Method and system for optimizing welding parameters of laser welding gun

Abstract

The invention relates to the technical field of laser welding gun welding parameter optimization, in particular to a laser welding gun welding parameter optimization method and system.The method comprises the following steps that when a laser welding gun executes a welding task, a non-key area in a welding seam is selected, and detection disturbance is applied to core welding parameters; when the detective disturbance is applied, capturing transient response characteristics of the molten pool to the detective disturbance by using a sensor; the transient response characteristics are compared with the physical correlation mode, and interference factors causing welding seam quality fluctuation are recognized; and performing compensation adjustment on the core welding parameters according to the interference factors. The accuracy of compensation adjustment of the welding parameters is improved.

Description

Technical Field

[0001] This invention relates to the technical field of laser welding torch welding parameter optimization, specifically to a method and system for optimizing laser welding torch welding parameters. Background Technology

[0002] In modern industrial manufacturing, laser welding technology is widely used in the production of critical components where structural integrity is paramount due to its high precision and high energy density. To ensure weld quality, advanced laser welding systems are typically equipped with specialized methods to optimize welding parameters, aiming to consistently produce welds with ideal penetration depth, uniform width, and minimal internal defects. However, the situation becomes complex when these standard parameters, determined in ideal laboratory environments, are implemented on large-scale, continuous industrial production lines. Summary of the Invention

[0003] The purpose of this invention is to address the aforementioned shortcomings by proposing a method and system for optimizing laser welding gun welding parameters.

[0004] The present invention adopts the following technical solution:

[0005] A method for optimizing laser welding torch welding parameters, the method comprising the following steps:

[0006] When performing welding tasks with a laser welding gun, select non-critical areas in the weld and apply probing perturbations to the core welding parameters;

[0007] When a probing disturbance is applied, sensors are used to capture the instantaneous response characteristics of the molten pool to the probing disturbance;

[0008] By comparing instantaneous response characteristics with physical correlation patterns, interference factors that cause weld quality fluctuations can be identified.

[0009] Based on the interference factors, the core welding parameters are compensated and adjusted.

[0010] Through this technical solution, this application can actively apply probing disturbances and compare them with the instantaneous response characteristics of the molten pool and the physical correlation pattern, thereby accurately identifying the interference factors that cause weld quality fluctuations. This overcomes the limitations of traditional online monitoring systems in accurately attributing and diagnosing the causes, and realizes precise compensation and adjustment of welding parameters, effectively solving the problem of welding quality fluctuations caused by the superposition of multiple complex factors in actual production.

[0011] Furthermore, the steps of comparing instantaneous response characteristics with physical correlation patterns to identify interfering factors causing weld quality fluctuations include:

[0012] The instantaneous response features are processed into multidimensional feature vectors to obtain multidimensional feature vectors.

[0013] Multidimensional feature vectors are matched with pre-established physical association patterns that include combinations of various interference factors in a multidimensional manner.

[0014] Based on the multi-dimensional matching results, the combination of interfering factors that cause weld quality fluctuations is identified, and then the interfering factors that cause weld quality fluctuations are derived.

[0015] Furthermore, the steps for performing multi-dimensional matching between the multi-dimensional feature vector and a pre-established physical association pattern containing multiple combinations of interference factors include:

[0016] The multidimensional feature vectors are initially matched with physical association patterns to obtain candidate patterns;

[0017] Based on the combination of interference factors corresponding to the candidate modes, the detection perturbation strategy is adjusted to obtain the adjusted detection perturbation.

[0018] Capture the instantaneous response of the molten pool to the adjusted probing disturbance and extract discriminative features;

[0019] The discriminant features are compared with the specific response fingerprints of the candidate patterns to identify combinations of interfering factors that cause fluctuations in weld quality.

[0020] Furthermore, when the identified interfering factor is a highly localized micro-inhomogeneity, the steps for compensating and adjusting the core welding parameters based on the interfering factor include:

[0021] Based on the spatial location and size information of the disturbance factors with highly localized micro-uniformity, a local compensation region is generated.

[0022] By using robot trajectory planning and microsecond-level synchronous control of laser power, the laser power can be increased within the local compensation area.

[0023] By using robot trajectory planning and microsecond-level synchronous control of laser power, the welding speed can be reduced within the local compensation area.

[0024] Furthermore, the steps to perform multidimensional feature vectorization on the instantaneous response features to obtain multidimensional feature vectors include:

[0025] Time series analysis was performed on the instantaneous response characteristics to extract the rate of change curves of the geometric features of the molten pool over time;

[0026] Calculate the slope, curvature, oscillation period, and decay constant of the rate of change curve of the molten pool geometry over time within a specific time window;

[0027] Time series analysis of instantaneous response characteristics is performed to extract the curves of spectral line intensity of specific elements over time;

[0028] Calculate the instantaneous rate of change, peak occurrence time, and duration of the spectral line intensity curve of a specific element over time;

[0029] The slope, curvature, oscillation period, and decay constant of the rate of change curve of the geometric features of the molten pool over time within a specific time window are combined with the static features of the molten pool. The instantaneous rate of change, peak occurrence time, and duration of the rate of change curve of the intensity of a specific element's spectral line over time are combined with the static features of the molten pool. All the combined results are fused to obtain a multidimensional feature vector.

[0030] Furthermore, the steps for calculating the slope, curvature, oscillation period, and decay constant of the rate of change curve of the molten pool geometry over a specific time window include:

[0031] The rate of change curve of the geometric features of the molten pool over time is piecewise fitted, and the local slope and local curvature are calculated in each segment.

[0032] Fourier transform is performed on the rate of change curve of the geometric features of the molten pool over time to extract the main oscillation frequency and the corresponding amplitude;

[0033] Wavelet analysis was performed on the rate of change curve of the geometric features of the molten pool over time to extract the attenuation constants at different scales;

[0034] By combining the local slope, local curvature, main oscillation frequency, amplitude, and decay constant at different scales, we can obtain the slope, curvature, oscillation period, and decay constant.

[0035] Furthermore, the steps of performing wavelet analysis on the rate of change curve of the molten pool geometry over time to extract the attenuation constants at different scales include:

[0036] Based on the local characteristics of the rate of change curve of the molten pool geometry over time, the wavelet basis function is dynamically selected;

[0037] By dynamically selecting wavelet basis functions, wavelet decomposition is performed on the rate of change curve of the geometric features of the molten pool over time to extract the attenuation constants at different scales.

[0038] Furthermore, based on the local characteristics of the rate of change curve of the molten pool geometry over time, the steps for dynamically selecting the wavelet basis function include:

[0039] Based on the local stationarity, local oscillation, and local sharpness of the rate of change curve of the molten pool geometry over time, select the corresponding wavelet basis function type;

[0040] When the rate of change curve of the molten pool geometry over time exhibits local stationary characteristics, the Daubechies wavelet is selected as the wavelet basis function.

[0041] When the rate of change curve of the molten pool geometry over time exhibits local oscillatory characteristics, the Coiflet wavelet is selected as the wavelet basis function.

[0042] When the rate of change curve of the molten pool geometry over time exhibits sharp local characteristics, the Symlet wavelet is selected as the wavelet basis function.

[0043] Furthermore, the steps of performing wavelet analysis on the rate of change curve of the molten pool geometry over time to extract the attenuation constants at different scales include:

[0044] Before performing wavelet analysis, the rate of change curve of the geometric features of the molten pool over time is adaptively filtered to remove high-frequency noise and outliers.

[0045] Based on the characteristics of the rate of change curve of the geometric features of the molten pool over time after filtering, the wavelet basis function and the number of decomposition layers are selected.

[0046] By selecting the wavelet basis function and the number of decomposition layers, wavelet analysis is performed on the rate of change curve of the filtered molten pool geometric features over time to extract the attenuation constants at different scales.

[0047] This application also discloses a laser welding torch welding parameter optimization system, applied to a laser welding torch welding parameter optimization method, the system comprising:

[0048] The application module selects non-critical areas in the weld and applies probing perturbations to the core welding parameters when the laser welding gun performs the welding task.

[0049] The capture module uses sensors to capture the instantaneous response characteristics of the molten pool to the probe disturbance when it is applied.

