Weld joint quality detection method and system for intelligent laser welding

Abstract

The invention discloses a welding seam quality detection method and system for intelligent laser welding, and relates to the technical field of welding technology and quality detection.The method comprises the following steps that multi-modal signal data of the surface of a welding seam are obtained, and a multi-modal feature extraction network is defined; an initial feature representation is generated based on the multi-modal feature extraction network in combination with the multi-modal signal data. According to the method, the parameter dynamic adjustment module is constructed through the transfer learning framework and the evolutionary algorithm, efficient conversion between different materials can be achieved, and the flexibility and generalization ability of a detection system are improved. The variational Gaussian mixture model is applied in combination with Bayesian inference, so that the detection system can extract a valuable variation trend from complex data distribution of the layered feedback adjustment unit, core parameters are dynamically adjusted, and the adaptability of the detection system to different material characteristics is further enhanced.

Description

Technical Field

[0001] The present invention relates to the field of welding technology and quality detection technology, and specifically to a weld quality detection method and system for laser intelligent welding. Background Art

[0002] In industrial manufacturing, laser intelligent welding, with its advantages of high precision and high efficiency, is widely used in industries such as automotive and aerospace. Weld quality directly impacts product safety, durability, and service life, making weld quality inspection crucial. Traditional manual visual inspection or conventional nondestructive testing techniques, such as ultrasonic and X-ray inspection, suffer from low efficiency, high cost, reliance on experience, and difficulty retaining data. These methods are no longer able to fully meet the demands of modern industry for high-precision, intelligent testing.

[0003] Although a variety of laser weld inspection technologies have been developed, such as inspection systems that integrate 3D laser scanning and AI analysis technology, which can achieve millimeter-level 3D modeling, dual-mode inspection, and intelligent defect classification, in actual applications, the inspection system still presents significant challenges in multi-material adaptability when faced with different metal materials. Different metals, such as copper, aluminum, and steel, have significant differences in reflectivity and thermal properties. For example, the high reflectivity of copper will cause a large amount of laser signal reflection, which will weaken the effective signal received by the sensor and affect the accurate acquisition of weld information. Aluminum has a high thermal conductivity and heat diffusion is rapid during welding. The temperature field and thermal stress distribution at the weld are significantly different from those of other metals. The existing detection thresholds and algorithms have limited adaptability to such differences, resulting in reduced detection accuracy and stability, thereby affecting the consistent assessment of the weld quality of various metal materials and the high-precision inspection effect. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the present invention provides a weld quality detection method and system for laser intelligent welding, which is used to solve the problems of insufficient adaptability to multiple materials and decreased detection accuracy and stability in the existing technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for detecting weld quality of laser intelligent welding, comprising the following steps:

[0006] Acquiring multimodal signal data of the weld surface, defining a multimodal feature extraction network, and generating an initial feature representation based on the multimodal feature extraction network and the multimodal signal data;

[0007] Using a cross-modal information fusion mechanism and an adaptive compensation algorithm to enhance the correlation between the optical signal and the thermal field signal in the initial feature representation, and determining a multimodal perception module;

[0008] Based on the multimodal perception module, a hierarchical feedback adjustment unit is constructed, and the hierarchical feedback adjustment unit is used to process feedback signals from different levels to generate an optimized detection path;

[0009] Based on the hierarchical feedback regulation unit, a parameter dynamic adjustment module is constructed using a transfer learning framework and an evolutionary algorithm, and core parameters are adjusted based on the parameter dynamic adjustment module and the optimized detection path to generate an adaptive detection strategy;

[0010] Based on the optimized detection path and the adaptive detection strategy, an incremental feature expansion module is constructed using a structured analysis technology based on graph embedding. Based on the incremental feature expansion module, the multimodal feature extraction network, the multimodal perception module, the hierarchical feedback adjustment unit and the parameter dynamic adjustment module, a weld quality detection system is constructed.

[0011] Furthermore, the hierarchical feedback adjustment unit is based on the transfer learning framework and the evolutionary algorithm to construct a parameter dynamic adjustment module, and the core parameters are adjusted based on the parameter dynamic adjustment module and the optimized detection path to generate an adaptive detection strategy, including the following steps:

[0012] Using a transfer learning framework and an evolutionary algorithm, an initial parameter dynamic adjustment module is constructed based on the hierarchical feedback adjustment unit, and the adaptability of the initial parameter dynamic adjustment module between different materials is optimized by a differential evolution algorithm and a genetic algorithm;

[0013] Based on the optimized detection path, a variational Gaussian mixture model is used to analyze the change pattern from the data distribution of feedback signals at different levels in the hierarchical feedback adjustment unit, and the core parameters in the initial parameter dynamic adjustment module are dynamically adjusted through Bayesian inference to generate a target parameter dynamic adjustment module;

[0014] The context-sensitive mechanism is used to enable the target parameter dynamic adjustment module to adjust the detection strategy according to the characteristics of different materials to generate an adaptive detection strategy.

