A Vision-Sensing-Based Online Monitoring and Adaptive Control Method for Laser Welding Quality

Abstract

This invention discloses a method for online monitoring and adaptive control of laser welding quality based on visual sensing, relating to the field of intelligent welding technology. The method includes: acquiring high-speed multispectral image sequences of the molten pool region during laser welding to obtain a raw visual dataset; extracting the contour of the molten pool region from the raw visual dataset and calculating dynamic data on the area, width, and length of the molten pool region over time to form temporal features of the molten pool morphology; analyzing the energy distribution within the defect-sensitive frequency band and calculating the spatial asymmetry of the edge gradient based on the temporal features of the molten pool morphology to obtain the molten pool oscillation energy ratio and spatial structure features; fusing the temporal features of the molten pool morphology, the molten pool oscillation energy ratio, and the spatial structure features to construct a three-domain joint defect-sensitive feature vector; and training and deploying a multivariate collaborative control model offline based on historical welding data and preset quality evaluation standards.

Description

Technical Field

[0001] This invention relates to the field of intelligent welding technology, and in particular to a method for online monitoring and adaptive control of laser welding quality based on vision sensing. Background Technology

[0002] Laser welding, with its advantages of high energy density, small heat-affected zone, large aspect ratio, and ease of automation, has been widely used in high-end manufacturing fields such as aerospace, new energy vehicles, and precision electronics. In recent years, with the deepening of Industry 4.0 and intelligent manufacturing, higher requirements have been placed on the stability of the welding process and the consistency of weld quality, prompting researchers to introduce visual sensing, signal processing, and intelligent control technologies into laser welding closed-loop systems.

[0003] Most systems rely solely on the temporal variation of the molten pool area or width, neglecting key precursors such as high-frequency oscillations and spatial geometric asymmetry. This results in insufficient sensitivity to early, subtle anomalies. Furthermore, existing adaptive control systems are mostly based on PID architectures, which only adjust a single parameter, failing to achieve coordinated optimization of multiple variables such as laser power, welding speed, and defocusing amount. Consequently, they struggle to cope with complex disturbances in strongly coupled nonlinear welding processes. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a vision-sensing-based online monitoring and adaptive control method for laser welding quality. This addresses the problem that most systems rely solely on the temporal changes in the molten pool area or width, neglecting key defect precursors such as high-frequency oscillations and spatial geometric asymmetry. This results in insufficient sensitivity to early, subtle anomalies. Furthermore, existing adaptive control systems are mostly based on PID architectures, which only adjust a single parameter, failing to achieve coordinated optimization of multiple variables such as laser power, welding speed, and defocusing amount. Consequently, they struggle to cope with complex disturbances in strongly coupled nonlinear welding processes.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for online monitoring and adaptive control of laser welding quality based on vision sensing, which includes the following steps: High-speed multispectral image sequences of the molten pool region during laser welding were acquired to obtain the raw visual dataset; Extract the outline of the molten pool region from the original visual dataset, and calculate the dynamic data of the area, width and length of the molten pool region as a function of time to form the temporal features of the molten pool morphology. Based on the temporal characteristics of the molten pool morphology, the energy distribution in the defect-sensitive frequency band is analyzed and the spatial asymmetry of the edge gradient is calculated to obtain the molten pool oscillation energy ratio and spatial structure characteristics. By fusing the temporal characteristics of the molten pool morphology, the molten pool oscillation energy ratio, and the spatial structure characteristics, a three-domain joint defect-sensitive feature vector is constructed. Based on historical welding data and preset quality evaluation standards, a multivariate collaborative control model is trained and deployed offline. The three-domain joint defect-sensitive feature vector is input into the multivariate collaborative control model, and the output includes multivariate collaborative control commands including laser power adjustment, welding speed correction and defocus adjustment. Based on multi-variable collaborative control commands, the laser output power, workpiece movement speed, and optical focusing position are synchronously adjusted to achieve closed-loop intervention of the dynamic behavior of the molten pool.