[0050] The identification module compares instantaneous response characteristics with physical correlation patterns to identify interference factors that cause fluctuations in weld quality.

[0051] The adjustment module compensates for and adjusts the core welding parameters based on interference factors.

[0052] Through modular design, it enables real-time monitoring of the welding process, intelligent identification of interference factors, and adaptive adjustment of welding parameters, providing an efficient and precise welding quality control solution for industrial production.

[0053] This application effectively solves the problems of welding quality fluctuations and fuzzy diagnosis caused by the superposition of multiple complex factors such as workpiece assembly accuracy, surface contaminants, robot trajectory deviation, laser performance attenuation, and protective airflow disturbance in actual industrial production through a closed-loop optimization mechanism of active detection, precise diagnosis, and intelligent compensation. It significantly improves the quality stability, production efficiency, and product qualification rate of laser welding.

[0054] To further understand the features and technical content of the present invention, please refer to the following detailed description and drawings of the present invention. However, the drawings provided are for reference and illustration only and are not intended to limit the present invention. Attached Figure Description

[0055] Figure 1 This is a flowchart of a laser welding gun welding parameter optimization method according to the present invention;

[0056] Figure 2 This is a schematic diagram of the structure of a laser welding torch welding parameter optimization system according to the present invention. Detailed Implementation

[0057] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can understand the advantages and effects of the present invention from the content disclosed in this specification. The present invention can be implemented or applied through other different specific embodiments, and various details in this specification can also be modified and changed based on different viewpoints and applications without departing from the spirit of the present invention. Furthermore, the accompanying drawings of the present invention are for simple illustrative purposes only and are not depictions of actual dimensions; this is stated in advance. The following embodiments will further describe the relevant technical content of the present invention in detail, but the disclosed content is not intended to limit the scope of protection of the present invention.

[0058] This embodiment provides a method and system for optimizing laser welding torch welding parameters, combined with... Figure 1 and Figure 2 As shown.

[0059] refer to Figure 1 A method for optimizing laser welding torch welding parameters, the method comprising the following steps:

[0060] When performing welding tasks with a laser welding gun, select non-critical areas in the weld and apply probing perturbations to the core welding parameters;

[0061] When a probing disturbance is applied, sensors are used to capture the instantaneous response characteristics of the molten pool to the probing disturbance;

[0062] By comparing instantaneous response characteristics with physical correlation patterns, interference factors that cause weld quality fluctuations can be identified.

[0063] Based on the interference factors, the core welding parameters are compensated and adjusted.

[0064] This application effectively solves the problem that traditional methods are difficult to accurately diagnose the root cause of welding defects in complex production environments by applying probing disturbances to non-critical areas and capturing the instantaneous response characteristics of the molten pool, thereby identifying interference factors and making compensation adjustments, and significantly improving welding quality and production efficiency.

[0065] Among them, "core welding parameters" typically refer to the parameters that have the most significant impact on weld formation and quality, such as laser power, welding speed, defocusing amount, and shielding gas flow rate. Even small fluctuations in these parameters can lead to significant changes in the molten pool behavior, thereby affecting the weld penetration, width, shape, and the generation of internal defects. "Detective perturbation" refers to a small, controllable, instantaneous change applied to the core welding parameters without significantly affecting the current weld quality. This perturbation can be periodic, random, or step-like, and its purpose is to elicit a specific response in the molten pool for capture and analysis by sensors. "Instantaneous response characteristics" refer to the observable properties exhibited by the molten pool after being subjected to a detective perturbation, resulting in changes in its physical state (such as temperature field, flow field, geometry, elemental distribution, etc.) within a very short time. These characteristics can be captured by various sensors such as high-speed cameras, infrared thermal imagers, spectrometers, and acoustic sensors. "Physical correlation patterns" refer to pre-established models or databases describing the intrinsic relationship between specific perturbation factors and the instantaneous response characteristics of the molten pool. This model can be obtained based on physical simulation, experimental data, or machine learning training, and is used to map the observed molten pool response characteristics to specific disturbance factors.

[0066] When performing welding tasks with a laser welding torch, it is first necessary to select non-critical areas in the weld seam and apply probing perturbations to the core welding parameters. The purpose of selecting non-critical areas is to perform parameter perturbation and data acquisition without affecting the final product quality. For example, the beginning or end section of a long weld seam, or areas in the product structure where strength requirements are not high, can be considered non-critical areas. There are various ways to apply probing perturbations. For example, the power output of the laser can be controlled to increase or decrease slightly over a very short period of time (e.g., tens to hundreds of microseconds), forming a power pulse or power step. Another method is to fine-tune the welding speed to produce slight acceleration or deceleration in a localized area. The focal point of the laser beam can also be moved back and forth within a small range by changing the defocusing amount. These perturbations are usually preset, and their amplitude and duration are carefully designed to ensure an effective response in the molten pool while avoiding irreversible damage to the weld seam.

[0067] When a probing disturbance is applied, sensors are used to capture the instantaneous response characteristics of the molten pool to the disturbance. The choice of sensor is crucial; it needs to be able to capture the dynamic changes of the molten pool in real time with high precision. For example, high-speed cameras can be used to record the morphological changes of the molten pool surface, spatter patterns, and the dynamic behavior of the keyhole at thousands or even tens of thousands of frames per second. Image processing techniques can be used to extract geometric features of the molten pool, such as its width, length, depth (indirectly estimated through the keyhole opening size), and oscillation frequency. Furthermore, infrared thermal imagers can be used to capture the temperature distribution and temperature gradient changes on the molten pool surface, reflecting the efficiency of heat input and conduction. Spectrometers are also effective sensors; by analyzing the spectrum emitted by the plasma or metal vapor above the molten pool, changes in the intensity of spectral lines of specific elements can be identified, thereby inferring the presence of chemical reactions, element burn-off, or contaminants within the molten pool. Acoustic sensors can capture sound wave signals generated by molten pool oscillations, bubble bursts, or spatter; these signals are also closely related to the stability of the molten pool.

[0068] This method compares instantaneous response characteristics with physical correlation patterns to identify interfering factors causing weld quality fluctuations. The physical correlation pattern is the core of this approach; it establishes a mapping relationship between the instantaneous response characteristics of the molten pool and specific interfering factors (such as workpiece gap, surface contaminants, laser power attenuation, and unstable shielding gas flow). For example, if a probing power disturbance causes a rapid increase in the molten pool width but no significant change in weld depth, this may be related to a decrease in laser absorptivity due to the oxide layer on the workpiece surface. If the molten pool exhibits severe oscillations after a disturbance, accompanied by an abnormal increase in the intensity of specific elemental spectral lines, it may indicate unstable shielding gas flow or the presence of specific contaminants on the workpiece surface. The physical correlation pattern can be a complex machine learning model, such as a support vector machine, neural network, or decision tree, trained on extensive experimental data to learn patterns in the molten pool response characteristics under different interfering factors. Alternatively, it can be an expert system based on physical principles, containing a series of rules for logical reasoning and diagnosis based on observed features.

[0069] The system compensates for and adjusts core welding parameters based on interfering factors. Once a factor causing fluctuations in weld quality is identified, the system can specifically adjust core welding parameters to counteract its effects and restore the welding process to its optimal state. For example, if laser power attenuation is detected, the system can moderately increase the laser's output power to compensate. If contaminants on the workpiece surface cause a decrease in absorptivity, the system can fine-tune the laser power or welding speed to ensure sufficient energy input. If the workpiece gap is too large, the system can adjust the laser beam defocusing or adopt an oscillating welding mode to better bridge the gap. This compensation adjustment is dynamic, real-time, and based on accurate diagnosis of interfering factors, rather than blind trial and error.

[0070] This application further proposes a step for identifying interfering factors that cause weld quality fluctuations by comparing instantaneous response characteristics with physical correlation patterns:

[0071] The instantaneous response features are processed into multidimensional feature vectors to obtain multidimensional feature vectors.

[0072] Multidimensional feature vectors are matched with pre-established physical association patterns that include combinations of various interference factors in a multidimensional manner.