[0015] Furthermore, the context-sensitive mechanism is used to enable the target parameter dynamic adjustment module to adjust the detection strategy according to the characteristics of different materials to generate an adaptive detection strategy, including the following steps:

[0016] Utilizing a context-sensitive mechanism combined with graph embedding technology and an attention allocation mechanism, key properties are analyzed from the characteristics of different materials in the hierarchical feedback regulation unit, including reflectivity distribution, thermal conductivity differences, and thermal stress distribution characteristics;

[0017] Based on the key attributes, the detection threshold in the target parameter dynamic adjustment module is dynamically adjusted in combination with Bayesian inference and Gaussian process modeling to generate a preliminary adaptive detection strategy;

[0018] optimizing the preliminary adaptive detection strategy through reinforcement learning using a time series decomposition model to generate an optimized detection strategy;

[0019] Searching for an optimal detection strategy from the optimized detection strategies by combining a particle swarm optimization algorithm and a simulated annealing algorithm;

[0020] An active sampling mechanism is introduced to select target samples from the optimal detection strategy to generate an adaptive detection strategy.

[0021] Furthermore, the method of using a context-sensitive mechanism in combination with graph embedding technology and an attention allocation mechanism to analyze key attributes from the characteristics of different materials in the hierarchical feedback regulation unit includes the following steps:

[0022] Using signal preprocessing technology to collect and clean the characteristics of different materials in the hierarchical feedback regulation unit to obtain material characteristic data of the hierarchical feedback regulation unit;

[0023] A context-sensitive mechanism is introduced to combine principal component analysis and independent component analysis to perform preliminary screening and dimensionality reduction on the material characteristic data to obtain target material characteristic information;

[0024] Applying graph embedding technology and complex network modeling to analyze the correlation between materials from the target material characteristic information to construct a material relationship graph;

[0025] The attention allocation mechanism is combined with deep reinforcement learning to highlight the important attributes in the material relationship graph, and the key attribute weight distribution result is obtained;

[0026] The material relationship diagram and the key attribute weight distribution result are analyzed to extract key attributes.

[0027] Furthermore, based on the optimized detection path, a variational Gaussian mixture model is used to analyze the change pattern from the data distribution of feedback signals at different levels in the hierarchical feedback adjustment unit, and the core parameters in the initial parameter dynamic adjustment module are dynamically adjusted through Bayesian inference to generate a target parameter dynamic adjustment module, including the following steps:

[0028] Based on the optimized detection path, a variational Gaussian mixture model and statistical modeling tools are used to model the data distribution of feedback signals at different levels in the hierarchical feedback regulation unit, and potential patterns and change trends are analyzed from the feedback signals to generate data distribution change rules;

[0029] Based on the data distribution change law, the core parameters in the initial parameter dynamic adjustment module are dynamically adjusted through Bayesian inference, Gaussian process modeling and random forest to generate a target parameter dynamic adjustment module.

[0030] Furthermore, the method of enhancing the correlation between the optical signal and the thermal field signal in the initial feature representation by using a cross-modal information fusion mechanism and an adaptive compensation algorithm to determine a multimodal perception module includes the following steps:

[0031] An initial multimodal feature extraction network is defined using a hierarchical graph embedding network, and an initial feature representation is analyzed from the intrinsic relationship between the optical signal and the thermal field signal of the initial multimodal feature extraction network;

[0032] Based on the initial feature representation, a cross-modal information fusion mechanism is introduced in combination with a bidirectional gated recurrent unit to perform joint encoding processing on the optical signal and the thermal field signal in the initial feature representation to obtain a multimodal information representation;

[0033] constructing a generative adversarial network based on the multimodal information representation, and analyzing key data from an adversarial training process between a generator and a discriminator in the generative adversarial network;

[0034] The initial multimodal feature extraction network is optimized based on the key data in combination with a variational Gaussian mixture model to obtain a target multimodal feature extraction network, and a multimodal perception module is determined based on the target multimodal feature extraction network.

[0035] Furthermore, based on the optimized detection path and the adaptive detection strategy, an incremental feature expansion module is constructed using a structured analysis technology based on graph embedding, including the following steps:

[0036] Determining incremental update rules using the optimized detection path and the adaptive detection strategy in combination with time series modeling and Bayesian inference;

[0037] Based on the incremental update rule, the multimodal data in the multimodal perception module is modeled and processed using a structured analysis technique based on graph embedding combined with a hierarchical clustering algorithm to obtain a structured representation of the multimodal data;

[0038] encoding data obtained from the hierarchical feedback regulation unit according to the multimodal data structured representation in combination with a dynamic graph embedding network and a variational Gaussian mixture model, and inputting the data into the multimodal data structured representation to output an updated feature representation;

[0039] Applying an attention allocation mechanism combined with deep reinforcement learning to highlight key attributes in the updated feature representation to obtain an optimized feature representation;

[0040] Combining reinforcement learning and particle swarm optimization algorithms to optimize and adjust the parameters in the incremental update process of the parameter dynamic adjustment module to generate an optimized parameter configuration;

[0041] An incremental feature expansion module is generated based on the optimized feature representation and the optimized parameter configuration.

