Metal repair defect online detection system and method

Abstract

The invention discloses a metal repair defect online detection system and method, the system comprises six units including a multi-dimensional heterogeneous composite sensing acquisition unit and a precise dynamic motion adaptation unit, and multi-modal data is processed by integrating multiple detection technologies and utilizing an optimized generative adversarial network and a Transform model. Wherein the multi-dimensional heterogeneous composite sensing acquisition unit acquires data from multiple dimensions, the precise dynamic motion adaptation unit provides precise position control, the intelligent signal processing and analysis central unit uses a model to carry out signal processing and defect feature mining, and the multi-modal data fusion and visual presentation unit realizes data fusion and visual presentation. The method is based on the system, and detection is completed through the steps of equipment installation, data acquisition, signal processing, result presentation, parameter adjustment, standard exceeding alarm and the like. According to the invention, the accuracy and efficiency of metal repair defect detection are effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of metal repair defect online detection, and in particular to a metal repair defect online detection system and method. Background Art

[0002] Metal repair technology is widely used in the industrial sector, and quality inspection of repaired metals is crucial to ensuring the safe and stable operation of equipment. As the manufacturing industry continues to demand higher precision and efficiency in metal repair, traditional online inspection technologies face numerous challenges and struggle to meet increasingly complex inspection needs.

[0003] Existing online metal repair defect detection technologies primarily rely on a single sensing method, such as ultrasonic or visual inspection. This single-sensor approach cannot fully capture defect information, has limited detection capabilities for complex structures and hidden defects, and is prone to missed and false detections. Furthermore, traditional detection systems often utilize fixed algorithms and models for data processing and analysis, lacking adaptive learning and optimization capabilities. When testing metals of varying materials and repair processes, it is difficult to dynamically adjust detection parameters and algorithms based on actual conditions, making it difficult to guarantee detection efficiency and accuracy. Furthermore, the limited visual presentation of test results hinders technicians from intuitively and quickly assessing defect conditions.

[0004] Based on this, there is an urgent need for an innovative online detection system and method for metal repair defects. Through the application of multimodal sensor fusion and advanced algorithms, the shortcomings of traditional technologies can be solved and the reliability and intelligence level of metal repair defect detection can be improved. Summary of the Invention

[0005] In order to overcome the shortcomings and deficiencies of the prior art, the present invention provides a metal repair defect online detection system and method.

[0006] The technical solution adopted by the present invention is an online detection system for metal repair defects, comprising:

[0007] A multi-dimensional heterogeneous composite sensing acquisition unit, which integrates a tunable electromagnetic ultrasonic phased array probe cluster with a frequency band of 2-10MHz, a pulsed eddy current induction detection module with adaptive frequency adjustment function, an industrial vision acquisition device with a pixel count of no less than 5 million and equipped with a multispectral imaging component, and a laser interferometry module with a wavelength of 532nm. Each component is connected to the system main control unit via a dedicated signal transmission cable;

[0008] A precision dynamic motion adaptation unit, comprising a six-axis precision robotic arm platform with a repeatability accuracy of ±0.01mm, an adaptive coupling contact mechanism with an elastic compensation structure, and an auxiliary support positioning structure for maintaining a stable detection environment. The six-axis precision robotic arm platform is fixedly connected to the multi-dimensional heterogeneous composite sensing acquisition unit via rigid connection components.

[0009] The intelligent signal processing and analysis central unit consists of a signal preprocessing subnet based on an optimized generative adversarial network and Transformer architecture, a defect feature mining network with multi-scale feature extraction capabilities, and a decision-making synthesis network that integrates multimodal information. This unit is connected to the multi-dimensional heterogeneous composite sensor acquisition unit and the subsequent visualization output unit via a high-speed data bus;

[0010] The multimodal data fusion and visualization presentation unit includes an image fusion processing module for B / C scanning image fusion display, a defect stereo modeling component based on a 3D reconstruction algorithm, and a document generation device capable of automatically generating quality rating reports. This unit interacts with the intelligent signal processing and analysis central unit through a data interface.

[0011] The system operation control and parameter adjustment unit includes a central control system for controlling the coordinated operation of each unit, a parameter adjustment module for real-time adjustment of detection parameters, and a system status monitoring and fault diagnosis device. It is connected to other units through control cables and communication protocols.

[0012] The environmental adaptation and auxiliary function unit includes a water circulation filtration system that maintains a constant water film layer between the probe and the repair area, a mobile water retaining mechanism to prevent water flow interference, and a temperature regulation and protection device to ensure stable operation of the system. It has physical and functional connections with the multi-dimensional heterogeneous composite sensing acquisition unit and the precision dynamic motion adaptation unit.

[0013] Furthermore, the signal preprocessing subnet in the intelligent signal processing and analysis central unit adopts a signal enhancement model based on an optimized generative adversarial network, and the formula is:

[0014]

[0015] Among them, X enhanced represents the enhanced signal, X original is the original acquisition signal, is the generator in the generative adversarial network, θ G is the parameter set of the generator, Z is the random noise vector, T θT is the signal transformation function of the Transformer model, θ Tis a parameter of the Transformer model. The defect feature mining network uses the Transformer's multi-head attention mechanism to build a feature extraction model. The formula is as follows:

[0016] F=Concat(head1, head2,..., head h )W O

[0017]

[0018] Among them, F is the extracted defect feature vector, head i is the output of the i-th attention head, h is the number of attention heads, W O is the output weight matrix, Q, K, V are query, key and value matrices respectively, d k is the dimension of the key vector.