[0007] As a preferred embodiment of the vision-sensing-based online monitoring and adaptive control method for laser welding quality described in this invention, the specific steps for acquiring high-speed multispectral image sequences of the molten pool region during laser welding to obtain the original visual dataset are as follows: A high-speed CMOS camera is integrated into the coaxial optical path of the laser welding head, and a dual-channel optical filter module is configured at the imaging front end; The dual-channel images are exposed and acquired at the same welding moment by controlling the hardware synchronous trigger signal. The images obtained in the two bands in time sequence are aligned by timestamp and packaged to form the original visual dataset.

[0008] As a preferred embodiment of the online monitoring and adaptive control method for laser welding quality based on vision sensing described in this invention, the specific steps for extracting the molten pool region contour from the original visual dataset and calculating the dynamic data of the molten pool region area, width, and length changing over time to form temporal features of the molten pool morphology are as follows: The dual-channel images at each moment in the original visual dataset are weighted and fused. The fusion weights are pre-calibrated based on the surface reflectivity of the welded material in the visible light band and the thermal radiation intensity in the near-infrared band, resulting in a single-channel molten pool image with enhanced contrast. The single-channel molten pool image is first subjected to convolution filtering using a two-dimensional Gaussian kernel to suppress high-frequency noise and preserve edge details; The Canny edge detection algorithm is used to adaptively set a high threshold and determine a low threshold in the gradient magnitude histogram. After suppression and hysteresis threshold processing, an initial binary edge map is obtained. Then, morphological closing operations are performed on the edge map. Circular structuring elements are first expanded to connect the broken edges and fill the internal holes, and then erosion is performed to restore the edge position, thus obtaining the molten pool outline. The minimum bounding rectangle is calculated based on the molten pool profile, and the longer side of the rectangle is defined as the molten pool length. The shorter side is defined as the width of the molten pool. Simultaneously, the number of all pixels within the contour is counted and multiplied by the actual physical area corresponding to a single pixel to obtain the molten pool area. ; Will , and Arranged sequentially according to welding time, they form a temporal characteristic sequence of molten pool morphology.

[0009] As a preferred embodiment of the vision-sensing-based online monitoring and adaptive control method for laser welding quality described in this invention, the specific steps of analyzing the energy distribution in the defect-sensitive frequency band and calculating the spatial asymmetry of the edge gradient based on the temporal characteristics of the molten pool morphology to obtain the molten pool oscillation energy ratio and spatial structure characteristics are as follows: For the molten pool area sequence Perform a discrete Fourier transform to obtain the frequency domain representation. ,in For frequency variables; The defect-sensitive frequency band is defined as the frequency range that has been experimentally calibrated. The energy level within the range is higher in defect samples than in normal welds; The proportion of energy within the calculated frequency band to the total frequency domain energy is used as the molten pool oscillation energy ratio, expressed as: ; in, The oscillation energy ratio of the molten pool represents the molten pool area sequence. The energy percentage within a specific defect-sensitive frequency band is used to quantify the high-frequency oscillation level of the molten pool. It is the lower limit frequency of the defect-sensitive frequency band. It is the upper limit frequency of the defect-sensitive frequency band. This represents the Fourier transform operator used to calculate the time series of molten pool area. The spectral distribution, in which It is a frequency variable, numerator part The calculation is performed in the defect-sensitive frequency band. The total energy within, the denominator It is the entire resolvable frequency domain. It is the Nyquist frequency corresponding to the image acquisition frame rate; Uniform sampling along the molten pool contour in a clockwise direction One point, For each sampling point, calculate the image gradient magnitude if it is an even number greater than or equal to 64. Angle with contour normal And define the spatial asymmetry index as follows, with the expression: ; in, It is the spatial asymmetry of the gradient at the edge of the molten pool. It is the total number of points sampled uniformly along the molten pool contour. It is the first Image gradient magnitude at each sampling point Is with the first The gradient magnitude of a point at a radial position on the contour. It is the first The angle between the contour normal and the horizontal axis at each sampling point It is the normal angle at the radial point.