[0042] The present invention also provides a laser intelligent welding weld quality detection system, comprising:

[0043] An acquisition module is used to obtain multimodal signal data of the weld surface, define a multimodal feature extraction network, and generate an initial feature representation based on the multimodal feature extraction network and the multimodal signal data;

[0044] an enhancement module, configured to enhance the correlation between the optical signal and the thermal field signal in the initial feature representation by utilizing a cross-modal information fusion mechanism and an adaptive compensation algorithm, and determine a multimodal perception module;

[0045] A construction module, configured to construct a hierarchical feedback adjustment unit based on the multimodal perception module, and utilize the hierarchical feedback adjustment unit to process feedback signals from different levels to generate an optimized detection path;

[0046] An adjustment module, configured to construct a parameter dynamic adjustment module based on the hierarchical feedback adjustment unit using a transfer learning framework and an evolutionary algorithm, and to adjust core parameters based on the parameter dynamic adjustment module and the optimized detection path to generate an adaptive detection strategy;

[0047] A module is used to construct an incremental feature expansion module based on the optimized detection path and the adaptive detection strategy using a structured analysis technology based on graph embedding, and a weld quality detection system is constructed based on the incremental feature expansion module, the multimodal feature extraction network, the multimodal perception module, the hierarchical feedback adjustment unit and the parameter dynamic adjustment module.

[0048] Furthermore, the acquisition module includes a high-resolution optical camera and an infrared thermal imager, which are used to capture the optical reflection characteristics and thermal field distribution characteristics of the weld surface respectively.

[0049] The present invention further provides a computing device, comprising a processor and a memory, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute any one of the above methods.

[0050] Beneficial effects

[0051] 1. The present invention uses a dynamic parameter adjustment module constructed through a transfer learning framework and an evolutionary algorithm, which can efficiently convert between different materials and improve the flexibility and generalization ability of the detection system. The application of the variational Gaussian mixture model combined with Bayesian inference enables the detection system to extract valuable change trends from the complex data distribution of the hierarchical feedback regulation unit, dynamically adjust the core parameters, and further enhance the adaptability of the detection system to different material properties. In addition, the application of the context-sensitive mechanism enables the detection system to flexibly adjust the detection strategy according to the specific material characteristics and generate the optimal detection strategy, thereby ensuring the high performance and rapid convergence of the detection system in various application scenarios. In summary, this method effectively solves the problems of insufficient multi-material adaptability, static parameter configuration, and imperfect feedback mechanism in the existing scheme, and provides a more intelligent and efficient weld quality detection solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 Schematic diagram of the module structure of the weld quality detection system for laser intelligent welding in an embodiment of the present invention;

[0053] Figure 2 Schematic diagram of the working principle of the multimodal perception module in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0055] The present invention provides a method and system for detecting weld quality of laser intelligent welding. Figure 1 and attached Figure 2 Provide detailed explanation.

[0056] Figure 1 The module structure diagram of the weld quality inspection system is shown, which includes the acquisition module, enhancement module, construction module, adjustment module and utilization module. These modules realize the specific process of weld quality inspection through logical connection relationships. Figure 2 This is a schematic diagram of the working principle of the multimodal perception module, which describes in detail the joint encoding processing process of optical signals and thermal field signals under the cross-modal information fusion mechanism and the role of generative adversarial networks in optimizing feature representation.

[0057] In practical applications, first, the acquisition module obtains the multi-modal signal data of the weld surface, including optical signals and thermal field signals. The acquisition module monitors the weld surface in real time through a sensor array, which includes a high-resolution optical camera and an infrared thermal imager, respectively used to capture the optical reflection characteristics and thermal field distribution characteristics of the weld surface. The collected multi-modal signal data is transmitted to the enhancement module for processing. The acquisition module and the enhancement module are connected through a gigabit Ethernet interface (transmission rate ≥ 1 Gbps, delay ≤ 10 ms), and at the same time, the CRC check mechanism is used to ensure that the signal data is not lost and error code during transmission, ensuring real-time and integrity.

