Narrow gap welding system

Abstract

The present application discloses a narrow gap welding system, comprising a gun head, a laser device and a posture control unit, the gun head is internally provided with an eccentric nozzle, the nozzle is driven by a rotating mechanism to reciprocate, so that the welding wire can move along a predetermined track in the welding work area; the laser device comprises at least two groups or two working states of laser emitting units and laser receiving units, wherein at least one group or one working state of the laser emitting units and the laser receiving units are used to obtain the surface image of the welding work area to analyze the welding working state; at least one other group or one working state of the laser emitting units and the laser receiving units act on the molten pool to assist welding; the posture control unit is used to receive real-time working parameters including the surface image, and adjust the working parameters of the laser device, the nozzle and the welding wire. The present application can track and optimize the welding quality of the narrow gap weld throughout the whole process.

Description

TECHNICAL FIELD

[0001] The present application relates to welding technology, in particular to a narrow gap welding system and method. BACKGROUND

[0002] Narrow gap welding (NGW) is a high-efficiency welding method suitable for thick-wall butt joints, which can reduce the amount of filler metal, reduce heat input, shorten welding time, and reduce deformation and residual stress. Narrow gap welding is widely used in heavy industry fields such as nuclear energy, aerospace, shipbuilding, and offshore pipelines. Among them, structural steel is the most commonly used material because it has good metallurgical stability, high corrosion resistance, and good creep and ductility at high temperatures.

[0003] The main narrow gap welding techniques include narrow gap submerged arc welding (NGSAW), narrow gap gas metal arc welding (NGGMAW), narrow gap gas tungsten arc welding (NGGTAW), narrow gap laser welding (NGLW), narrow gap electron beam welding (NGEBW), and hybrid welding methods. In order to achieve stable welding in a deep and narrow groove, special narrow gap welding torches and automatic tracking systems are required for the above-mentioned techniques.

[0004] However, narrow gap welding also faces some challenges and problems, such as: due to insufficient heat conduction between the groove sidewall and the molten pool, incomplete fusion or penetration defects may occur. In order to improve the sidewall fusion quality, a swinging, tilting or wavy wire is required to increase the effect of the arc on the sidewall. However, these methods also increase the instability of the arc and heat input, affecting the formation and mechanical properties of the weld. In some cases, the groove depth is greater than the length of the welding torch, resulting in ineffective coverage of the molten pool surface by the protective gas, causing poor or insufficient gas protection. In order to improve the gas protection effect, it is necessary to increase the gas flow or use a special-shaped welding torch. However, these methods also increase gas consumption and cost, as well as arc blowing force and deformation. Due to the small groove width compared to the welding torch diameter, the space between the welding torch and the groove is very limited. In order to prevent the welding torch from overheating or colliding with the groove, a water-cooled or insulated welding torch is required, equipped with a precise tracking system. However, these methods also increase the complexity and failure rate of the equipment.

[0005] Therefore, further research and innovation are needed to solve the above-mentioned problems of the prior art. SUMMARY

[0006] The present application relates to welding technology, in particular to a narrow gap welding system and method.

[0007] Technical solution, according to one aspect of the present application, a narrow gap welding system is provided, comprising:

[0008] The gun head is internally provided with an eccentric gun nozzle, which is driven by a rotating mechanism to reciprocate so as to move the welding wire along a predetermined track in a welding operation area;

[0009] The laser device comprises at least two groups or two working states of laser emitting units and laser receiving units,

[0010] At least one group or one working state of laser emitting units and laser receiving units is used to acquire a surface image of a welding operation area to analyze a welding working state;

[0011] At least another group or one working state of laser emitting units and laser receiving units acts on a molten pool to assist welding;

[0012] The posture control unit is used to receive real-time working parameters including the surface image and adjust working parameters of the laser device, the gun nozzle and the welding wire.

[0013] According to one aspect of the present application, the laser device comprises two groups, which comprise:

[0014] The first laser emitting unit and the second laser emitting unit are respectively used to emit first laser and second laser of a predetermined frequency;

[0015] The beam combiner combines the first laser and the second laser into a coaxial light;

[0016] The emitting light guide directs the coaxial light to a predetermined working area;

[0017] The receiving light guide receives laser reflected by the working area and guides the laser to the laser detector and then to the industrial computer.