[0073] Based on the multi-dimensional matching results, the combination of interfering factors that cause weld quality fluctuations is identified, and then the interfering factors that cause weld quality fluctuations are derived.

[0074] Specifically, transient response characteristics refer to the instantaneous response data of the molten pool to a probed disturbance, captured by sensors. This data may include various forms, such as optical, acoustic, and thermal signals. To enable unified and effective analysis of this heterogeneous data, multidimensional feature vectorization is required. Multidimensional feature vectorization involves transforming the raw, complex transient response characteristic data into a vector with fixed dimensions and numerical representation through a series of signal processing, feature extraction, and data fusion techniques. Each dimension of this vector represents a specific attribute or combination of attributes of the transient response characteristic, such as the rate of change of the molten pool's geometric dimensions, temperature gradient, or elemental spectral line intensity, thereby comprehensively and quantitatively describing the dynamic behavior of the molten pool.

[0075] Among them, physical correlation patterns are pre-established knowledge bases or models containing various combinations of disturbance factors. These patterns are constructed through extensive experimental data, physical modeling, or expert experience to describe the unique "fingerprints" or patterns generated by specific disturbance factors (such as material inhomogeneity, insufficient gas protection, laser power fluctuations, etc.) in the instantaneous response characteristics of the molten pool. Each disturbance factor or combination thereof corresponds to one or more specific physical correlation patterns.

[0076] In practical applications, multi-dimensional matching refers to comparing and analyzing a vectorized multi-dimensional feature vector with various patterns in a physical correlation pattern library. This matching can be implemented using various algorithms, such as distance-based matching, machine learning-based classifiers, and pattern recognition algorithms. Through multi-dimensional matching, it is possible to evaluate which preset combination of interference factors best matches the instantaneous response characteristics of the current melt pool.

[0077] Therefore, based on the multi-dimensional matching results, combinations of interfering factors causing weld quality fluctuations can be identified. For example, if the matching results show that the current molten pool response highly matches the combination pattern of "local material inclusions" and "shielding gas flow fluctuations," then these two factors can be identified as interfering factors causing the current weld quality fluctuations. Furthermore, by analyzing these combinations of factors, specific interfering factors causing weld quality fluctuations can be derived, providing a precise basis for subsequent parameter compensation adjustments.

[0078] The proposed solution quantifies and standardizes the complex dynamic behavior of the molten pool by structurally vectorizing the original instantaneous response features into multidimensional features. This standardized representation enables subsequent multidimensional matching with pre-established physical correlation patterns. Through this refined matching process, specific interfering factors and their combinations that cause weld quality fluctuations can be accurately identified from numerous potential interference factors. This mechanism avoids the fuzzy judgments or experience-based dependencies that may exist in traditional methods, significantly improving the accuracy and efficiency of interfering factor identification, and providing a solid data foundation and decision-making basis for subsequent parameter compensation adjustments.

[0079] This application further proposes a multi-dimensional matching process for multi-dimensional feature vectors with pre-established physical association patterns containing multiple combinations of interference factors, including:

[0080] The multidimensional feature vectors are initially matched with physical association patterns to obtain candidate patterns;

[0081] Based on the combination of interference factors corresponding to the candidate modes, the detection perturbation strategy is adjusted to obtain the adjusted detection perturbation.

[0082] Capture the instantaneous response of the molten pool to the adjusted probing disturbance and extract discriminative features;

[0083] The discriminant features are compared with the specific response fingerprints of the candidate patterns to identify combinations of interfering factors that cause fluctuations in weld quality.

[0084] Specifically, the initial matching of multidimensional feature vectors with physical correlation patterns to obtain candidate patterns refers to the process of initially comparing the multidimensional feature vectors obtained by multidimensional feature vectorization of the instantaneous response features of the melt pool to probing disturbances with a pre-established physical correlation pattern library. This physical correlation pattern library contains various known combinations of interference factors and their corresponding melt pool response feature fingerprints. The purpose of the initial matching is to filter out several patterns with high similarity to the current multidimensional feature vector from the large pattern library. The combinations of interference factors represented by these patterns are the candidate patterns. For example, methods such as cosine similarity, Euclidean distance, or machine learning classifiers can be used for initial matching to quickly narrow down the range of potential interference factors.

[0085] In this process, the probing perturbation strategy is adjusted based on the combination of interfering factors corresponding to the candidate modes, resulting in the adjusted probing perturbation. This can be understood as designing or selecting a specific probing perturbation for each candidate mode obtained from the initial matching. This adjusted probing perturbation aims to amplify or highlight the specificity of the interfering factors represented by the candidate mode in the molten pool response, making it easier for subsequent sensors to capture. For example, if a candidate mode indicates the possible presence of a specific type of porosity defect, the perturbation frequency or amplitude of the laser power or welding speed can be adjusted to generate a more pronounced transient response in the molten pool related to that porosity formation mechanism.

[0086] In practical applications, capturing the instantaneous response of the molten pool to an adjusted probing perturbation and extracting discriminative features involves, specifically, monitoring the dynamic behavior of the molten pool in real time using a sensor system (such as a high-speed camera, spectrometer, or acoustic emission sensor) after applying the adjusted probing perturbation. The captured instantaneous response will then be more targeted, reflecting the unique behavior of specific perturbation factors under specific disturbances. The discriminative features extracted from these instantaneous responses are optimized and screened, effectively distinguishing key indicators of different candidate modes. For example, specific frequency components of the molten pool oscillation mode, intensity trends of specific elemental spectral lines, or minute fluctuations in the molten pool surface morphology can be extracted.

[0087] Furthermore, a secondary comparison is performed between the discriminant features and the specific response fingerprints of the candidate patterns to identify the combinations of interfering factors causing weld quality fluctuations. This involves a refined comparison of the extracted discriminant features with the pre-stored specific response fingerprints of each candidate pattern. A specific response fingerprint is a unique molten pool response pattern exhibited by a specific combination of interfering factors under a specific adjustment perturbation strategy. This secondary comparison eliminates ambiguities in the initial matching, accurately identifies the combination of interfering factors that best matches the current molten pool behavior, and ultimately determines the specific interfering factors causing weld quality fluctuations.

[0088] This application's solution effectively addresses the ambiguity and uncertainty inherent in single-dimensional multi-dimensional matching for identifying interfering factors in complex environments by introducing a two-stage matching and verification mechanism. First, preliminary matching quickly filters out potential combinations of interfering factors, forming a candidate pattern set, thus avoiding the enormous computational overhead of performing high-precision matching on all possible patterns. Second, for these candidate patterns, the system intelligently adjusts its probing perturbation strategy, making the melt pool's response to specific interfering factors more significant and discriminative. It is precisely this adaptive perturbation adjustment that allows for the extraction of more discriminative features from subsequently captured instantaneous responses. Finally, by performing a secondary comparison of these discriminative features with the specific response fingerprints of the candidate patterns, the preliminary matching results can be refined and confirmed, significantly improving the accuracy and reliability of interfering factor identification. This iterative and verification mechanism enables the system to more accurately pinpoint the root cause of problems when faced with similar or complex melt pool responses.

[0089] In some preferred embodiments, it is assumed that during laser welding, preliminary matching results show two candidate modes: Mode A (possibly caused by microscopic inclusions within the material) and Mode B (possibly caused by fluctuations in the shielding gas flow rate). These two factors may exhibit similarities in certain instantaneous response characteristics. In this case, the system adjusts the probing perturbation strategy according to the characteristics of Mode A and Mode B respectively. For example, for Mode A, a laser power perturbation of a specific frequency and amplitude can be applied to excite a specific oscillation mode in the molten pool around the inclusions; while for Mode B, a shielding gas flow rate perturbation of a specific frequency and amplitude can be applied to generate specific fluctuations or elemental spectral changes on the molten pool surface. Subsequently, the sensor captures the instantaneous response of the molten pool to these two adjusted probing perturbations and extracts their respective discriminant features. For example, for Mode A, the discriminant feature might be the acoustic emission signal intensity of a specific region at the bottom of the molten pool; for Mode B, the discriminant feature might be the rate of change in the intensity of the evaporation spectral lines of specific elements on the molten pool surface. Finally, these discriminant features are compared a second time with the specific response fingerprints of Mode A and Mode B respectively. If the actual response of the molten pool closely matches the specific response fingerprint of Mode B, but has a lower degree of similarity to Mode A, the system will ultimately identify the shielding gas flow fluctuation as the interfering factor causing weld quality fluctuations. This iterative and verification process ensures that the true interfering factors can be accurately and reliably identified in complex and ever-changing environments.