[0058] The enhancement module processes the collected multi-modal signal data based on a multi-modal feature extraction network to generate an initial feature representation. The multi-modal feature extraction network adopts a hierarchical design: a convolutional neural network (CNN) containing 3 convolutional layers (convolution kernel size 3×3, 5×5, 3×3 in turn, step size 1) and 2 pooling layers (maximum pooling, 2×2), used to extract the spatial features of the optical signal (such as crack edge, pore morphology); an autoencoder containing an input layer (dimension = number of thermal field signal sampling points × time step), a hidden layer (3 layers, dimension 512, 256, 128 in turn) and an output layer, used to extract the time series features of the thermal field signal (such as temperature gradient change rate). The enhancement module further introduces a cross-modal information fusion mechanism, which combines a bidirectional gated recurrent unit to jointly encode and process the optical signal and the thermal field signal in the initial feature representation to obtain a multi-modal information representation. The cross-modal information fusion mechanism highlights key attributes through an attention distribution mechanism to ensure that the correlation between the optical signal and the thermal field signal is fully explored. Attention weight calculation (based on scaled dot-product attention): where q i is the query vector (such as the optical signal feature), k j is the key vector (such as the thermal field signal feature), d k is the key vector dimension, and a i,j is the attention weight of the i-th query to the j-th key.

[0059] Subsequently, a generative adversarial network is introduced to optimize the initial multi-modal feature extraction network, generate a target multi-modal feature extraction network, and determine a multi-modal perception module. The generator in the generative adversarial network is responsible for generating latent feature representations, and the discriminator evaluates the generated feature representations, continuously optimizing the quality of the feature representations through an adversarial training process.

[0060] The construction module constructs a hierarchical feedback regulation unit based on the multimodal perception module, and uses the hierarchical feedback regulation unit to process feedback signals from different levels to generate an optimized detection path. The hierarchical feedback regulation unit adopts a three-layer tree structure design: the first layer (root node) receives the original multimodal signal summary data; the second layer (6 sub-nodes) corresponds to optical reflection intensity, optical texture distribution, thermal field peak temperature, thermal field diffusion rate, stress concentration coefficient, and stress distribution uniformity; the third layer (12 sub-nodes) is the subdivision dimension of each indicator in the second layer (such as the thermal field diffusion rate is subdivided into lateral diffusion rate and longitudinal diffusion rate). Each layer of nodes aggregates the lower layer signals through a weighted average algorithm (weights are dynamically assigned based on feature importance) to generate upper layer feedback results. The hierarchical feedback regulation unit analyzes the data distribution of the feedback signal through statistical modeling tools, identifies potential patterns and change trends, and generates an optimized detection path based on these patterns and trends. Evidence lower bound (ELBO) objective function of the variational Gaussian mixture model: Among them, X is the observed data (feedback signal), Z is the latent variable (cluster label), θ is the model parameter (mean, covariance), q(Z,θ) is the variational distribution, and the goal is to maximize To approximate the true posterior p(Z,θ|X). The construction module and the adjustment module are connected via a data bus to ensure that the output of the hierarchical feedback adjustment unit can be transmitted to the parameter dynamic adjustment module in real time.

[0061] The adjustment module uses a transfer learning framework and evolutionary algorithms to construct a dynamic parameter adjustment module based on a hierarchical feedback regulation unit. This module then adjusts core parameters based on the optimized detection path to generate an adaptive detection strategy. The construction of the dynamic parameter adjustment module is divided into three steps.

[0062] In the first step, the initial parameter dynamic adjustment module is constructed based on the hierarchical feedback regulation unit using the transfer learning framework and evolutionary algorithm. The differential evolution algorithm and genetic algorithm are used to optimize the adaptability of this module between different materials. The mutation operation formula of the differential evolution algorithm is: i (t) = x r1 (t)+F·(x r2 (t)-x r3 (t)), where v i (t) is the ith mutant individual in the tth generation, x r1 ,x r2 ,x r3 are three randomly selected individuals, and F∈[0,2] is the scaling factor (here F=0.8). Crossover operation formula: Among them, CR∈[0,1] is the crossover probability (here CR=0.6), j rand is a random dimension.

[0063] In the second step, based on the optimized detection path, the variational Gaussian mixture model is used to extract the change rules from the data distribution of feedback signals at different levels in the hierarchical feedback regulation unit, and the core parameters in the initial parameter dynamic adjustment module are dynamically adjusted through Bayesian inference to generate the target parameter dynamic adjustment module.

[0064] In the third step, the context-sensitive mechanism is used to enable the target parameter dynamic adjustment module to adjust the detection strategy according to the characteristics of different materials to generate an adaptive detection strategy. The context-sensitive mechanism combines graph embedding technology and attention allocation mechanism to analyze key attributes from the characteristics of different materials in the hierarchical feedback adjustment unit, including reflectivity distribution, thermal conductivity differences and thermal stress distribution characteristics. These key attributes are used to dynamically adjust the detection threshold through Bayesian inference and Gaussian process modeling to generate a preliminary adaptive detection strategy. The preliminary adaptive detection strategy is further optimized through time series decomposition model and reinforcement learning, and finally the particle swarm optimization algorithm and simulated annealing algorithm are combined to search for the optimal detection strategy. The speed and position update formula of the particle swarm optimization algorithm are:

[0065]

[0066] Among them, w is the inertia weight (here w = 0.729), c_1 = c_2 = 1.494 is the acceleration coefficient, r1, r2∈[0,1] are random numbers, pbest i is the individual optimal position, gbest is the global optimal position, and k is the number of iterations.

