, the hot spot concentration index of the thermal imaging channel is dynamically calculated. When the hot spot concentration exceeds the adaptive saturation threshold, the slope of the response curve and the upper limit parameter of the saturation of the infrared thermal imaging signal are adjusted in real time, so that the infrared thermal imaging signal can improve the overall contrast of the low-temperature area while maintaining the details of the high-temperature area. The enhanced visual image and the infrared thermal imaging signal with the adjusted response curve are aligned at the pixel level and geometrically registered between channels to form composite enhanced frame data with spatial consistency and modal complementarity. The composite enhanced frame data synchronously inherits its corresponding trajectory position label and risk prediction level information. The composite enhanced frame data is structured and marked as high-confidence feature segments, and texture complexity index and hotspot adjustment parameters are added as auxiliary modeling factors during the structuring process to provide a basis for region refinement weights when outputting to the calculation module. The high-confidence feature fragments are uniformly output as part of the physical feature information used to characterize the degree of metal mixing, interface diffusion behavior and dynamic fluctuations of the molten pool, and are fused with the standard area acquisition results and input into the calculation module to enhance the accuracy and response sensitivity of the modeling results at the high-risk trajectory location points.

Images