[0090] This application further proposes a step for compensating and adjusting the core welding parameters based on the identified interfering factors when the identified interfering factors are highly localized micro-inhomogeneities.

[0091] Based on the spatial location and size information of the disturbance factors with highly localized micro-uniformity, a local compensation region is generated.

[0092] By using robot trajectory planning and microsecond-level synchronous control of laser power, the laser power can be increased within the local compensation area.

[0093] By using robot trajectory planning and microsecond-level synchronous control of laser power, the welding speed can be reduced within the local compensation area.

[0094] Specifically, highly localized micro-inhomogeneities refer to small-sized and unevenly distributed defects or abnormal regions appearing inside or on the surface of the weld, such as micron-sized pores, inclusions, unfused areas, or localized grain anomalies. These inhomogeneities can significantly affect the mechanical properties and fatigue life of the weld. The spatial location and size information of the interfering factors can be obtained through multi-sensor data fusion techniques. For example, refined analysis can be performed using instantaneous response characteristics captured by high-speed cameras, spectrometers, or acoustic emission sensors to accurately locate the center coordinates, boundary contours, and dimensions of the defect in the welding direction and transverse direction. Therefore, the local compensation region refers to a specific length and width area defined on the weld path surrounding the identified highly localized micro-inhomogeneous defect. The generation of this region aims to ensure that compensation measures can accurately act on the defect itself and its affected surrounding area, avoiding unnecessary interference with other normal areas. In practical applications, robot trajectory planning refers to the precise pre-setting or real-time adjustment of the laser welding torch's movement path to ensure that the laser focus accurately covers the local compensation region. Microsecond-level synchronous control of laser power refers to the ability to rapidly and precisely adjust the laser output power within an extremely short timescale (e.g., at the microsecond level), achieving a high degree of synchronization with the robot's trajectory movement. This synchronous control ensures that the laser power can be instantly increased when the welding torch passes through a local compensation area and quickly return to normal levels when leaving that area. Within the local compensation area, increasing the laser power aims to increase the energy input in that area, promoting the full melting and resolidification of the material in the defective region, thereby effectively eliminating defects such as porosity and poor fusion, and improving the local metallurgical quality. Simultaneously, reducing the welding speed within the local compensation area aims to prolong the interaction time between the laser and the material, allowing more energy to be absorbed by the defective region and providing more sufficient escape time for defects (such as bubbles) within the molten pool, further improving the defect repair effect and weld density.

[0095] This application's solution generates a locally compensated region that precisely matches the defect area by accurately spatially locating and sizing the identified highly localized micro-inhomogeneous interference factors. This refined region division allows subsequent compensation measures to be highly focused on the defect itself, avoiding the potential side effects of traditional large-scale compensation. Furthermore, through microsecond-level synchronous control of robot trajectory planning and laser power, instantaneous increase in laser power and synchronous reduction in welding speed are achieved within the locally compensated region. This collaborative control mechanism ensures higher energy density and longer treatment time in the defect area, thereby promoting complete melting of the material, effective gas escape, and more uniform resolidification, effectively repairing highly localized micro-inhomogeneous defects.

[0096] In some preferred embodiments, it is assumed that during laser welding, the instantaneous response characteristics of the molten pool captured by sensors identify a micropore with a diameter of approximately 200 micrometers, centered at a specific location on the weld path. Based on the spatial location and size of this micropore, the system automatically generates a local compensation zone centered on the micropore, with a length of 1 mm and a width of 0.5 mm. When the laser welding torch moves to the starting point of this local compensation zone, the robot trajectory planning system synchronizes with the laser power control system at the microsecond level. Specifically, the laser power is increased from 3 kW (normal welding) to 3.5 kW within 10 microseconds, while the welding speed is reduced from 50 mm / s to 40 mm / s. The welding torch passes through the local compensation zone with adjusted parameters, ensuring that the micropore area receives additional energy input and a longer melting time. When the welding torch leaves the zone, the laser power and welding speed quickly return to normal levels. Through this precise local compensation, the micropore can be effectively eliminated, thereby significantly improving the local quality and overall reliability of the weld.

[0097] The steps to perform multidimensional feature vectorization on the instantaneous response features to obtain multidimensional feature vectors include:

[0098] Time series analysis was performed on the instantaneous response characteristics to extract the rate of change curves of the geometric features of the molten pool over time;

[0099] Calculate the slope, curvature, oscillation period, and decay constant of the rate of change curve of the molten pool geometry over time within a specific time window;

[0100] Time series analysis of instantaneous response characteristics is performed to extract the curves of spectral line intensity of specific elements over time;

[0101] Calculate the instantaneous rate of change, peak occurrence time, and duration of the spectral line intensity curve of a specific element over time;

[0102] The slope, curvature, oscillation period, and decay constant of the rate of change curve of the geometric features of the molten pool over time within a specific time window are combined with the static features of the molten pool. The instantaneous rate of change, peak occurrence time, and duration of the rate of change curve of the intensity of a specific element's spectral line over time are combined with the static features of the molten pool. All the combined results are fused to obtain a multidimensional feature vector.

[0103] Instantaneous response characteristics refer to the dynamic behavior data of the molten pool captured by sensors when a probing disturbance is applied. This data may include, but is not limited to, the surface morphology, size changes, temperature distribution, and emission spectra of specific elements within the molten pool. Time-series analysis of instantaneous response characteristics aims to reveal the evolutionary patterns of the molten pool's dynamic behavior. The rate of change curve of the molten pool's geometric characteristics over time can reflect the macroscopic dynamic characteristics of the molten pool under disturbance, such as expansion, contraction, and oscillation. For example, high-speed cameras or laser displacement sensors can be used to acquire molten pool surface contour data, and then the rate of change of its geometric parameters such as area, depth, and width over time can be calculated. The rate of change curve of the intensity of specific element spectral lines over time can reflect the microscopic dynamic changes in the internal material composition and temperature field of the molten pool. For example, the intensity of characteristic spectral lines of key elements such as iron, aluminum, and oxygen can be captured by a spectrometer to monitor metallurgical reactions or gas inclusions within the molten pool.

[0104] Furthermore, the slope, curvature, oscillation period, and decay constant of the rate of change curves of the molten pool's geometric characteristics over time are calculated within a specific time window to quantify the characteristics of the molten pool's macroscopic dynamic behavior. The slope characterizes the rate of change of the molten pool, the curvature reflects the steepness of the change trend, and the oscillation period and decay constant reveal the frequency and stability of the molten pool's oscillating behavior. Similarly, the instantaneous rate of change, peak occurrence time, and duration of the rate of change curves of the intensity of a specific element's spectral line over time are calculated to quantify the characteristics of the molten pool's microscopic dynamic behavior. The instantaneous rate of change reflects the intensity of the elemental reaction, the peak occurrence time indicates the timing of a specific event (such as bubble formation or collapse), and the duration measures the scope of the event's influence.

[0105] Finally, the extracted dynamic features are combined with the static features of the molten pool. The static features of the molten pool can include steady-state parameters such as initial size, initial temperature, and material composition before the disturbance is applied. By combining the dynamic and static features, a comprehensive multidimensional feature vector can be formed. This vector can more completely describe the instantaneous response of the molten pool under probing disturbances, thus providing a richer and more accurate data foundation for subsequent identification of disturbance factors. The fusion of all combined results can be achieved using various methods such as feature concatenation, weighted averaging, or machine learning to ensure that the multidimensional feature vector can comprehensively and effectively characterize the instantaneous response of the molten pool.