[0067] The module uses graph embedding-based structured analysis technology to construct an incremental feature expansion module based on an optimized detection path and adaptive detection strategy. The construction process of the incremental feature expansion module includes the following steps. First, the optimized detection path and adaptive detection strategy are combined with time series modeling (using an ARIMA model with order p=3 and q=2) and Bayesian inference to determine the incremental update rule: an incremental update is triggered when the distribution difference (KL divergence) between the newly collected data and the historical data is ≥0.1. The update frequency is every 50ms, and each update retains only the core features of the first 1000 historical data (filtered by L1 regularization) to reduce redundancy. Second, based on the incremental update rule, the multimodal data in the multimodal perception module is modeled and processed using graph embedding-based structured analysis technology combined with a hierarchical clustering algorithm to obtain a structured representation of the multimodal data. Third, based on the structured representation of the multimodal data, a dynamic graph embedding network and a variational Gaussian mixture model are used to encode the data obtained from the hierarchical feedback regulation unit. The data is then input into the structured representation of the multimodal data to output the updated feature representation. Subsequently, an attention allocation mechanism combined with deep reinforcement learning is applied to highlight key attributes in the updated feature representation, resulting in an optimized feature representation. Finally, reinforcement learning and a particle swarm optimization algorithm are combined to optimize the parameters during the incremental update process in the dynamic parameter adjustment module, generating an optimized parameter configuration. Based on this optimized feature representation and parameter configuration, an incremental feature expansion module is then generated.

[0068] In actual application scenarios, the collaborative workflow of the above modules is as follows. After the multimodal signal data of the weld surface is acquired by the acquisition module, the data is first transmitted to the enhancement module for feature extraction and cross-modal information fusion processing to generate a multimodal perception module. Subsequently, the construction module constructs a hierarchical feedback adjustment unit based on the multimodal perception module, generates an optimized detection path and passes it to the adjustment module. The adjustment module uses the transfer learning framework and evolutionary algorithm to construct a parameter dynamic adjustment module, and generates an adaptive detection strategy based on the optimized detection path. Finally, the utilization module constructs an incremental feature expansion module based on the optimized detection path and adaptive detection strategy, completing the construction of the entire weld quality inspection system.

[0069] In terms of hardware implementation, this system is executed by a computing device comprising a processor and memory. The memory stores a computer program, which the processor runs to execute the weld quality inspection method described above. The computing device connects to the acquisition module via a communication interface, receives multimodal signal data from the sensor array, and transmits this data to the various functional modules via an internal bus for processing. The memory also stores key algorithmic modules, such as pre-trained multimodal feature extraction network models, generative adversarial network models, and variational Gaussian mixture models, ensuring the system's efficient operation under diverse material conditions.

[0070] In this embodiment, the connection relationship and position relationship between each module are reasonably designed to ensure the transmission efficiency and processing accuracy of signal data. For example, a high-speed data transmission interface is used between the acquisition module and the enhancement module to avoid data loss and delay problems. The enhancement module and the construction module are connected through a special data channel to ensure that the output of the multi-modal perception module can be quickly transmitted to the hierarchical feedback regulation unit. The adjustment module and the utilization module are connected through a double-channel data bus to support the bidirectional transmission of the optimized detection path and the adaptive detection strategy. In addition, the coordination relationship between each module is also optimized to ensure that the system can maintain high performance in complex environments. For example, the cross-modal information fusion mechanism in the enhancement module works cooperatively with the generative adversarial network to ensure that the deep association of multi-modal signal data is effectively mined. The hierarchical feedback regulation unit in the construction module closely cooperates with the parameter dynamic adjustment module in the adjustment module to ensure that the dynamic adjustment of the detection path and the core parameters can respond to environmental changes in real time.

[0071] In order to better enable relevant persons in the art to fully understand and implement the present application, the specific implementation principles of the present application are further supplemented below in conjunction with specific application scenarios.

[0072] In actual industrial manufacturing scenarios, taking the welding of aluminum alloy frames in the automotive manufacturing field as an example, a weld quality detection system is applied for real-time monitoring on the production line. After the welding robot completes the laser intelligent welding of a section of aluminum alloy frame, there may be defects such as micro-cracks, pores or incomplete fusion on the weld surface. At this time, the weld quality detection system starts and gradually completes multi-modal signal acquisition, feature extraction and fusion, dynamic path optimization and adaptive strategy generation according to the preset process.