[0018] According to one aspect of the present application, a beam splitter is further arranged between the receiving light guide and the laser detector, and output ends of the beam splitter are respectively connected to the laser detector.

[0019] According to one aspect of the present application, the first laser emitting unit is multiple and arranged in the laser device body in an off-axis manner, and the second laser emitting unit is rotationally connected to the laser device body, and the position of the second laser emitting unit is adjusted according to the working state of each first laser emitting unit, and the second laser emitting unit forms a predetermined angle with one of the first laser emitting units when the second laser emitting unit rotates to a working position.

[0020] According to one aspect of the present application, the industrial computer is provided with a laser data processing unit, and the laser data processing unit comprises:

[0021] The preprocessing module acquires electrical signal data in the laser detector and performs filtering, enhancement and extraction of characteristic parameters of the signal; the characteristic parameters include amplitude, frequency, phase and polarization;

[0022] a pre-configuration model module, analyzing laser beam characteristics, establishing physical models of each stage in the propagation process of the laser beam;

[0023] a data analysis module, based on the physical models, constructing a data-driven laser image analysis and detection module, extracting time sequence features and region features, calculating and outputting welding quality information to the control module of the industrial computer;

[0024] a weld detection module, obtaining weld parameters from the laser image.

[0025] According to one aspect of the present application, the laser device is provided with an adaptive laser modulation unit for calculating optimal working parameters according to pre-stored historical data, and controlling the working parameters of the first and second laser emission units according to the optimal working parameters;

[0026] The historical data includes at least welding parameters, weld shape, welding material and environmental temperature; and the optimal working parameters include at least wavelength, power, pulse width and frequency.

[0027] According to one aspect of the present application, the laser detector is at least two, respectively for receiving laser of different wavelengths and converting into electrical signal;

[0028] The number of pre-processing modules is the same as that of laser detectors;

[0029] The pre-processing module includes an image segmentation and fusion unit, which performs region segmentation and fusion on the image signal of each wavelength laser, obtains image information of at least two wavelengths, and performs spectral feature extraction, spectral classification and spectral matching on the spectral information of the multi-spectral laser image through the pre-processing module, for spectral identification and analysis of the welding area.

[0030] According to one aspect of the present application, it further includes an image compression module for compressing laser image data to reduce data preprocessing workload,

[0031] The image compression module includes compressing and sampling the laser image through a Hadamard matrix with random projection, reducing the data amount of the laser image, and using the equidistance property as a constraint to make the compressed image not distorted; then through a tracking algorithm based on norm minimization, the complete information of the laser image is restored, whether the quality and accuracy of the laser image meet the requirements are judged, and the solution closest to the original image is solved by using the sparse prior of the laser image.

[0032] According to one aspect of the present application, a cooperative control unit is further included for real-time tracking of the weld based on the weld parameters using deep learning and reinforcement learning according to the characteristics of the composite heat source of the laser and the arc, and assigning different attention weights according to the importance and difficulty of the weld to prioritize the key parts of the weld, dynamically adjusting the energy distribution and coupling effect of the laser and the arc, and realizing adaptive optimization of the welding process.

[0033] According to one aspect of the present application, the preprocessing module further includes a weld area quality tracking unit that sequentially receives image information from the image segmentation and fusion unit at a predetermined period, finds the base welding unit, analyzes the welding quality of the adjacent base welding unit, and adjusts the welding quality of the subsequent base welding unit based on the welding quality evaluation data of the previous welding unit.

[0034] The beneficial effects are achieved by the two groups or two working states of the laser emitting unit and the laser receiving unit, which realize multispectral laser imaging and detection of the welding area, and assist in welding, improve the information quantity and quality of the image and signal, and realize real-time tracking and optimization of the weld, realize adaptive adjustment and cooperative control of the laser and the arc, and improve the stability and reliability of the welding process. The advantages of the related art will be described below in conjunction with specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 is a structural schematic diagram of the present application.

[0036] Figure 2 is a workflow diagram of the present application. DETAILED DESCRIPTION

[0037] In order to solve the problems existing in the prior art, the applicant and the associated person have applied for CN2023230916454, narrow gap weld pool monitoring device and welding system, and CN2023226983390, narrow gap welding device, etc. In order to avoid repeated description, the above content will not be copied into the present application.