[0106] This application's approach, through multidimensional feature vectorization of instantaneous response characteristics, enables a more comprehensive and refined capture of the dynamic behavior of the molten pool under probing perturbations. Specifically, through time series analysis, the dynamic response characteristics of the molten pool can be extracted from two dimensions: macroscopic geometric changes and microscopic elemental spectral line intensity changes. The rate of change curves of the molten pool's geometric characteristics over time, along with their quantification parameters (slope, curvature, oscillation period, decay constant), reflect the overall deformation, flow, and stability of the molten pool. Meanwhile, the rate of change curves of specific elemental spectral line intensities over time, along with their quantification parameters (instantaneous rate of change, peak occurrence time, duration), reveal microscopic processes within the molten pool, such as metallurgical reactions, bubble behavior, or impurity distribution. Combining and fusing these dynamic characteristics with the static characteristics of the molten pool results in a multidimensional feature vector that not only contains the instantaneous dynamic information of the molten pool under perturbation but also incorporates its initial state information, thus constructing a more complete and robust molten pool response fingerprint. This refined feature extraction and vectorization process provides richer and more accurate input for subsequent comparison of instantaneous response features with physical correlation patterns to identify interference factors that cause weld quality fluctuations, significantly improving the accuracy and reliability of interference factor identification.

[0107] The above technical solution enables refined and multi-dimensional quantification of the instantaneous response characteristics of the molten pool. This multi-dimensional feature vectorization overcomes the limitations of single or coarse feature descriptions, allowing for a more comprehensive and accurate characterization of the dynamic behavior of the molten pool. Consequently, when comparing instantaneous response characteristics with physical correlation patterns, richer and more discriminative information is provided, significantly improving the accuracy and robustness of identifying interfering factors that cause weld quality fluctuations. For example, by simultaneously analyzing the macroscopic geometric changes and microscopic elemental spectral line intensity changes of the molten pool, similar molten pool responses caused by different interfering factors such as material inhomogeneity, insufficient gas protection, or laser energy fluctuations can be more effectively distinguished, providing a more precise basis for subsequent parameter compensation adjustments and ultimately improving the overall quality and stability of the weld.

[0108] The steps for calculating the slope, curvature, oscillation period, and decay constant of the rate of change curve of the molten pool geometry over a specific time window include:

[0109] The rate of change curve of the geometric features of the molten pool over time is piecewise fitted, and the local slope and local curvature are calculated in each segment.

[0110] Fourier transform is performed on the rate of change curve of the geometric features of the molten pool over time to extract the main oscillation frequency and the corresponding amplitude;

[0111] Wavelet analysis was performed on the rate of change curve of the geometric features of the molten pool over time to extract the attenuation constants at different scales;

[0112] By combining the local slope, local curvature, main oscillation frequency, amplitude, and decay constant at different scales, we can obtain the slope, curvature, oscillation period, and decay constant.

[0113] Piecewise fitting of the rate of change curve of the molten pool geometry over time involves dividing the entire time series curve into several smaller, relatively stable, or trend-specific sub-intervals, and independently fitting the curve within each sub-interval. For example, polynomial fitting or spline fitting methods can be used. Piecewise fitting allows for more accurate capture of the local variation characteristics of the curve, enabling the calculation of local slopes and curvatures. These parameters reflect the instantaneous change trend and degree of curvature of the molten pool geometry over different time periods.

[0114] Furthermore, a Fourier transform is performed on the rate of change curve of the molten pool geometry over time. The purpose is to convert the time-domain signal to the frequency domain to reveal the periodic components contained in the signal. Through the Fourier transform, the main oscillation frequency and the corresponding amplitude of the signal can be extracted. These parameters are crucial for understanding the periodic fluctuation behavior of the molten pool, such as molten pool oscillation and keyhole fluctuation.

[0115] Furthermore, wavelet analysis was performed on the rate of change curves of the molten pool's geometric characteristics over time, aiming to conduct time-frequency localization analysis of non-stationary signals. Wavelet analysis can provide frequency information of the signal at different time scales, thereby effectively extracting the attenuation constant at different scales. The attenuation constant can characterize the speed at which the molten pool recovers to a stable state after a disturbance or the characteristics of energy dissipation, which is of great significance for evaluating the stability of the molten pool.

[0116] Finally, the local slope, local curvature, main oscillation frequency, amplitude, and decay constants at different scales are combined to form a comprehensive and multi-dimensional feature set. This combination comprehensively reflects the instantaneous trend, curvature, periodic fluctuations, and stability decay characteristics of the molten pool's geometric features over time, providing a richer and more accurate data foundation for subsequent identification of interference factors.

[0117] This application employs multiple signal processing techniques, including piecewise fitting, Fourier transform, and wavelet analysis, to refine the analysis of the rate of change curve of the molten pool's geometric characteristics over time. Specifically, piecewise fitting effectively addresses the local nonlinear characteristics of the molten pool's dynamic response, avoiding errors that might arise from a single global fitting, thus more accurately capturing local slopes and curvatures and reflecting subtle differences in the instantaneous morphological changes of the molten pool. Fourier transform focuses on extracting the periodic components of the signal, quantifying the inherent oscillation frequency and amplitude of the molten pool, which is crucial for identifying disturbances related to periodic perturbations (such as airflow fluctuations and laser power pulsations). Wavelet analysis, with its excellent time-frequency localization capabilities, effectively handles the non-stationary characteristics of the molten pool response, extracting attenuation constants at different time scales, thereby revealing the energy dissipation and stability recovery process of the molten pool after perturbation. This has unique advantages for evaluating non-periodic, transient disturbances such as internal material defects or thermal stress accumulation. By combining these multi-dimensional and multi-scale features, the proposed solution overcomes the limitations of single analysis methods in dealing with complex molten pool dynamics, ensuring that the extracted parameters can comprehensively and accurately characterize the instantaneous state of the molten pool, thus laying a solid foundation for the accurate identification of subsequent interference factors.

[0118] Through the above technical solutions, this application can significantly improve the accuracy and robustness of extracting geometric feature parameters of the molten pool. Piecewise fitting allows for the precise capture of local variation trends, avoiding information loss that may result from global fitting; Fourier transform ensures the effective identification of periodic dynamics, providing a quantitative basis for judging oscillation-related interference factors; wavelet analysis demonstrates excellent performance in processing non-stationary signals, revealing attenuation characteristics at different scales, thereby providing a more comprehensive assessment of the stability of the molten pool. These refined, multi-dimensional parameter extraction methods enable a more comprehensive and in-depth characterization of the instantaneous response features of the molten pool, greatly enhancing the system's ability to identify interference factors that cause weld quality fluctuations. Especially when facing complex and variable welding environments and microscopic defects, it can provide more reliable diagnostic information, thereby achieving more precise welding parameter compensation and adjustment, ultimately improving the overall quality and stability of the weld.

[0119] In some preferred embodiments, it is assumed that during the welding task performed by the laser welding torch, a sensor captures the rate of change curve of the molten pool geometry over time. To accurately calculate the slope, curvature, oscillation period, and attenuation constant of this curve within a specific time window, firstly, the curve is divided into multiple time segments, for example, every 50 milliseconds. Within each segment, a quadratic polynomial is fitted to calculate the local slope and local curvature. Secondly, a Fast Fourier Transform is performed on the rate of change curve of the molten pool geometry over time throughout the entire time window. By analyzing the spectrum, the dominant oscillation frequency with the highest energy concentration is identified, and its corresponding amplitude is calculated. For example, if a dominant oscillation frequency of 200 Hz is detected, it is recorded. Thirdly, wavelet decomposition is performed on the curve, for example, using the Daubechies wavelet basis function for a three-level decomposition. By analyzing the detail coefficients and approximation coefficients at different scales, the attenuation constant characterizing the signal energy attenuation is extracted. For example, in the first-level decomposition, a constant reflecting rapid attenuation may be extracted, while in the third-level decomposition, a constant reflecting slow attenuation may be extracted. Finally, the local slopes and curvatures obtained through piecewise fitting, the master oscillation frequencies and amplitudes obtained through Fourier transform, and the attenuation constants at different scales obtained through wavelet analysis are integrated. For example, a feature vector can be constructed that includes the average local slope, maximum local curvature, master oscillation frequency, amplitude, and multiple attenuation constants. This integrated feature vector is then compared with a pre-established physical correlation pattern to identify interfering factors that cause weld quality fluctuations. This multi-technology fusion computational approach enables a more comprehensive and accurate characterization of the molten pool's dynamic behavior, thereby improving the accuracy of interfering factor identification.