[0073] First, the acquisition module monitors the weld surface in real time through a sensor array. A high-resolution optical camera captures the optical reflection characteristics of the weld surface, and an infrared thermal imager records the thermal field distribution characteristics of the weld area. Since aluminum alloy has high thermal conductivity, heat is rapidly dispersed during welding, resulting in significant differences in temperature field distribution in the weld area compared to other metal materials. Therefore, the thermal field signals collected need to be processed in conjunction with the optical signals to fully reflect the state information of the weld. The acquisition module transmits these multi-modal signal data to the enhancement module in real time through a high-speed data transmission interface to ensure the integrity and timeliness of the data.

[0074] Subsequently, the enhancement module processes the collected data based on the multimodal feature extraction network. The convolutional neural network extracts the spatial features of the weld surface from the optical signal, such as the geometric shape and distribution pattern of the crack; the autoencoder extracts the time series features from the thermal field signal, such as the temperature change trend in the weld area. The cross-modal information fusion mechanism jointly encodes the two signals through a bidirectional gated recurrent unit to highlight key attributes. For example, in response to the common thermal stress concentration phenomenon in the aluminum alloy welding process, the cross-modal information fusion mechanism uses the attention allocation mechanism to prioritize the abnormal temperature gradient area in the thermal field signal and associate it with the surface crack characteristics in the optical signal. The generative adversarial network further optimizes this process. The generator is responsible for generating potential feature representations, and the discriminator evaluates them. The quality of the feature representation is continuously improved through adversarial training, and the multimodal perception module is finally determined.

[0075] The construction module constructs a hierarchical feedback regulation unit based on the multimodal perception module. The hierarchical feedback regulation unit adopts a tree-like structure design, and each layer of nodes corresponds to a different source of feedback signals. For example, the first layer of nodes receives the optical reflection characteristics of the weld surface, the second layer of nodes analyzes the thermal field distribution characteristics, and the third layer of nodes focuses on the stress distribution characteristics inside the material. The statistical modeling tool analyzes the data distribution of the feedback signal to identify potential patterns and changing trends. For example, by analyzing the thermal field signals in the aluminum alloy weld area, it was found that there was a significant temperature gradient change during the welding cooling process, which may be related to the residual stress distribution inside the weld. Based on this analysis result, the construction module generates an optimized detection path to ensure that subsequent detection can cover the key areas of the weld.

[0076] The adjustment module constructs a dynamic parameter adjustment module based on the hierarchical feedback adjustment unit. First, an initial dynamic parameter adjustment module is constructed using a transfer learning framework and an evolutionary algorithm. Differential evolution and genetic algorithms are used to optimize the module's adaptability to different materials. For example, in an aluminum alloy welding scenario, the initial dynamic parameter adjustment module adjusts the weight coefficient of the thermal field signal based on the high thermal conductivity of aluminum alloy, giving it a more prominent position in the detection process. Next, based on the optimized detection path, a variational Gaussian mixture model is used to extract variation patterns from the data distribution of feedback signals at different levels of the hierarchical feedback adjustment unit. For example, core parameters are dynamically adjusted through Bayesian inference, enabling the detection system to adapt to the complex thermal field distribution in the aluminum alloy weld area. A context-sensitive mechanism, combined with graph embedding technology and an attention allocation mechanism, further analyzes the reflectivity distribution, thermal conductivity differences, and thermal stress distribution characteristics of the aluminum alloy to generate a preliminary adaptive detection strategy. This strategy is then optimized using a time series decomposition model and reinforcement learning. Finally, the optimal detection strategy is searched for using a particle swarm optimization algorithm and simulated annealing algorithm.

[0077] The module constructs an incremental feature expansion module based on the optimized detection path and adaptive detection strategy. First, time series modeling and Bayesian inference are used to determine the incremental update rules. For example, a dynamically updated time window is set for the thermal field signal in the aluminum alloy weld area to ensure that the system can capture the changing trends of the temperature field in real time. Second, a structured analysis technique based on graph embedding is combined with a hierarchical clustering algorithm to model and process the multimodal data in the multimodal perception module, resulting in a structured representation of the multimodal data. For example, surface crack features in the optical signal are associated with abnormal temperature gradient areas in the thermal field signal to form a unified feature representation. Third, a dynamic graph embedding network and a variational Gaussian mixture model are used to encode the data obtained from the hierarchical feedback regulation unit and input the encoded data into the structured representation of the multimodal data, outputting an updated feature representation. Subsequently, an attention allocation mechanism combined with deep reinforcement learning is applied to highlight key attributes in the updated feature representation, for example, prioritizing features related to weld cracks. Finally, the reinforcement learning and particle swarm optimization algorithms are combined to optimize the parameters in the incremental update process of the parameter dynamic adjustment module to generate the optimized parameter configuration, and an incremental feature expansion module is generated based on the optimized feature representation and optimized parameter configuration.