[0038] As Figure 1 shown, a narrow gap welding system is provided, comprising:

[0039] The gun head is internally provided with an eccentric gun nozzle, which is driven by the rotary mechanism to reciprocate, so that the welding wire can move in the welding operation area according to the predetermined trajectory;

[0040] The laser device comprises at least two groups or two working states of laser emitting units and laser receiving units,

[0041] At least one set or one working state of the laser emitting unit and the laser receiving unit is used to obtain a surface image of the welding working area to analyze the welding working state.

[0042] At least one set or one working state of the laser emitting unit and the laser receiving unit is used to obtain a surface image of the welding working area to analyze the welding working state.

[0043] The posture control unit is used to receive real-time working parameters including the surface image, and adjust working parameters including the laser device, the gun nozzle and the welding wire.

[0044] The above-mentioned problems in the prior art are mainly that the existing detection technology and adaptive adjustment of welding parameters cannot meet the requirements. In order to solve the above-mentioned problems in the prior art, real-time tracking monitoring and adjustment of welding parameters are performed during welding. Specifically, in the embodiment, the welding process is intelligentized and automated, the welding working state is obtained by using laser images and sensors, the welding parameters are analyzed and optimized by using data processing and machine learning, and the energy distribution and coupling effect of the laser and the electric arc are adjusted and matched by using posture control and cooperative control, so that the adaptive optimization of the welding process is realized.

[0045] By using at least two lasers, not only can the welding be formed into a composite, but also the image data of the welding area can be obtained by using the laser, the welding quality data in the welding process can be tracked in real time, and more rich quality information can be obtained by using lasers of different wavelengths, so that various information can be obtained and various effects can be realized.

[0046] According to an aspect of the present application, the laser device is two sets, including:

[0047] The first laser emitting unit and the second laser emitting unit are respectively used to emit first laser and second laser of a predetermined frequency;

[0048] The beam combiner combines the first laser and the second laser into a coaxial light;

[0049] The emitting light guide guides the coaxial light to the predetermined working area;

[0050] The receiving light guide receives the reflected laser of the working area and guides it into the laser detector and transmits it to the industrial computer.

[0051] The beam splitter is arranged between the receiving light guide and the laser detector, and the output ends of the beam splitter are respectively connected to the laser detector.

[0052] The first laser emitting units are multiple and arranged off-axis on the laser device body, and the second laser emitting unit is rotationally connected to the laser device body. According to the working state of each first laser emitting unit, the position of the second laser emitting unit is adjusted. When the second laser emitting unit rotates to a working position, it forms a predetermined angle with one of the first laser emitting units.

[0053] In this embodiment, the beam combiner is a key component for combining the first laser and the second laser into a coaxial light beam. It needs to ensure the spatiotemporal synchronization and phase matching of the two laser beams, as well as the quality and stability of the combined light. The design and production of the beam combiner need to consider parameters such as wavelength, power, pulse width, and polarization state of the laser, as well as factors such as material, structure, reflectivity, transmittance, and dispersion of the beam combiner. At the same time, the second laser emitting unit needs to rotate to the corresponding position according to the working state of each first laser emitting unit, forming a predetermined angle with the first laser emitting unit to meet the input requirements of the beam combiner. The rotation control of the second laser emitting unit needs to consider factors such as rotation speed, precision, stability, and synchronization, as well as factors such as reliability, durability, and anti-interference of the rotation mechanism. The laser detector is an important component for receiving the reflected laser from the working area and analyzing its characteristics and information. It needs to be connected to the output end of the beam splitter to achieve analysis of the reflected laser. The selection and configuration of the laser detector need to consider parameters such as wavelength, power, pulse width, and polarization state of the laser, as well as factors such as sensitivity, response speed, resolution, and dynamic range of the laser detector.

[0054] This embodiment utilizes the same set of laser devices to achieve laser processing and detection of the working area through the synergistic effect of the beam combiner, emitting light guide, receiving light guide, beam splitter, and laser detector, realizing the integration of laser processing and detection, improving work efficiency and quality, and reducing cost and risk. It solves the technical problems of separation of laser processing and detection, low efficiency, poor quality, high cost, and high risk in traditional methods.