[0120] This application further proposes a step for extracting the attenuation constant at different scales by performing wavelet analysis on the rate of change curve of the molten pool geometry over time:

[0121] Based on the local characteristics of the rate of change curve of the molten pool geometry over time, the wavelet basis function is dynamically selected;

[0122] By dynamically selecting wavelet basis functions, wavelet decomposition is performed on the rate of change curve of the geometric features of the molten pool over time to extract the attenuation constants at different scales.

[0123] Specifically, dynamic selection of wavelet basis functions, based on the local characteristics of the rate of change curve of the molten pool geometry over time, refers to the system analyzing and evaluating the local characteristics of the rate of change curve of the molten pool geometry over time before wavelet decomposition. These local characteristics may include, but are not limited to, the smoothness of the curve, oscillation frequency, sharpness, or the presence of abrupt changes. Based on these analysis results, the system intelligently selects the wavelet basis function most suitable for the current local characteristics. For example, for curve segments with strong smoothness, the Daubechies wavelet with good smoothness can be selected; for curve segments with strong oscillation, the Coiflet wavelet with good frequency localization can be selected; and for curve segments with sharp changes or abrupt changes, the Symlet wavelet with good time localization can be selected. The purpose of this dynamic selection is to ensure that the selected wavelet basis function can match the local structure of the signal to the greatest extent, thereby optimizing the decomposition effect.

[0124] The process of using dynamically selected wavelet basis functions to perform wavelet decomposition on the rate-of-change curve of the molten pool's geometric features over time, and extracting attenuation constants at different scales, involves determining the most suitable wavelet basis function for the current local characteristics. This function is then applied to the rate-of-change curve of the molten pool's geometric features over time for multi-scale wavelet decomposition. Wavelet decomposition breaks down the original signal into sub-bands of different frequencies (scales), each representing information about the signal within a specific frequency range. From these decomposed sub-bands, attenuation constants at different scales can be extracted. The attenuation constant is a parameter describing the rate of decay of signal energy or amplitude over time or scale, reflecting the dynamic characteristics of the molten pool's transient response. This method yields more refined and accurate attenuation constants, providing more reliable data support for subsequent identification of interference factors.

[0125] This application's solution effectively addresses the limitations of traditional fixed wavelet basis functions in processing complex and variable transient response signals of molten pools by introducing a step of "dynamically selecting wavelet basis functions based on the local characteristics of the rate of change curve of the molten pool's geometric features over time." When the rate of change curve of the molten pool's geometric features over time exhibits different local characteristics, such as stationary, oscillating, or sharp changes, a single fixed wavelet basis function cannot efficiently capture all these characteristics simultaneously, potentially leading to distorted decomposition results and affecting the accurate extraction of the attenuation constant. By dynamically selecting the wavelet basis function, a high degree of matching between the selected basis function and the signal's local characteristics can be ensured, enabling wavelet decomposition to more accurately capture the signal's details and transient behavior. This adaptive analysis method allows the attenuation constants extracted from different scales to more realistically reflect the physical processes of the molten pool, providing a more reliable and refined data foundation for subsequent identification of interference factors.

[0126] In some preferred embodiments, when performing wavelet analysis on the rate of change curve of the molten pool geometry over time, the system first performs local characteristic analysis on the curve. For example, if the curve exhibits a relatively stable and slowly changing trend over a certain time period, the system may dynamically select the Daubechies wavelet as the wavelet basis function because the Daubechies wavelet has good energy concentration when processing stationary signals. If the curve exhibits obvious periodic oscillations over another time period, the system may select the Coiflet wavelet, which has good localization characteristics in the frequency domain and is suitable for capturing oscillation modes. When sudden peaks or breaks appear in the curve, the Symlet wavelet may be selected because it provides a good balance between time and frequency domains and can effectively capture signal discontinuities. Through this dynamic selection, it is ensured that each local region is decomposed using the wavelet basis function most suitable for its characteristics, thereby extracting more accurate attenuation constants at different scales.

[0127] Based on the local characteristics of the rate of change curve of the molten pool geometry over time, the steps for dynamically selecting wavelet basis functions include:

[0128] Based on the local stationarity, local oscillation, and local sharpness of the rate of change curve of the molten pool geometry over time, select the corresponding wavelet basis function type;

[0129] When the rate of change curve of the molten pool geometry over time exhibits local stationary characteristics, the Daubechies wavelet is selected as the wavelet basis function.

[0130] When the rate of change curve of the molten pool geometry over time exhibits local oscillatory characteristics, the Coiflet wavelet is selected as the wavelet basis function.

[0131] When the rate of change curve of the molten pool geometry over time exhibits sharp local characteristics, the Symlet wavelet is selected as the wavelet basis function.

[0132] The local characteristics of the rate of change curve of the molten pool geometry over time refer to the signal behavior pattern exhibited by the curve in different time periods. These local characteristics can be summarized as local stationarity, local oscillation, and local sharpness. Local stationarity means that the curve changes slowly and has a stable trend within a certain time window; local oscillation means that the curve exhibits periodic or quasi-periodic fluctuations within a certain time window; local sharpness means that the curve has abrupt changes or instantaneous shocks within a certain time window.

[0133] In practical applications, the Daubechies wavelet is a type of discrete wavelet with tight support and orthogonality. Its characteristics include good smoothness, effectively capturing the stationary or slowly changing parts of a signal while suppressing noise. Therefore, when the rate of change curve of the molten pool geometry over time exhibits local stationary characteristics, choosing the Daubechies wavelet as the wavelet basis function allows for more accurate analysis of its stationary trend. The Coiflet wavelet is another type of wavelet with tight support and higher-order vanishing moments. Its characteristics include good symmetry and regularity, making it particularly suitable for analyzing signals with oscillating characteristics. Its waveform can better match periodic or quasi-periodic signals, thus capturing oscillating components more accurately during decomposition. Therefore, when the rate of change curve of the molten pool geometry over time exhibits local oscillating characteristics, choosing the Coiflet wavelet as the wavelet basis function allows for more effective extraction of information such as oscillation period and amplitude. The Symlet wavelet is an improved form of the Daubechies wavelet. While maintaining tight support and orthogonality, it adds symmetry, making it superior in processing sharp features of signals. Symlet wavelets are better able to locate abrupt changes or transient shocks in signals, reducing phase distortion. Therefore, when the rate of change curve of the molten pool geometry over time exhibits sharp local characteristics, choosing Symlet wavelets as wavelet basis functions can more accurately capture these transient changes and avoid information loss.

[0134] This application's solution addresses the problem in traditional wavelet analysis where a single wavelet basis function is insufficient to adapt to diverse local signal behaviors by meticulously analyzing the local characteristics of the molten pool's geometric features over time and dynamically matching the most suitable wavelet basis function based on these characteristics. Specifically, when the curve exhibits local stationary characteristics, the smoothness of the Daubechies wavelet effectively filters out high-frequency noise and accurately captures the stationary trend; when the curve exhibits local oscillatory characteristics, the symmetry of the Coiflet wavelet allows for accurate decomposition of the oscillatory components, avoiding phase distortion; and when the curve exhibits local sharp characteristics, the improved symmetry of the Symlet wavelet enables precise location and analysis of instantaneous changes, ensuring the integrity of key information. It is precisely this adaptive wavelet basis function selection mechanism based on local signal characteristics that allows the wavelet decomposition process to more accurately adapt to the complex dynamic changes of the molten pool's instantaneous response, thereby improving the accuracy and robustness of attenuation constant extraction.

[0135] Through the above technical solution, this application can intelligently select the best-matching wavelet basis function based on the actual local behavior of the rate of change curve of the molten pool's geometric characteristics over time. This adaptive selection mechanism significantly improves the accuracy and robustness of wavelet analysis in signal processing of complex welding processes. Compared with the method of using fixed wavelet basis functions, this scheme can more accurately capture different local details such as stability, oscillation, and sharpness in the instantaneous response characteristics of the molten pool, thereby enabling the extracted attenuation constant to more realistically reflect the physical characteristics of the molten pool. As a result, the identification of interference factors will be more accurate, providing a more reliable data foundation for subsequent compensation and adjustment of core welding parameters, ultimately helping to improve the stability and consistency of weld quality.