[0078] Through the above steps, the weld quality inspection system achieves efficient inspection of aluminum alloy frame welds. During the inspection process, the system accurately identifies tiny cracks on the weld surface and, combined with thermal field signals, determines whether the cracks are accompanied by internal thermal stress concentration. Furthermore, the system dynamically adjusts the inspection strategy based on the material properties of the aluminum alloy, ensuring high accuracy and stability under varying welding conditions. Ultimately, the entire inspection process is completed with the support of a computing device. The processor runs the computer program stored in memory and executes the above methods, ensuring the system can operate efficiently in complex industrial environments.

[0079] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0080] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting weld quality of laser intelligent welding, characterized in that: The following steps are involved: Acquiring multimodal signal data of the weld surface, defining a multimodal feature extraction network, and generating an initial feature representation based on the multimodal feature extraction network and the multimodal signal data; Using a cross-modal information fusion mechanism and an adaptive compensation algorithm to enhance the correlation between the optical signal and the thermal field signal in the initial feature representation, and determining a multimodal perception module; Based on the multimodal perception module, a hierarchical feedback adjustment unit is constructed, and the hierarchical feedback adjustment unit is used to process feedback signals from different levels to generate an optimized detection path; Based on the hierarchical feedback regulation unit, a parameter dynamic adjustment module is constructed using a transfer learning framework and an evolutionary algorithm, and core parameters are adjusted based on the parameter dynamic adjustment module and the optimized detection path to generate an adaptive detection strategy; Based on the optimized detection path and the adaptive detection strategy, an incremental feature expansion module is constructed using a structured analysis technology based on graph embedding. Based on the incremental feature expansion module, the multimodal feature extraction network, the multimodal perception module, the hierarchical feedback adjustment unit and the parameter dynamic adjustment module, a weld quality detection system is constructed.

2. The method according to claim 1, characterized in that The method comprises the following steps: constructing a parameter dynamic adjustment module based on the hierarchical feedback adjustment unit using a transfer learning framework and an evolutionary algorithm; and adjusting core parameters based on the parameter dynamic adjustment module and the optimized detection path to generate an adaptive detection strategy. Using a transfer learning framework and an evolutionary algorithm, an initial parameter dynamic adjustment module is constructed based on the hierarchical feedback adjustment unit, and the adaptability of the initial parameter dynamic adjustment module between different materials is optimized by a differential evolution algorithm and a genetic algorithm; Based on the optimized detection path, a variational Gaussian mixture model is used to analyze the change pattern from the data distribution of feedback signals at different levels in the hierarchical feedback adjustment unit, and the core parameters in the initial parameter dynamic adjustment module are dynamically adjusted through Bayesian inference to generate a target parameter dynamic adjustment module; The context-sensitive mechanism is used to enable the target parameter dynamic adjustment module to adjust the detection strategy according to the characteristics of different materials to generate an adaptive detection strategy.

3. The method according to claim 2, characterized in that The method of utilizing the context-sensitive mechanism to enable the target parameter dynamic adjustment module to adjust the detection strategy according to the characteristics of different materials to generate an adaptive detection strategy includes the following steps: Utilizing a context-sensitive mechanism combined with graph embedding technology and an attention allocation mechanism, key properties are analyzed from the characteristics of different materials in the hierarchical feedback regulation unit, including reflectivity distribution, thermal conductivity differences, and thermal stress distribution characteristics; Based on the key attributes, the detection threshold in the target parameter dynamic adjustment module is dynamically adjusted in combination with Bayesian inference and Gaussian process modeling to generate a preliminary adaptive detection strategy; optimizing the preliminary adaptive detection strategy through reinforcement learning using a time series decomposition model to generate an optimized detection strategy; Searching for an optimal detection strategy from the optimized detection strategies by combining a particle swarm optimization algorithm and a simulated annealing algorithm; An active sampling mechanism is introduced to select target samples from the optimal detection strategy to generate an adaptive detection strategy.

4. The method according to claim 3, characterized in that The method of analyzing key attributes from the characteristics of different materials in the hierarchical feedback regulation unit by using a context-sensitive mechanism combined with graph embedding technology and an attention allocation mechanism includes the following steps: Using signal preprocessing technology to collect and clean the characteristics of different materials in the hierarchical feedback regulation unit to obtain material characteristic data of the hierarchical feedback regulation unit; A context-sensitive mechanism is introduced to combine principal component analysis and independent component analysis to perform preliminary screening and dimensionality reduction on the material characteristic data to obtain target material characteristic information; Applying graph embedding technology and complex network modeling to analyze the correlation between materials from the target material characteristic information to construct a material relationship graph; The attention allocation mechanism is combined with deep reinforcement learning to highlight the important attributes in the material relationship graph, and the key attribute weight distribution result is obtained; The material relationship diagram and the key attribute weight distribution result are analyzed to extract key attributes.