[0010] As a preferred embodiment of the vision-sensing-based online monitoring and adaptive control method for laser welding quality described in this invention, the specific steps for fusing the temporal characteristics of the molten pool morphology, the molten pool oscillation energy ratio, and the spatial structure characteristics to construct a three-domain joint defect-sensitive feature vector are as follows: Set steady-state molten pool area For process calibration, the real-time molten pool area After normalization, the expression is: ; in, To normalize the molten pool area, For process calibration values, This represents the real-time molten pool area; The normalized melt pool morphology index is calculated using the following expression: ; in, Represents the normalized melt pool morphology index. Indicates the width of the molten pool. Indicates the length of the molten pool; Will , , and Concatenated in a fixed order to form a four-dimensional vector, the expression is: ; in, It is a three-domain joint defect-sensitive feature vector. To normalize the molten pool area, Represents the normalized melt pool morphology index. The gradient spatial asymmetry at the edge of the molten pool. It is the energy ratio of molten pool oscillation.

[0011] As a preferred embodiment of the vision-sensing-based online monitoring and adaptive control method for laser welding quality described in this invention, the following steps are included: Based on historical welding data and preset quality evaluation standards, a multivariate collaborative control model is trained offline and deployed. A three-domain joint defect-sensitive feature vector is input into the multivariate collaborative control model, and a multivariate collaborative control command including laser power adjustment, welding speed correction, and defocusing adjustment is output. The specific steps are as follows: Construct a labeled dataset containing joint three-domain defect-sensitive feature vectors of welding samples. Corresponding actual process parameters And the results of post-weld quality inspection; Define reward function based on labeled dataset When satisfied , and At that time, reward value This indicates that the molten pool maintains a high-quality, stable region; when this condition is not met... This indicates that the current control actions failed to maintain the stability of the molten pool. The target morphological index. , , The preset tolerance threshold; A deep neural network policy model is trained using a proximal policy optimization algorithm. This enables deep neural network policy models to... Given a state as input, the output is an action vector, expressed as: ; in, For multivariable collaborative control command vectors, Laser power adjustment amount This is a correction amount for welding speed. This is the defocus adjustment amount; After training, the strategy model is solidified into a multivariate collaborative control model and deployed on an embedded industrial control platform.

[0012] As a preferred embodiment of the vision-sensing-based online monitoring and adaptive control method for laser welding quality described in this invention, the specific steps of synchronously adjusting the laser output power, workpiece movement speed, and optical focusing position according to multi-variable collaborative control commands to achieve closed-loop intervention of the dynamic behavior of the molten pool are as follows: The output of the multivariable collaborative control model The signal is sent to the laser power supply driver module for real-time adjustment of the laser output power. Will The data is sent to the servo motion controller to correct the travel speed of the workpiece or welding head. Will Send to the Z-axis electric focusing mechanism to adjust the position of the focusing lens relative to the workpiece surface; The three actuators respond synchronously to control commands under a unified system clock, completing the coordinated control of laser energy density, heat input rate and spot size; After the adjustment is completed, the newly acquired molten pool image is automatically used as the raw visual dataset for the next cycle and enters the next round of monitoring and control, thus forming a continuous online closed-loop adaptive control process.

[0013] As a preferred embodiment of the vision-sensing-based online monitoring and adaptive control method for laser welding quality described in this invention, wherein: the defect-sensitive frequency band

[0014] For a specific combination of materials and processes, collect no fewer than 100 sets of welding samples with known defect types. Each set of samples includes a time series of the molten pool area and a defect type label. Perform spectral analysis on each sequence and calculate the energy percentage; The energy concentration of various defects in different frequency bands was statistically analyzed, and frequency ranges in which the energy proportion of at least two types of defects exceeded three times that of normal welds and the occurrence frequency was higher than 80% were selected as defect-sensitive frequency bands. ; The interval parameters are written into the system configuration file and directly called during the online monitoring phase to calculate the molten pool oscillation energy ratio. .

[0015] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the vision-sensing-based online monitoring and adaptive control method for laser welding quality as described in the first aspect of the present invention.

[0016] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the vision-sensing-based online monitoring and adaptive control method for laser welding quality as described in the first aspect of the present invention.