[0136] In some preferred embodiments, it is assumed that during the welding task performed by the laser welding torch, the rate of change curve of the molten pool geometry captured by the sensor over time exhibits different local characteristics in different time periods. Specifically, in time period t1-t2, the curve shows a relatively stable trend, possibly corresponding to the slow expansion of the molten pool during the stable heating phase; in time period t2-t3, the curve exhibits obvious periodic oscillations, which may be related to periodic fluctuations caused by Marangoni convection or bubble dynamics within the molten pool; while in time period t3-t4, the curve shows a sharp rise or fall, which may indicate instantaneous evaporation, spatter, or external disturbances within the molten pool. According to the scheme of this application, in time period t1-t2, due to the local stationary characteristics of the curve, the system will automatically select the Daubechies wavelet as the wavelet basis function for decomposition to accurately capture its stationary trend and effectively suppress noise. In time period t2-t3, given the local oscillation characteristics of the curve, the system will select the Coiflet wavelet for analysis, thereby more accurately extracting the frequency, amplitude, and phase information of the oscillations. Within the time interval t3-t4, to address the local sharpness of the curve, the system will select Symlet wavelet decomposition to accurately locate and analyze the instantaneous change, ensuring the integrity of the abrupt change information. This dynamic and adaptive wavelet basis function selection ensures that the analysis of the molten pool's instantaneous response characteristics maintains optimal matching and accuracy throughout the entire welding process, thus providing high-quality data support for subsequent interference factor identification and parameter optimization.

[0137] This application further proposes a step for extracting the attenuation constant at different scales by performing wavelet analysis on the rate of change curve of the molten pool geometry over time:

[0138] Before performing wavelet analysis, the rate of change curve of the geometric features of the molten pool over time is adaptively filtered to remove high-frequency noise and outliers.

[0139] Based on the characteristics of the rate of change curve of the geometric features of the molten pool over time after filtering, the wavelet basis function and the number of decomposition layers are selected.

[0140] By selecting the wavelet basis function and the number of decomposition layers, wavelet analysis is performed on the rate of change curve of the filtered molten pool geometric features over time to extract the attenuation constants at different scales.

[0141] Specifically, adaptive filtering refers to dynamically adjusting the filter parameters based on the local statistical characteristics of the rate of change curve of the molten pool geometry over time, in order to effectively remove noise while preserving the most useful information of the signal. For example, methods such as Kalman filtering, wavelet thresholding, or empirical mode decomposition can be used. The purpose of removing high-frequency noise and outliers is to purify the signal and prevent these interfering components from misleading the wavelet analysis results.

[0142] The selection of wavelet basis functions and decomposition levels based on the characteristics of the rate of change curve of the filtered molten pool geometry over time can be understood as choosing the most suitable wavelet basis function and determining the appropriate number of wavelet decomposition levels based on the curve's local stationarity, local oscillation, and local sharpness, to ensure effective capture of the signal's attenuation characteristics at different scales. For example, when the curve exhibits local stationarity, the Daubechies wavelet can be chosen as the wavelet basis function; when the curve exhibits local oscillation, the Coiflet wavelet can be chosen; and when the curve exhibits local sharpness, the Symlet wavelet can be chosen. The selection of the number of decomposition levels is usually based on the signal's frequency range and the required level of analysis precision.

[0143] In practical applications, wavelet analysis is performed on the rate of change curve of the filtered molten pool geometric features over time by selecting wavelet basis functions and decomposition levels, and the attenuation constants at different scales are extracted. The purpose is to accurately quantify the attenuation behavior of the molten pool transient response while minimizing noise interference, so as to provide accurate input for subsequent feature vectorization.

[0144] The proposed solution effectively suppresses high-frequency noise and outliers in the rate-of-change curve of the molten pool's geometric features over time by introducing adaptive filtering before wavelet analysis, ensuring the purity of the input signal. Based on this, the wavelet basis function and decomposition level are dynamically selected according to the characteristics of the filtered curve, enabling wavelet analysis to better match the local features of the signal and thus more accurately capture the attenuation constant at different scales. It is precisely because of the high-quality preprocessing of the input signal and the refined configuration of the analysis tools that the extracted attenuation constant is more representative and reliable, providing a solid data foundation for subsequent multidimensional feature vectorization and interference factor identification, significantly improving the accuracy and robustness of the entire welding parameter optimization method.

[0145] In some preferred embodiments, when a laser welding torch performs a welding task, the rate of change curve of the molten pool geometry captured by the sensor over time may contain high-frequency noise due to environmental vibration or electromagnetic interference. In this case, this application first performs adaptive Kalman filtering on the curve to smooth it and remove random noise. After filtering, if the curve exhibits obvious oscillatory characteristics locally, the system dynamically selects a Coiflet wavelet as the wavelet basis function and determines an appropriate number of decomposition levels based on the signal complexity, for example, 3 or 4 levels. Subsequently, wavelet decomposition is performed on the filtered curve using the selected Coiflet wavelet and the number of decomposition levels to accurately extract the attenuation constants at different scales. For example, a fast attenuation constant corresponding to the molten pool oscillation mode and a slow attenuation constant corresponding to the thermal diffusion process can be extracted. These precise attenuation constants will be used to construct a multidimensional feature vector, thereby more accurately identifying interfering factors that cause weld quality fluctuations, such as microscopic defects within the material or uneven local heat input.

[0146] refer to Figure 2 This application proposes a laser welding torch welding parameter optimization system, applied to a laser welding torch welding parameter optimization method. The system includes:

[0147] The application module selects non-critical areas in the weld and applies probing perturbations to the core welding parameters when the laser welding gun performs the welding task.

[0148] The capture module uses sensors to capture the instantaneous response characteristics of the molten pool to the probe disturbance when it is applied.

[0149] The identification module compares instantaneous response characteristics with physical correlation patterns to identify interference factors that cause fluctuations in weld quality.

[0150] The adjustment module compensates for and adjusts the core welding parameters based on interference factors.

[0151] The application module can be understood as a control unit used to generate and apply probing disturbances. Specifically, the application module can be a programmable logic controller or an embedded microcontroller, configured to apply minute, controllable, instantaneous changes to core welding parameters such as laser power, welding speed, and focal point position within a preset non-critical region. Its purpose is to induce specific responses in the molten pool through these controlled disturbances, enabling subsequent capture and analysis.

[0152] A capture module can be understood as a sensing device used to monitor the state of the molten pool in real time and acquire its response characteristics. Specifically, a capture module may include one or more sensors, such as a high-speed camera, spectrometer, acoustic sensor, or thermocouple, which are arranged near the laser welding torch to capture the instantaneous geometric changes, temperature distribution, elemental spectral line intensity, or acoustic signal response characteristics of the molten pool under probing disturbances. Its purpose is to accurately and in real time acquire the dynamic feedback of the molten pool to disturbances.

[0153] The identification module can be understood as a processing unit used to analyze the captured instantaneous response characteristics and identify interfering factors. In practical applications, the identification module can be an industrial computer or a high-performance digital signal processor, internally running specific algorithms to compare the instantaneous response characteristics acquired by the capture module with pre-established physical correlation patterns. This comparison process aims to identify interfering factors related to specific weld quality fluctuations, such as material inhomogeneity, insufficient gas protection, or equipment vibration. Its purpose is to accurately diagnose the root cause of welding defects.

[0154] The adjustment module can be understood as an execution unit that corrects core welding parameters based on identified interference factors. For example, the adjustment module can be a control interface connected to a laser controller, robot motion controller, or gas flow controller. Once the identification module determines the specific interference factor, the adjustment module adjusts core welding parameters such as laser power, welding speed, wire feed speed, or shielding gas flow rate in real time according to a preset compensation strategy to counteract the influence of the interference factor, thereby stabilizing weld quality. Its purpose is to achieve adaptive optimization and quality control of the welding process.