[0055] Using two groups or two working states of laser emitting units, the wavelength, power, pulse width, and polarization state of the laser can be adjusted according to different working requirements to adapt to different materials and processes, improving the flexibility and adaptability of laser processing. Through the switching and adjustment of the two groups or two working states of laser emitting units, the multi-parameter variation of the laser is realized, solving the technical problem of laser processing being limited by a single parameter and being difficult to adapt to different materials and processes in traditional methods.

[0056] The reflected laser can be analyzed by the beam splitter and the laser detector to obtain the surface image of the working area and the processing parameters such as the shape, temperature, depth, etc., thereby enhancing the accuracy and reliability of the laser detection. Through the analysis and measurement of the beam splitter and the laser detector, multi-dimensional information of the reflected laser is obtained, and the technical problem that the laser detection is limited by single information and cannot reflect the real state of the working area in the traditional method is solved.

[0057] The industrial computer is configured with a laser data processing unit, and the laser data processing unit comprises:

[0058] A preprocessing module acquires electrical signal data in the laser detector and filters, enhances, and extracts characteristic parameters of the signal; the characteristic parameters include amplitude, frequency, phase, and polarization; the preprocessing module is a key component for acquiring electrical signal data in the laser detector and filtering, enhancing, and extracting characteristic parameters of the signal, and it needs to ensure the quality and availability of the signal, as well as the accuracy and stability of the characteristic parameters. The design and implementation of the preprocessing module need to consider the type, amplitude, frequency, phase, polarization, etc. of the signal, as well as the filtering, enhancement, extraction method, algorithm, efficiency, etc.

[0059] A preconfigured model module analyzes the characteristics of the laser beam and establishes a physical model of each stage in the propagation process of the laser beam; in order to ensure the rationality and adaptability of the physical model, as well as the consistency with the actual situation. The establishment and verification of the preconfigured model module need to consider the wavelength, power, pulse width, polarization state, etc. of the laser beam, as well as the form, equation, parameter, boundary condition, etc. of the physical model.

[0060] A data analysis module is based on the physical model to construct a data-driven laser image analysis and detection module, extract time series features and regional features, and calculate and output the welding quality information to the control module of the industrial computer;

[0061] In order to ensure the accuracy and reliability of the data analysis, as well as the effectiveness and practicality of the welding quality information. The construction and optimization of the data analysis module require engineers to consider the type, format, size, distribution, etc. of the data, as well as the analysis and detection method, algorithm, effect, performance, etc.

[0062] The weld detection module obtains the weld parameters from the laser image. In order to ensure the integrity and accuracy of the weld parameters and the correspondence with the actual weld, the acquisition and processing of the weld detection module need to consider the quality, clarity, resolution, etc. of the laser image, as well as the definition, extraction, calculation, representation, etc. of the weld parameters. In this application, the industrial computer and the attitude adjustment unit are set as two modules, and if the cost budget of the device is higher, the image processing module can be directly configured on the machine. The advantage of using an industrial computer is that a higher industrial computer can serve multiple welding systems, thereby reducing the cost of the system.

[0063] This embodiment realizes the automation and intelligentization of laser image analysis and detection of the welding work area, improves the work efficiency and quality, reduces the possibility of manual intervention and error, realizes the integration of laser image analysis and detection of the welding work area, solves the technical problems of separation of laser processing and detection, low efficiency, poor quality, high cost and high risk in traditional methods. High-quality processing and efficient analysis of laser images are realized, which improves the precision and reliability of laser image analysis and detection, and solves the technical problems of poor laser image quality, many noise points and large errors in traditional methods. Not only can the time sequence features and regional features be extracted, but also the welding quality information such as weld shape, depth and temperature can be calculated, which provides a basis for the optimization and control of the welding process, and can also be used for the evaluation and monitoring of the welding quality. Multi-dimensional information extraction and welding quality information calculation of laser images are realized, which solves the technical problems of single laser image information and lack of welding quality information in traditional methods.

[0064] In this embodiment, by combining physical models with data analysis such as neural network analysis modules, the problem that existing neural network modules cannot reflect the actual physical process is solved, and the detection efficiency and quality are improved.