[0017] The beneficial effects of this invention are as follows: By sequentially performing Gaussian filtering, Canny edge detection, and morphological closing operations on the single-channel molten pool image after dual-channel weighted fusion, closed, continuous, and geometrically accurate molten pool contour extraction is achieved. Then, based on the contour, the minimum bounding rectangle and internal pixel area are calculated to obtain a dynamic sequence with high spatiotemporal resolution. By concatenating the normalized area, morphological index, oscillation energy ratio, and spatial asymmetry into a four-dimensional vector, deep feature fusion of the three dimensions of time domain, frequency domain, and spatial domain is achieved. The feature vector not only retains the key criteria of each domain, but also eliminates the scale drift caused by the difference in process parameters through normalization, so that the same control model can be reused across materials and power levels, enhancing the system's generalization ability and engineering practicality. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a vision-sensing-based online monitoring and adaptive control method for laser welding quality. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0023] Reference Figure 1 This embodiment of the invention provides a method for online monitoring and adaptive control of laser welding quality based on vision sensing, comprising the following steps: S1. Acquire high-speed multispectral image sequences of the molten pool region during laser welding to obtain the original visual dataset.

[0024] Furthermore, a high-speed CMOS camera is integrated into the coaxial optical path of the laser welding head, and a dual-channel optical filter module is configured at the imaging front end; The dual-channel images are exposed and acquired at the same welding moment by controlling the hardware synchronous trigger signal. The images obtained in the two bands in time sequence are aligned by timestamp and packaged to form the original visual dataset.

[0025] It should be noted that by integrating a high-speed CMOS camera with a dual-channel optical filter module and using hardware synchronous triggering signals to ensure exposure and image acquisition at the same welding moment, the accuracy and real-time performance of feature extraction in the molten pool area can be effectively improved, laying the foundation for subsequent high-quality visual data analysis.

[0026] S2. Extract the outline of the molten pool region from the original visual dataset, and calculate the dynamic data of the area, width and length of the molten pool region changing over time to form the temporal features of the molten pool morphology.

[0027] Furthermore, the dual-channel images at each moment in the original visual dataset are weighted and fused. The fusion weights are pre-calibrated based on the surface reflectivity of the welded material in the visible light band and the thermal radiation intensity in the near-infrared band, resulting in a single-channel molten pool image with enhanced contrast. The single-channel molten pool image is first subjected to convolution filtering using a two-dimensional Gaussian kernel to suppress high-frequency noise and preserve edge details; The Canny edge detection algorithm is used to adaptively set a high threshold and determine a low threshold in the gradient magnitude histogram. After suppression and hysteresis threshold processing, an initial binary edge map is obtained. Then, morphological closing operations are performed on the edge map. Circular structuring elements are first expanded to connect the broken edges and fill the internal holes, and then erosion is performed to restore the edge position, thus obtaining the molten pool outline. The minimum bounding rectangle is calculated based on the molten pool profile, and the longer side of the rectangle is defined as the molten pool length. The shorter side is defined as the width of the molten pool. Simultaneously, the number of all pixels within the contour is counted and multiplied by the actual physical area corresponding to a single pixel to obtain the molten pool area. ; Will , and Arranged sequentially according to welding time, they form a temporal characteristic sequence of molten pool morphology.

[0028] It should be noted that weighted fusion, convolutional filtering, and edge detection of the original visual data can enhance the contrast of the molten pool image and accurately extract the molten pool contour, thereby enabling high-precision dynamic monitoring of changes in the area, width, and length of the molten pool, which helps to detect potential defects in a timely manner.

[0029] S3. Based on the temporal characteristics of the molten pool morphology, analyze the energy distribution in the defect-sensitive frequency band and calculate the spatial asymmetry of the edge gradient to obtain the molten pool oscillation energy ratio and spatial structure characteristics.