[0155] The laser welding torch welding parameter optimization system of this application achieves real-time monitoring and adaptive adjustment of the laser welding process through the collaborative work of modules. First, the application module actively introduces minute probing disturbances into non-critical areas of the weld to activate the dynamic response of the molten pool. Then, the capture module uses high-sensitivity sensors to acquire the instantaneous response characteristics of the molten pool to these disturbances in real time; these characteristics contain rich information about the internal state of the molten pool and external interference. Next, the identification module performs in-depth analysis of these instantaneous response characteristics, comparing them with preset physical correlation patterns to accurately identify the specific interference factors causing weld quality fluctuations. Finally, the adjustment module precisely compensates and adjusts the core welding parameters based on the identified interference factors to eliminate or mitigate the adverse effects of interference factors on weld quality. Through this closed-loop control mechanism, the system can achieve intelligent management and optimization of the welding process.

[0156] Through the above technical solution, this application provides a system capable of practically deploying and operating a laser welding torch welding parameter optimization method. This system, by concretizing each step of the method into independent, collaborative modules, achieves automated, real-time, and intelligent control of the welding process. This not only improves the accuracy and response speed of welding parameter adjustments but also significantly reduces reliance on manual experience, enhancing the automation level and production efficiency of the welding production line. Furthermore, the system can dynamically adjust welding parameters based on real-time feedback, effectively coping with various complex welding environments and material changes, thereby ensuring the stability and consistency of weld quality and reducing scrap rates and rework costs.

[0157] The content disclosed above is only a preferred and feasible embodiment of the present invention, and is not intended to limit the scope of protection of the present invention. Therefore, all equivalent technical changes made based on the content of the present invention specification and drawings are included within the scope of protection of the present invention. Furthermore, the elements therein can be updated as technology develops.

Claims

1. A method for optimizing laser welding torch welding parameters, characterized in that, The method includes the following steps: When performing welding tasks with a laser welding gun, select non-critical areas in the weld and apply probing perturbations to the core welding parameters; When a probing disturbance is applied, sensors are used to capture the instantaneous response characteristics of the molten pool to the probing disturbance; By comparing instantaneous response characteristics with physical correlation patterns, interference factors that cause weld quality fluctuations can be identified. Based on the interference factors, the core welding parameters are compensated and adjusted.

2. The laser welding torch welding parameter optimization method as described in claim 1, characterized in that, The steps for identifying interfering factors causing weld quality fluctuations by comparing instantaneous response characteristics with physical correlation patterns include: The instantaneous response features are processed into multidimensional feature vectors to obtain multidimensional feature vectors. Multidimensional feature vectors are matched with pre-established physical association patterns that include combinations of various interference factors in a multidimensional manner. Based on the multi-dimensional matching results, the combination of interfering factors that cause weld quality fluctuations is identified, and then the interfering factors that cause weld quality fluctuations are derived.

3. The laser welding torch welding parameter optimization method as described in claim 2, characterized in that, The steps for performing multidimensional matching between multidimensional feature vectors and pre-established physical association patterns containing combinations of various interference factors include: The multidimensional feature vectors are initially matched with physical association patterns to obtain candidate patterns; Based on the combination of interference factors corresponding to the candidate modes, the detection perturbation strategy is adjusted to obtain the adjusted detection perturbation. Capture the instantaneous response of the molten pool to the adjusted probing disturbance and extract discriminative features; The discriminant features are compared with the specific response fingerprints of the candidate patterns to identify combinations of interfering factors that cause fluctuations in weld quality.

4. The laser welding torch welding parameter optimization method as described in claim 1, characterized in that, When the identified interfering factor is a highly localized micro-inhomogeneity, the steps for compensating and adjusting the core welding parameters based on the interfering factor include: Based on the spatial location and size information of the disturbance factors with highly localized micro-uniformity, a local compensation region is generated. By using robot trajectory planning and microsecond-level synchronous control of laser power, the laser power can be increased within the local compensation area. By using robot trajectory planning and microsecond-level synchronous control of laser power, the welding speed can be reduced within the local compensation area.

5. The laser welding torch welding parameter optimization method as described in claim 2, characterized in that, The steps to perform multidimensional feature vectorization on the instantaneous response features to obtain multidimensional feature vectors include: Time series analysis was performed on the instantaneous response characteristics to extract the rate of change curves of the geometric features of the molten pool over time; Calculate the slope, curvature, oscillation period, and decay constant of the rate of change curve of the molten pool geometry over time within a specific time window; Time series analysis of instantaneous response characteristics is performed to extract the curves of spectral line intensity of specific elements over time; Calculate the instantaneous rate of change, peak occurrence time, and duration of the spectral line intensity curve of a specific element over time; The slope, curvature, oscillation period, and decay constant of the rate of change curve of the geometric features of the molten pool over time within a specific time window are combined with the static features of the molten pool. The instantaneous rate of change, peak occurrence time, and duration of the spectral line intensity of a specific element over time are combined with the static features of the molten pool. All the combined results are fused to obtain a multidimensional feature vector.

6. The laser welding torch welding parameter optimization method as described in claim 5, characterized in that, The steps for calculating the slope, curvature, oscillation period, and decay constant of the rate of change curve of the molten pool geometry over a specific time window include: The rate of change curve of the geometric features of the molten pool over time is piecewise fitted, and the local slope and local curvature are calculated in each segment. Fourier transform is performed on the rate of change curve of the geometric features of the molten pool over time to extract the main oscillation frequency and the corresponding amplitude; Wavelet analysis was performed on the rate of change curve of the geometric features of the molten pool over time to extract the attenuation constants at different scales; By combining the local slope, local curvature, main oscillation frequency, amplitude, and decay constant at different scales, we can obtain the slope, curvature, oscillation period, and decay constant.

7. The laser welding torch welding parameter optimization method as described in claim 6, characterized in that, The steps for performing wavelet analysis on the rate of change curves of molten pool geometry over time to extract attenuation constants at different scales include: Based on the local characteristics of the rate of change curve of the molten pool geometry over time, the wavelet basis function is dynamically selected; By dynamically selecting wavelet basis functions, wavelet decomposition is performed on the rate of change curve of the geometric features of the molten pool over time to extract the attenuation constants at different scales.

8. The laser welding torch welding parameter optimization method as described in claim 7, characterized in that, Based on the local characteristics of the rate of change curve of the molten pool geometry over time, the steps for dynamically selecting wavelet basis functions include: Based on the local stationarity, local oscillation, and local sharpness of the rate of change curve of the molten pool geometry over time, select the corresponding wavelet basis function type; When the rate of change curve of the molten pool geometry over time exhibits local stationary characteristics, the Daubechies wavelet is selected as the wavelet basis function. When the rate of change curve of the molten pool geometry over time exhibits local oscillatory characteristics, the Coiflet wavelet is selected as the wavelet basis function. When the rate of change curve of the molten pool geometry over time exhibits sharp local characteristics, the Symlet wavelet is selected as the wavelet basis function.

9. The laser welding torch welding parameter optimization method as described in claim 6, characterized in that, The steps for performing wavelet analysis on the rate of change curves of molten pool geometry over time to extract attenuation constants at different scales include: Before performing wavelet analysis, the rate of change curve of the geometric features of the molten pool over time is adaptively filtered to remove high-frequency noise and outliers. Based on the characteristics of the rate of change curve of the geometric features of the molten pool over time after filtering, the wavelet basis function and the number of decomposition layers are selected. By selecting the wavelet basis function and the number of decomposition layers, wavelet analysis is performed on the rate of change curve of the filtered molten pool geometric features over time to extract the attenuation constants at different scales.

10. A laser welding torch welding parameter optimization system, applied to the laser welding torch welding parameter optimization method as described in claim 1, characterized in that, The system includes: The application module selects non-critical areas in the weld and applies probing perturbations to the core welding parameters when the laser welding gun performs the welding task. The capture module uses sensors to capture the instantaneous response characteristics of the molten pool to the probe disturbance when it is applied. The identification module compares instantaneous response characteristics with physical correlation patterns to identify interference factors that cause fluctuations in weld quality. The adjustment module compensates for and adjusts the core welding parameters based on interference factors.

Images