5. The method according to claim 2, characterized in that The method comprises the following steps: analyzing the variation pattern of the data distribution of feedback signals at different levels in the hierarchical feedback adjustment unit using a variational Gaussian mixture model based on the optimized detection path, and dynamically adjusting the core parameters in the initial parameter dynamic adjustment module through Bayesian inference to generate a target parameter dynamic adjustment module. Based on the optimized detection path, a variational Gaussian mixture model and statistical modeling tools are used to model the data distribution of feedback signals at different levels in the hierarchical feedback regulation unit, and potential patterns and change trends are analyzed from the feedback signals to generate data distribution change rules; Based on the data distribution change law, the core parameters in the initial parameter dynamic adjustment module are dynamically adjusted through Bayesian inference, Gaussian process modeling and random forest to generate a target parameter dynamic adjustment module.

6. The method according to claim 1, characterized in that The method of utilizing a cross-modal information fusion mechanism and an adaptive compensation algorithm to enhance the correlation between the optical signal and the thermal field signal in the initial feature representation and determining a multimodal perception module includes the following steps: An initial multimodal feature extraction network is defined using a hierarchical graph embedding network, and an initial feature representation is analyzed from the intrinsic relationship between the optical signal and the thermal field signal of the initial multimodal feature extraction network; Based on the initial feature representation, a cross-modal information fusion mechanism is introduced in combination with a bidirectional gated recurrent unit to perform joint encoding processing on the optical signal and the thermal field signal in the initial feature representation to obtain a multimodal information representation; constructing a generative adversarial network based on the multimodal information representation, and analyzing key data from an adversarial training process between a generator and a discriminator in the generative adversarial network; The initial multimodal feature extraction network is optimized based on the key data in combination with a variational Gaussian mixture model to obtain a target multimodal feature extraction network, and a multimodal perception module is determined based on the target multimodal feature extraction network.

7. The method according to claim 1, characterized in that The method of constructing an incremental feature expansion module based on the optimized detection path and the adaptive detection strategy using a structured analysis technology based on graph embedding includes the following steps: Determining incremental update rules using the optimized detection path and the adaptive detection strategy in combination with time series modeling and Bayesian inference; Based on the incremental update rule, the multimodal data in the multimodal perception module is modeled and processed using a structured analysis technique based on graph embedding combined with a hierarchical clustering algorithm to obtain a structured representation of the multimodal data; encoding data obtained from the hierarchical feedback regulation unit according to the multimodal data structured representation in combination with a dynamic graph embedding network and a variational Gaussian mixture model, and inputting the data into the multimodal data structured representation to output an updated feature representation; Applying an attention allocation mechanism combined with deep reinforcement learning to highlight key attributes in the updated feature representation to obtain an optimized feature representation; Combining reinforcement learning and particle swarm optimization algorithms to optimize and adjust the parameters in the incremental update process of the parameter dynamic adjustment module to generate an optimized parameter configuration; An incremental feature expansion module is generated based on the optimized feature representation and the optimized parameter configuration.

8. A laser intelligent welding weld quality detection system, characterized in that: include: An acquisition module is used to obtain multimodal signal data of the weld surface, define a multimodal feature extraction network, and generate an initial feature representation based on the multimodal feature extraction network and the multimodal signal data; an enhancement module, configured to enhance the correlation between the optical signal and the thermal field signal in the initial feature representation by utilizing a cross-modal information fusion mechanism and an adaptive compensation algorithm, and determine a multimodal perception module; A construction module, configured to construct a hierarchical feedback adjustment unit based on the multimodal perception module, and utilize the hierarchical feedback adjustment unit to process feedback signals from different levels to generate an optimized detection path; An adjustment module, configured to construct a parameter dynamic adjustment module based on the hierarchical feedback adjustment unit using a transfer learning framework and an evolutionary algorithm, and to adjust core parameters based on the parameter dynamic adjustment module and the optimized detection path to generate an adaptive detection strategy; A module is used to construct an incremental feature expansion module based on the optimized detection path and the adaptive detection strategy using a structured analysis technology based on graph embedding, and a weld quality detection system is constructed based on the incremental feature expansion module, the multimodal feature extraction network, the multimodal perception module, the hierarchical feedback adjustment unit and the parameter dynamic adjustment module.

9. The system according to claim 8, characterized in that The acquisition module includes a high-resolution optical camera and an infrared thermal imager, which are used to capture the optical reflection characteristics and thermal field distribution characteristics of the weld surface respectively.

10. A computing device, characterized in that The method comprises a processor and a memory, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to perform the method according to any one of claims 1 to 7.

Images