[0030] Furthermore, regarding the sequence of molten pool areas... Perform a discrete Fourier transform to obtain the frequency domain representation. ,in For frequency variables; The defect-sensitive frequency band is defined as the frequency range that has been experimentally calibrated. The energy level within the range is higher in defect samples than in normal welds; The proportion of energy within the calculated frequency band to the total frequency domain energy is used as the molten pool oscillation energy ratio, expressed as: ; in, The oscillation energy ratio of the molten pool represents the molten pool area sequence. The energy percentage within a specific defect-sensitive frequency band is used to quantify the high-frequency oscillation level of the molten pool. It is the lower limit frequency of the defect-sensitive frequency band. It is the upper limit frequency of the defect-sensitive frequency band. This represents the Fourier transform operator used to calculate the time series of molten pool area. The spectral distribution, in which It is a frequency variable, numerator part The calculation is performed in the defect-sensitive frequency band. The total energy within, the denominator It is the entire resolvable frequency domain. It is the Nyquist frequency corresponding to the image acquisition frame rate; Uniform sampling along the molten pool contour in a clockwise direction One point, For each sampling point, calculate the image gradient magnitude if it is an even number greater than or equal to 64. Angle with contour normal And define the spatial asymmetry index as follows, with the expression: ; in, It is the spatial asymmetry of the gradient at the edge of the molten pool. It is the total number of points sampled uniformly along the molten pool contour. It is the first Image gradient magnitude at each sampling point Is with the first The gradient magnitude of a point at a radial position on the contour. It is the first The angle between the contour normal and the horizontal axis at each sampling point It is the normal angle at the radial point; For a specific combination of materials and processes, collect no fewer than 100 sets of welding samples with known defect types. Each set of samples includes a time series of the molten pool area and a defect type label. Perform spectral analysis on each sequence and calculate the energy percentage; The energy concentration of various defects in different frequency bands was statistically analyzed, and frequency ranges in which the energy proportion of at least two types of defects exceeded three times that of normal welds and the occurrence frequency was higher than 80% were selected as defect-sensitive frequency bands. ; The interval parameters are written into the system configuration file and directly called during the online monitoring phase to calculate the molten pool oscillation energy ratio. .

[0031] It should be noted that analyzing the energy distribution and computational space asymmetry of the time-domain characteristics of the molten pool morphology within the defect-sensitive frequency band can not only quantify the degree of high-frequency oscillation of the molten pool, but also identify spatial structural anomalies that may cause welding defects, thus providing a scientific basis for early warning and control adjustments.

[0032] S4. The temporal characteristics of the molten pool morphology, the molten pool oscillation energy ratio, and the spatial structure characteristics are fused to construct a three-domain joint defect-sensitive feature vector.

[0033] Furthermore, the steady-state molten pool area is set. For process calibration, the real-time molten pool area After normalization, the expression is: ; in, To normalize the molten pool area, For process calibration values, This represents the real-time molten pool area; The normalized melt pool morphology index is calculated using the following expression: ; in, Represents the normalized melt pool morphology index. Indicates the width of the molten pool. Indicates the length of the molten pool; Will , , and Concatenated in a fixed order to form a four-dimensional vector, the expression is: ; in, It is a three-domain joint defect-sensitive feature vector. To normalize the molten pool area, Represents the normalized melt pool morphology index. The gradient spatial asymmetry at the edge of the molten pool. It is the energy ratio of molten pool oscillation.

[0034] It should be noted that constructing a three-domain joint defect-sensitive feature vector organically combines the temporal characteristics of the molten pool morphology, the oscillation energy ratio, and the spatial structure characteristics. This multi-dimensional information fusion method can more comprehensively reflect the state of the molten pool and improve the identification ability and prediction accuracy of different types of welding defects.

[0035] S5. Based on historical welding data and preset quality evaluation standards, a multivariate collaborative control model is trained and deployed offline. The three-domain joint defect-sensitive feature vector is input into the multivariate collaborative control model, and the output includes multivariate collaborative control commands including laser power adjustment, welding speed correction, and defocus adjustment.

[0036] Furthermore, an annotated dataset is constructed, containing joint three-domain defect-sensitive feature vectors of welding samples. Corresponding actual process parameters And the results of post-weld quality inspection; Define reward function based on labeled dataset When satisfied , and At that time, reward value This indicates that the molten pool maintains a high-quality, stable region; when this condition is not met... This indicates that the current control actions failed to maintain the stability of the molten pool. The target morphological index. , , The preset tolerance threshold; A deep neural network policy model is trained using a proximal policy optimization algorithm. This enables deep neural network policy models to... Given a state as input, the output is an action vector, expressed as: ; in, For multivariable collaborative control command vectors, Laser power adjustment amount This is a correction amount for welding speed. This is the defocus adjustment amount; After training, the strategy model is solidified into a multivariate collaborative control model and deployed on an embedded industrial control platform.