[0065] According to one aspect of the present application, the laser device is provided with an adaptive laser modulation unit for calculating the optimal working parameters according to the pre-stored historical data, and controlling the working parameters of the first and second laser emitting units according to the optimal working parameters;

[0066] The historical data includes at least welding parameters, weld shape, welding materials and environmental temperature, and the acquisition and storage of the historical data need to consider the source, quality, magnitude, format, safety, etc. of the data; the optimal working parameters at least include wavelength, power, pulse width and frequency, and the calculation and optimization of the optimal working parameters need to consider the calculation method, algorithm, efficiency, accuracy, etc.

[0067] In this embodiment, the wavelength, power, pulse width and frequency of the first and second laser emitting units are adjusted in real time by the adaptive laser modulation unit to adapt to different welding conditions and requirements. The adaptive laser modulation unit includes a data processing module and an optical modulation module. The data processing module is responsible for extracting relevant welding parameters, weld shape, welding material and environmental temperature information from pre-stored historical data, and calculating the optimal working parameters according to the predetermined algorithm and model. The optical modulation module is responsible for adjusting the working parameters of the first and second laser emitting units according to the output of the data processing module to achieve the best welding effect. The optical modulation module can use different ways to adjust the laser parameters, such as electro-optical modulator, acousto-optical modulator, liquid crystal spatial light modulator, etc. The electro-optical modulator uses the influence of electric field on the refractive index of the material to realize the frequency conversion and intensity modulation of laser. By changing the voltage and frequency of the electro-optical modulator, the wavelength and power of the laser can be accurately controlled. The acousto-optical modulator uses the influence of sound wave on the refractive index of the material to realize the time conversion and phase modulation of laser. By changing the sound pressure and sound frequency of the acousto-optical modulator, the pulse width and frequency of the laser can be accurately controlled. The data processing module in the industrial computer can calculate the welding quality information such as weld shape, depth, temperature according to the characteristics of reflected laser such as amplitude, frequency, phase and polarization, and compare it with the pre-set target value. If there is a deviation, the adaptive laser modulation unit can be adjusted through feedback control to optimize and control the welding quality.

[0068] By using the adaptive laser modulation unit, the optimal working parameters are calculated according to the pre-stored historical data and the current welding conditions and requirements, and the working parameters of the laser emitting unit are controlled according to the optimal working parameters, so that the laser beam can adapt to different welding materials, weld shapes, environmental temperatures and other factors, improving the adaptability and flexibility of laser welding. According to the optimal working parameters, the working parameters of the laser emitting unit are adjusted and controlled in real time, so that the laser beam can achieve the best welding effect, improving the efficiency and quality of laser welding. According to the pre-stored historical data, the number and time of experiments and verifications of laser welding are reduced, and the cost and risk of laser welding are reduced. By quickly calculating the optimal working parameters such as wavelength, power, pulse width and frequency, the technical problem of needing a large number of experiments and verifications in traditional methods is solved. By using optical modulators such as electro-optical modulators and acousto-optical modulators, the working parameters of the laser emitting unit such as wavelength, power, pulse width and frequency are accurately controlled, solving the technical problem of difficult adjustment and control of the working parameters of the laser emitting unit in traditional methods. The coaxial light is guided to the predetermined working area by the emitting light guide, solving the technical problem of difficult beam combination and guidance of laser beam in traditional methods.

[0069] According to an aspect of the present application, the laser detector is at least two, respectively used for receiving laser of different wavelengths and converting into electrical signal;

[0070] The number of the pre-processing module is the same as the laser detector;

[0071] The pre-processing module includes an image segmentation and fusion unit, which performs region segmentation and fusion on the image signal of each wavelength laser, obtains image information of at least two wavelengths, and performs spectral feature extraction, spectral classification and spectral matching through the pre-processing module based on the spectral information of the multi-spectral laser image, to complete spectral identification and analysis of the welding area.