[0019] Furthermore, the decision synthesis network in the intelligent signal processing and analysis central unit constructs a multimodal decision fusion model based on optimized generative adversarial network and Transformer, and the formula is:

[0020]

[0021] Among them, D final is the final decision result, F opt is the defect feature extracted by the optical channel, F ult is the defect feature extracted by the ultrasonic channel, [F opt , F ult ] means to splice the two together. is the generative adversarial network generator for feature fusion, γ G2 is the control parameter, T θT2 is the Transformer model used for decision transformation, θ T2 are the parameters of the Transformer model.

[0022] Furthermore, the electromagnetic ultrasonic phased array probe cluster in the multi-dimensional heterogeneous composite sensing acquisition unit adopts a probe excitation optimization model based on an optimized generative adversarial network, and the formula is:

[0023]

[0024] Among them, I excitation is the probe excitation current, P defect is the preset defect parameter vector, including defect type and size information, S env is the environmental parameter vector, including temperature and material properties, is the generative adversarial network generator, θ G3is the control parameter, and M is the excitation pattern matrix.

[0025] Furthermore, the image fusion processing module in the multimodal data fusion and visualization presentation unit uses a Transformer-based image fusion model, and the formula is:

[0026] I fused =T θT3 ([I opt , I ult ])

[0027] Among them, I fused is the fused image, I opt is the optically collected image, I ult is the image detected by ultrasound, [I opt , I ult ] represents image stitching, T θT3 is the Transformer model used for image fusion transformation, θ T3 are the parameters of the Transformer model.

[0028] Furthermore, the intelligent signal processing and analysis central unit has an adaptive learning and updating mechanism based on optimized generative adversarial networks and Transformer, and the model formula is:

[0029]

[0030] Among them, θ new is the updated model parameter, γ old is the original model parameter, α is the learning rate, is the gradient of the loss function with respect to the parameter γ, is the generative adversarial network generator used to generate new data, γ G4 For the control parameters, is the Transformer model used to process new data, θ T4 is the parameter of the Transformer model, X new For newly collected data.

[0031] Furthermore, the signal preprocessing subnet in the intelligent signal processing and analysis central unit includes: a signal noise reduction unit for removing noise interference in the collected signal; a signal normalization unit for normalizing signals of different scales; a preliminary signal feature extraction unit for extracting basic features of the signal; and a signal quality evaluation unit for evaluating the quality of the preprocessed signal.

[0032] Furthermore, the defect three-dimensional modeling component in the multimodal data fusion and visualization presentation unit includes: a three-dimensional point cloud generation unit, which converts multimodal data into a three-dimensional point cloud; a point cloud optimization unit, which smoothes and denoises the generated point cloud; a three-dimensional model construction unit, which constructs a defect three-dimensional model based on the optimized point cloud; and a model rendering unit, which renders and observes the three-dimensional model.

[0033] Furthermore, the parameter adjustment module in the system operation control and parameter adjustment unit includes: a detection parameter automatic adjustment unit, which automatically adjusts the sensing unit parameters according to the detection situation; a motion parameter adjustment unit, which adjusts the motion parameters of the precision dynamic motion adaptation unit; an algorithm parameter optimization unit, which optimizes the algorithm parameters in the intelligent signal processing and analysis central unit; and a system performance adjustment unit, which comprehensively adjusts the overall performance parameters of the system.

[0034] A method for online detection of metal repair defects, comprising the following steps:

[0035] Step S1: Install the multi-dimensional heterogeneous composite sensing acquisition unit on the six-axis precision robotic arm platform of the precision dynamic motion adaptation unit, fix each sensing component through a dedicated connection component, and enable the oil / water circulation filtration system to build an oil / water film layer of preset thickness between the probe and the repair area, while activating the mobile oil / water blocking mechanism;

[0036] Step S2: Controlling the precision dynamic motion adaptation unit to drive the multi-dimensional heterogeneous composite sensing acquisition unit at a preset speed to scan the metal repair area. During the scanning process, the electromagnetic ultrasonic phased array probe cluster, the pulsed eddy current induction detection module, the industrial vision acquisition device, and the laser interferometry measurement module are synchronously triggered to collect data;

[0037] Step S3: The collected multimodal data is transmitted to the intelligent signal processing and analysis central unit, and processed in sequence by the signal preprocessing subnet, the defect feature mining network, and the decision synthesis network to identify and analyze the defects;

[0038] Step S4: The results processed by the intelligent signal processing and analysis central unit are transmitted to the multimodal data fusion and visualization presentation unit, and data fusion, three-dimensional modeling and quality rating report generation are performed through the image fusion processing module, defect three-dimensional modeling component and document generation device;

[0039] Step S5: The system operation control and parameter adjustment unit monitors the operation status of each unit in real time and adjusts the system parameters accordingly based on the monitoring results;

[0040] Step S6: If an excessive defect is detected, the system operation control and parameter adjustment unit triggers the corresponding alarm device, and uploads the defect location coordinates and detection data to the external management system.