[0037] It should be noted that by defining a reward function based on a labeled dataset and training a deep neural network policy model, the system can automatically learn the optimal control strategy during complex welding processes, achieving intelligent closed-loop control from perception to decision-making, and significantly improving welding quality and production efficiency.

[0038] S6. Based on multi-variable collaborative control commands, the laser output power, workpiece movement speed and optical focusing position are synchronously adjusted to achieve closed-loop intervention of the dynamic behavior of the molten pool.

[0039] Furthermore, the output of the multivariable collaborative control model... The signal is sent to the laser power supply driver module for real-time adjustment of the laser output power. Will The data is sent to the servo motion controller to correct the travel speed of the workpiece or welding head. Will Send to the Z-axis electric focusing mechanism to adjust the position of the focusing lens relative to the workpiece surface; The three actuators respond synchronously to control commands under a unified system clock, completing the coordinated control of laser energy density, heat input rate and spot size; After the adjustment is completed, the newly acquired molten pool image is automatically used as the raw visual dataset for the next cycle and enters the next round of monitoring and control, thus forming a continuous online closed-loop adaptive control process.

[0040] It should be noted that by synchronously adjusting the laser output power, workpiece movement speed, and optical focusing position according to the multi-variable collaborative control command, precise intervention in the dynamic behavior of the molten pool can be achieved, ensuring that the welding process is always within the optimal process parameter range, effectively preventing welding defects, and ensuring the consistency and stability of weld quality.

[0041] This embodiment also provides a computer device applicable to the online monitoring and adaptive control method for laser welding quality based on vision sensing, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the online monitoring and adaptive control method for laser welding quality based on vision sensing proposed in the above embodiment.

[0042] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0043] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the vision-sensing-based online monitoring and adaptive control method for laser welding quality as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0044] In summary, this invention achieves closed, continuous, and geometrically accurate molten pool contour extraction by sequentially performing Gaussian filtering, Canny edge detection, and morphological closing operations on a single-channel molten pool image obtained through weighted fusion of dual channels. Then, based on the contour, the minimum bounding rectangle and internal pixel area are calculated to obtain a dynamic sequence with high spatiotemporal resolution. By concatenating normalized area, morphological index, oscillation energy ratio, and spatial asymmetry into a four-dimensional vector, deep feature fusion across the time, frequency, and spatial domains is achieved. The feature vector not only retains key criteria from each domain but also eliminates scale drift caused by differences in process parameters through normalization. This allows the same control model to be reused across materials and power levels, enhancing the system's generalization ability and engineering practicality.

[0045] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for online monitoring and adaptive control of laser welding quality based on vision sensing, characterized in that: Includes the following steps: High-speed multispectral image sequences of the molten pool region during laser welding were acquired to obtain the raw visual dataset; Extract the outline of the molten pool region from the original visual dataset, and calculate the dynamic data of the area, width and length of the molten pool region as a function of time to form the temporal features of the molten pool morphology. Based on the temporal characteristics of the molten pool morphology, the energy distribution in the defect-sensitive frequency band is analyzed and the spatial asymmetry of the edge gradient is calculated to obtain the molten pool oscillation energy ratio and spatial structure characteristics. By fusing the temporal characteristics of the molten pool morphology, the molten pool oscillation energy ratio, and the spatial structure characteristics, a three-domain joint defect-sensitive feature vector is constructed. Based on historical welding data and preset quality evaluation standards, a multivariate collaborative control model is trained and deployed offline. The three-domain joint defect-sensitive feature vector is input into the multivariate collaborative control model, and the output includes multivariate collaborative control commands including laser power adjustment, welding speed correction and defocus adjustment. Based on multi-variable collaborative control commands, the laser output power, workpiece movement speed, and optical focusing position are synchronously adjusted to achieve closed-loop intervention of the dynamic behavior of the molten pool.