[0072] By using the spectral information of the multi-spectral laser image, spectral feature extraction, spectral classification and spectral matching are performed through the pre-processing module to complete spectral identification and analysis of the welding area, which can obtain spectral information of multiple wavelengths of the welding area at the same time, improve the accuracy and efficiency of spectral identification and analysis, improve the richness and integrity of spectral information, and present the spectral information of the welding area in the form of image through spectral identification and analysis of the welding area, improve the visualization and interpretability of spectral information, solve the technical problems of ambiguity and instability of spectral classification in traditional methods, extract features such as spectral curve, spectral peak and spectral width from image information, which can reflect the spectral characteristics of the welding area, solve the technical problems of difficulty and inaccuracy of spectral feature extraction in traditional methods, and use the results of spectral classification to compare the spectrum of the welding area with the pre-established spectral library according to predetermined criteria, find the most similar spectrum, and judge the physical, chemical and mechanical properties of the welding area, solve the technical problems of complexity and unreliability of spectral matching in traditional methods.

[0073] According to an aspect of the present application, it further includes an image compression module for compressing laser image data to reduce data pre-processing workload,

[0074] The image compression module includes compressing and sampling the laser image through a Hadamard matrix with random projection, reducing the data amount of the laser image, and using the equidistance property as a constraint to make the compressed image not distorted; then the complete information of the laser image is recovered through a tracking algorithm based on norm minimization to judge whether the quality and accuracy of the laser image meet the requirements, and the solution closest to the original image is solved by using the sparse prior of the laser image.

[0075] The laser image is compressed and sampled using Hadamard matrix, which can greatly reduce the data volume and storage space of the laser image, and improve the efficiency of data transmission and processing. The isometric property is used as a constraint to ensure that the compressed image is not distorted, and the tracking algorithm is used to recover the complete information of the laser image, so that the recovered image is as close as possible to the original image, ensuring the integrity and distortionlessness of the laser image. Through the Hadamard matrix, efficient compression sampling and compressed sensing are realized, which can greatly reduce the data volume of the image without losing image information, solving the complexity and low efficiency of compression sampling and compressed sensing in traditional methods. The over-tracking algorithm realizes high-quality image recovery and reconstruction, which can use the sparsity prior of the laser image to solve the solution closest to the original image, solving the precision and stability problems of image recovery and reconstruction in traditional methods. By expanding the spatial information of the laser image in the time domain, the two-dimensional image information is recovered from the one-dimensional time domain signal, solving the complexity and cost problems of the imaging system in traditional methods.

[0076] According to one aspect of the present application, a cooperative control unit is further included for tracking the weld in real time based on the weld parameters using deep learning and reinforcement learning according to the characteristics of the combined heat source of laser and arc, and assigning different attention weights according to the importance and difficulty of the weld to preferentially process the key parts of the weld and dynamically adjust the energy distribution and coupling effect of laser and arc to realize adaptive optimization of the welding process.

[0077] The cooperative control unit is used to cooperatively control and optimize the combined heat source of laser and arc, which can dynamically adjust the energy distribution and coupling effect of laser and arc according to the real-time state and characteristics of the weld to realize adaptive optimization of the welding process and improve the welding quality and efficiency. The cooperative control and optimization of the combined heat source of laser and arc can be realized, solving the difficulty and complexity of the control and optimization of the combined heat source of laser and arc in traditional methods. The real-time tracking and prediction of the weld parameters are realized using deep learning and reinforcement learning algorithms, different attention weights are assigned according to the importance and difficulty of the weld to preferentially process the key parts of the weld, and intelligent decision-making and control of the welding process are realized to improve the intelligent and automated level of welding. The real-time tracking and prediction of the weld parameters solve the inaccuracy and timeliness problems of weld parameter detection and analysis in traditional methods. Through the deep learning and reinforcement learning algorithms, adaptive optimization of the welding process can be realized, solving the fixed and low efficiency problems of the welding process in traditional methods.

[0078] In some embodiments, the following process can be used:

[0079] S1: Data acquisition is completed through the above steps. Real-time acquisition of parameters such as power, speed, temperature, current, voltage, etc. of the combined heat source of laser and electric arc, as well as the shape, position, width, depth, etc. of the weld. The collected data is converted into digital signals and stored in the database.

[0080] S2: Data preprocessing is completed through the above steps. The collected data is cleaned, filtered, normalized, and dimensionally reduced, etc. to remove noise, outliers, redundant information, etc. to improve the quality and usability of the data. The preprocessed data is divided into training set, validation set and test set for subsequent model training and evaluation.