2. The method for online monitoring and adaptive control of laser welding quality based on vision sensing as described in claim 1, characterized in that: The process of acquiring high-speed multispectral image sequences of the molten pool region during laser welding to obtain the original visual dataset involves the following steps: A high-speed CMOS camera is integrated into the coaxial optical path of the laser welding head, and a dual-channel optical filter module is configured at the imaging front end; The dual-channel images are exposed and acquired at the same welding moment by controlling the hardware synchronous trigger signal. The images obtained in the two bands in time sequence are aligned by timestamp and packaged to form the original visual dataset.

3. The method for online monitoring and adaptive control of laser welding quality based on vision sensing as described in claim 2, characterized in that: The steps for extracting the molten pool region contour from the original visual dataset and calculating the dynamic data of the molten pool region area, width, and length changing over time to form the temporal features of the molten pool morphology are as follows: The dual-channel images at each moment in the original visual dataset are weighted and fused. The fusion weights are pre-calibrated based on the surface reflectivity of the welded material in the visible light band and the thermal radiation intensity in the near-infrared band, resulting in a single-channel molten pool image with enhanced contrast. The single-channel molten pool image is first subjected to convolution filtering using a two-dimensional Gaussian kernel to suppress high-frequency noise and preserve edge details; The Canny edge detection algorithm is used to adaptively set a high threshold and determine a low threshold in the gradient magnitude histogram. After suppression and hysteresis threshold processing, an initial binary edge map is obtained. Then, morphological closing operations are performed on the edge map. Circular structuring elements are first expanded to connect the broken edges and fill the internal holes, and then erosion is performed to restore the edge position, thus obtaining the molten pool outline. The minimum bounding rectangle is calculated based on the molten pool profile, and the longer side of the rectangle is defined as the molten pool length. The shorter side is defined as the width of the molten pool. Simultaneously, the number of all pixels within the contour is counted and multiplied by the actual physical area corresponding to a single pixel to obtain the molten pool area. ; Will , and Arranged sequentially according to welding time, they form a temporal characteristic sequence of molten pool morphology.

4. The method for online monitoring and adaptive control of laser welding quality based on vision sensing as described in claim 3, characterized in that: The steps are as follows: Based on the temporal characteristics of the molten pool morphology, the energy distribution within the defect-sensitive frequency band is analyzed, and the spatial asymmetry of the edge gradient is calculated to obtain the molten pool oscillation energy ratio and spatial structure characteristics. For the molten pool area sequence Perform a discrete Fourier transform to obtain the frequency domain representation. ,in For frequency variables; The defect-sensitive frequency band is defined as the frequency range that has been experimentally calibrated. The energy level within the range is higher in defect samples than in normal welds; The proportion of energy within the calculated frequency band to the total frequency domain energy is used as the molten pool oscillation energy ratio, expressed as: ; in, The oscillation energy ratio of the molten pool represents the molten pool area sequence. The energy percentage within a specific defect-sensitive frequency band is used to quantify the high-frequency oscillation level of the molten pool. It is the lower limit frequency of the defect-sensitive frequency band. It is the upper limit frequency of the defect-sensitive frequency band. This represents the Fourier transform operator used to calculate the time series of molten pool area. The spectral distribution, in which It is a frequency variable, numerator part The calculation is performed in the defect-sensitive frequency band. The total energy within, the denominator It is the entire resolvable frequency domain. It is the Nyquist frequency corresponding to the image acquisition frame rate; Uniform sampling along the molten pool contour in a clockwise direction One point, For each sampling point, calculate the image gradient magnitude if it is an even number greater than or equal to 64. Angle with contour normal And define the spatial asymmetry index as follows, with the expression: ; in, It is the spatial asymmetry of the gradient at the edge of the molten pool. It is the total number of points sampled uniformly along the molten pool contour. It is the first Image gradient magnitude at each sampling point Is with the first The gradient magnitude of a point at a radial position on the contour. It is the first The angle between the contour normal and the horizontal axis at each sampling point It is the normal angle at the radial point.