[0081] S3: Data analysis. Using deep learning and reinforcement learning, build and train the model to achieve the following functions:

[0082] S31: Use convolutional neural network (CNN) and recurrent neural network (RNN) combined with physical model to form PINN module to realize real-time identification and prediction of the shape, position, width, depth, etc. of the weld, and realize real-time tracking of the weld.

[0083] S32: Use attention mechanism to assign different attention weights according to the importance and difficulty of the weld, and prioritize processing of the key parts of the weld to improve welding quality and efficiency.

[0084] S33: Use reinforcement learning to dynamically adjust the energy distribution and coupling effect of laser and electric arc according to the real-time state and target state of the weld to realize adaptive optimization of the welding process.

[0085] S4: Data visualization and application. Use graphical interface to display the results of data analysis in the form of charts, curves, images, etc. to users to facilitate monitoring and control of the welding process and evaluation of the welding effect. Apply the results of data analysis to the control and optimization of the welding process to achieve accurate tracking of the weld and high-quality welding by adjusting the parameters of laser and electric arc.

[0086] In one embodiment, the PINN module can be specifically:

[0087] According to the complexity and characteristics of the partial differential equations in the welding process, the neural network structure is selected, including ResNet or MLP; according to the relationship between the space-time coordinates and the state variables in the welding process, the input and output of the network are defined, the space-time coordinates in the welding process are taken as the input of the network, and the state variables in the welding process are taken as the output of the network. According to the error between the network output and the actual data, and the error between the network output and the physical model, the cross-entropy is used as the loss function to measure the performance and accuracy of the network. According to the complexity of the network and the characteristics of the data, the weight decay is selected as the regularization technique to prevent overfitting or underfitting of the network and improve the generalization ability of the network.

[0088] According to the structure of the network and the characteristics of the loss function, the stochastic gradient descent method is selected as the optimization algorithm to update the parameters of the network to make the loss function reach the minimum value; according to the amount and quality of the data, and the limitation of the computing resources, the training parameters such as learning rate, batch size, iteration number, and validation frequency are selected. According to the optimization algorithm and the training parameters, the training process is executed to update the parameters of the network constantly to make the loss function reach the minimum value. In the training process, the validation set or test set can be used to evaluate the performance and accuracy of the network, and to detect whether the network has overfitting or underfitting. According to the results of the training process, the parameters of the trained network and the related performance and accuracy indicators are saved for subsequent application and analysis.

[0089] In another embodiment of the present application, the preprocessing module further comprises a welding area quality tracking unit, which sequentially receives the image information of the image segmentation and fusion unit according to a predetermined period, finds the base welding unit, analyzes the welding quality of the adjacent base welding unit, and adjusts the welding quality of the subsequent base welding unit through the welding quality evaluation data of the previous welding unit. In this embodiment, the welding is characterized by periodic oscillation welding, and in many welding processes, there are arc-shaped, fish-scale-shaped, and periodically repeated welding areas. Because heat and light have a great impact on image acquisition during welding, the operation mode of this embodiment is provided, i.e., the image is segmented according to time and area during a welding process. The welding area of the previous welding is moved to a non-welding work area after the welding is completed. Therefore, according to the previously collected image information and subsequent quality evaluation, the quality evaluation parameters of the welding work area can be adjusted to improve the evaluation quality. For example, the first to 200th welding areas periodically appear. When welding, the first welding is completed and is moved away from the welding area, thereby facilitating quality evaluation. Moreover, this area has collected corresponding image and spectral information during the welding operation, so that the real-time evaluation parameters can be checked through subsequent quality evaluation, thereby ensuring that the real-time evaluation of the subsequent welding operation area is more accurate. The part that is moved away from the welding area can be achieved by other image or ultrasonic acquisition units.

[0090] The preferred embodiments of the present application are described in detail above, but the present application is not limited to the specific details of the above-described embodiments. Within the technical concept of the present application, various equivalent transformations of the technical solutions of the present application can be made, and these equivalent transformations all belong to the protection scope of the present application.

Claims

1. A narrow gap welding system characterized by, The gun head is internally provided with an eccentric gun nozzle, which is driven by a rotating mechanism to reciprocate so as to move the welding wire along a predetermined trajectory in a welding work area. The laser device comprises at least two groups or two working states of laser emitting units and laser receiving units, At least one group or one working state of the laser emitting units and the laser receiving units is used to obtain a surface image of the welding work area to analyze the welding working state; At least one other group or one working state of the laser emitting units and the laser receiving units acts on the molten pool to assist welding; The two groups or two working states of the laser emitting units can be switched; The attitude control unit is used to receive real-time working parameters including the surface image and adjust working parameters of the laser device, the gun nozzle and the welding wire. The laser device comprises:

2. The narrow gap welding system of claim 1, wherein, The first laser emitting unit and the second laser emitting unit are respectively used to emit first laser and second laser of a predetermined frequency; The beam combiner combines the first laser and the second laser into a coaxial light; The emitting light guide directs the coaxial light to a predetermined working area; The receiving light guide receives the reflected laser of the working area and guides it into the laser detector and transmits it to the industrial computer. The beam splitter is further arranged between the receiving light guide and the laser detector, and the output ends of the beam splitter are respectively connected to the laser detector.

3. The narrow gap welding system of claim 2, wherein, The first laser emitting unit is multiple and arranged off-axis in the laser device body, and the second laser emitting unit is rotationally connected to the laser device body. According to the working state of each first laser emitting unit, the position of the second laser emitting unit is adjusted. When the second laser emitting unit rotates to a working position, it forms a predetermined angle with one of the first laser emitting units.

4. The narrow gap welding system of claim 3, wherein, The industrial computer is configured with a laser data processing unit, which comprises:

5. The narrow gap welding system of claim 4, wherein, The preprocessing module obtains electrical signal data in the laser detector and performs filtering, enhancement and extraction of signal characteristic parameters; the characteristic parameters include amplitude, frequency, phase and polarization; The preconfigured model module analyzes the characteristics of the laser beam and establishes a physical model of each stage in the propagation process of the laser beam; The data analysis module is based on the physical model to construct a data-driven laser image analysis and detection module, extracts time sequence features and region features, calculates and outputs welding quality information to the control module of the industrial computer; The weld detection module obtains weld parameters from the laser image. The laser device is provided with an adaptive laser modulation unit for calculating optimal working parameters according to pre-stored historical data and controlling the working parameters of the first laser emitting unit and the second laser emitting unit according to the optimal working parameters; 6. The narrow gap welding system of claim 5, wherein, The historical data includes at least welding parameters, weld shape, welding materials and environmental temperature; the optimal working parameters include at least wavelength, power, pulse width and frequency. The laser detector is at least two, respectively used to receive laser of different wavelengths and convert into electrical signal; 7. The narrow gap welding system of claim 6, wherein, The number of the preprocessing module is the same as that of the laser detector; ​ The preprocessing module comprises an image segmentation and fusion unit, which performs regional segmentation and fusion on the image signals of each wavelength laser to obtain image information of at least two wavelengths, and performs spectral feature extraction, spectral classification and spectral matching on the spectral information of the multispectral laser image through the preprocessing module to perform spectral recognition and analysis on the welding area.

8. The narrow gap welding system of claim 7, wherein, An image compression module is further included to compress the laser image data and reduce the data preprocessing workload. The image compression module comprises a Hadamard matrix with random projection to compress and sample the laser image, reduce the data volume of the laser image, and use the equidistance property as a constraint to ensure that the compressed image is not distorted; then a tracking algorithm based on norm minimization is used to restore the complete information of the laser image, determine whether the quality and precision of the laser image meet the requirements, and use the sparse prior of the laser image to solve the solution closest to the original image.

9. The narrow gap welding system of claim 8, wherein, A cooperative control unit is further included to use deep learning and reinforcement learning to track the weld in real time based on the weld parameters according to the characteristics of the combined heat source of the laser and the electric arc, assign different attention weights according to the importance and difficulty of the weld, prioritize the key parts of the weld, dynamically adjust the energy distribution and coupling effect of the laser and the electric arc, and realize adaptive optimization of the welding process.

10. The narrow gap welding system of claim 8, wherein, The preprocessing module further comprises a welding area quality tracking unit that sequentially receives image information from the image segmentation and fusion unit at a predetermined period, finds the basic welding unit, analyzes the welding quality of the adjacent basic welding unit, and adjusts the welding quality of the subsequent basic welding unit based on the welding quality evaluation data of the previous welding unit.

Images