5. The method for online monitoring and adaptive control of laser welding quality based on vision sensing as described in claim 4, characterized in that: The specific steps for fusing the temporal characteristics of the molten pool morphology, the molten pool oscillation energy ratio, and the spatial structure characteristics to construct a three-domain joint defect-sensitive feature vector are as follows: Set steady-state molten pool area For process calibration, the real-time molten pool area After normalization, the expression is: ； in, To normalize the molten pool area, For process calibration values, This represents the real-time molten pool area; The normalized melt pool morphology index is calculated using the following expression: ； in, Represents the normalized melt pool morphology index. Indicates the width of the molten pool. Indicates the length of the molten pool; Will , , and Concatenated in a fixed order to form a four-dimensional vector, the expression is: ; in, It is a three-domain joint defect-sensitive feature vector. To normalize the molten pool area, Represents the normalized melt pool morphology index. The gradient spatial asymmetry at the edge of the molten pool. It is the energy ratio of molten pool oscillation.

6. The method for online monitoring and adaptive control of laser welding quality based on vision sensing as described in claim 5, characterized in that: The process involves offline training and deployment of a multivariate collaborative control model based on historical welding data and preset quality evaluation standards. The three-domain joint defect-sensitive feature vector is input into the multivariate collaborative control model, which outputs multivariate collaborative control commands including laser power adjustment, welding speed correction, and defocusing adjustment. The specific steps are as follows: Construct a labeled dataset containing joint three-domain defect-sensitive feature vectors of welding samples. Corresponding actual process parameters And the results of post-weld quality inspection; Define reward function based on labeled dataset When satisfied , and At that time, reward value This indicates that the molten pool maintains a high-quality, stable region; when this condition is not met... This indicates that the current control actions failed to maintain the stability of the molten pool. The target morphological index. , , The preset tolerance threshold; A deep neural network policy model is trained using a proximal policy optimization algorithm. This enables deep neural network policy models to... Given a state as input, the output is an action vector, expressed as: ; in, For multivariable collaborative control command vectors, Laser power adjustment amount This is a correction amount for welding speed. This is the defocus adjustment amount; After training, the strategy model is solidified into a multivariate collaborative control model and deployed on an embedded industrial control platform.

7. The method for online monitoring and adaptive control of laser welding quality based on vision sensing as described in claim 6, characterized in that: The method of synchronously adjusting laser output power, workpiece movement speed, and optical focusing position based on multi-variable collaborative control commands to achieve closed-loop intervention of the dynamic behavior of the molten pool includes the following steps: The output of the multivariable collaborative control model The signal is sent to the laser power supply driver module for real-time adjustment of the laser output power. Will The data is sent to the servo motion controller to correct the travel speed of the workpiece or welding head. Will Send to the Z-axis electric focusing mechanism to adjust the position of the focusing lens relative to the workpiece surface; The three actuators respond synchronously to control commands under a unified system clock, completing the coordinated control of laser energy density, heat input rate and spot size; After the adjustment is completed, the newly acquired molten pool image is automatically used as the raw visual dataset for the next cycle and enters the next round of monitoring and control, thus forming a continuous online closed-loop adaptive control process.

8. The method for online monitoring and adaptive control of laser welding quality based on vision sensing as described in claim 7, characterized in that, The method for determining the defect-sensitive frequency band is as follows: For a specific combination of materials and processes, collect no fewer than 100 sets of welding samples with known defect types. Each set of samples includes a time series of the molten pool area and a defect type label. Perform spectral analysis on each sequence and calculate the energy percentage; The energy concentration of various defects in different frequency bands was statistically analyzed, and frequency ranges in which the energy proportion of at least two types of defects exceeded three times that of normal welds and the occurrence frequency was higher than 80% were selected as defect-sensitive frequency bands. ; The interval parameters are written into the system configuration file and directly called during the online monitoring phase to calculate the molten pool oscillation energy ratio. .

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the online monitoring and adaptive control method for laser welding quality based on vision sensing as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the online monitoring and adaptive control method for laser welding quality based on vision sensing as described in any one of claims 1 to 8.

Images