[0041] Beneficial Effects: This invention proposes an online detection system and method for metal repair defects. This system integrates multiple detection components through a multi-dimensional, heterogeneous composite sensor acquisition unit, overcoming the shortcomings of traditional single-sensor detection, which often lacks complete information acquisition. A cluster of electromagnetic ultrasonic phased array probes, a pulsed eddy current induction detection module, an industrial vision acquisition device, and a laser interferometry measurement module work together to collect data from multiple dimensions, including electromagnetic, optical, and mechanical, comprehensively covering the various characteristics of metal repair defects and significantly reducing the probability of missed and false detections. In terms of data processing, the intelligent signal processing and analysis central unit utilizes an optimized generative adversarial network and Transformer model, replacing traditional fixed algorithms. The signal preprocessing subnet enhances the signal, the defect feature mining network extracts multi-scale features, and the decision synthesis network fuses multimodal information to achieve adaptive detection of metals of different materials and processes. Furthermore, an adaptive learning and update mechanism enables the system to dynamically optimize parameters based on new data, significantly improving detection efficiency and accuracy. The multimodal data fusion and visualization unit fuses images for display, 3D modeling, and generates quality reports. Compared to traditional single-mode presentations, technicians can more intuitively and accurately determine defect conditions. The system operation control and parameter adjustment unit monitors and adjusts each unit in real time. The environmental adaptation and auxiliary function unit ensures the stability of the detection environment, ensuring that the entire detection process is efficient and reliable, comprehensively solving the shortcomings of traditional technologies, and providing advanced and effective technical solutions for metal repair defect detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a diagram of the system unit composition of the present invention;

[0043] Figure 2 The figure is a flow chart of the method steps of the present invention. DETAILED DESCRIPTION

[0044] It should be noted that, unless there is a conflict, the embodiments in this application and the features described in the embodiments can be combined with each other. The application is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0045] like Figure 1 As shown, a metal repair defect online detection system includes:

[0046] A multi-dimensional heterogeneous composite sensing acquisition unit integrates a tunable electromagnetic ultrasonic phased array probe cluster with a frequency band of 2-10MHz, a pulsed eddy current induction detection module with adaptive frequency adjustment, an industrial vision acquisition device with a pixel count of no less than 5 million and equipped with a multispectral imaging component, and a laser interferometry module with a wavelength of 532nm. Each component is connected to the system's main control unit via a dedicated signal transmission cable.

[0047] Specifically, the multi-dimensional heterogeneous composite sensing acquisition unit achieves all-round data acquisition by integrating multiple detection technologies. The tunable electromagnetic ultrasonic phased array probe cluster operates in the 2-10MHz frequency band and can dynamically adjust the frequency according to detection requirements to adapt to the detection of metal parts of different thicknesses and materials. The pulsed eddy current induction detection module has an adaptive frequency adjustment function, which can optimize the excitation signal to enhance the detection sensitivity of surface and near-surface defects. The industrial vision acquisition device has a pixel count of no less than 5 million and is equipped with a multispectral imaging component. It can acquire metal surface images in multiple spectral bands, improving the ability to identify tiny cracks and surface discontinuities. The laser interferometry measurement module uses a 532nm wavelength laser to achieve high-precision measurement of tiny deformations and vibrations on the metal surface by analyzing changes in interference fringes. Each component is connected to the system main control unit via a dedicated signal transmission cable to ensure the stability and reliability of data transmission.

[0048] The unit's implementation method is as follows: First, the initial parameters of each sensor component are configured based on the material, shape, and possible defect types of the metal part to be inspected. During the inspection process, the electromagnetic ultrasonic phased array probe cluster emits ultrasonic waves at a set frequency and angle, receiving reflected signals to detect internal defects. The pulsed eddy current induction detection module optimizes the detection of defects at different depths through adaptive frequency adjustment. The industrial vision acquisition device captures metal surface images in multiple spectral bands, capturing subtle defects that are difficult to detect with the naked eye. The laser interferometry measurement module scans the metal surface to detect surface flatness and minor deformations. Each component synchronously collects data and transmits it in real time to the system's main control unit via dedicated signal transmission cables, providing a comprehensive and accurate data foundation for subsequent defect analysis.

[0049] A precision dynamic motion adaptation unit, comprising a six-axis precision robotic arm platform with a repeatability accuracy of ±0.01mm, an adaptive coupling contact mechanism with an elastic compensation structure, and an auxiliary support positioning structure for maintaining a stable detection environment. The six-axis precision robotic arm platform is fixedly connected to the multi-dimensional heterogeneous composite sensing acquisition unit via rigid connection components.

[0050] Specifically, the precision dynamic motion adaptation unit provides precise position control and a stable operating environment for sensor acquisition. The six-axis precision robotic arm platform has a repeatability accuracy of ±0.01mm, and can accurately move the multi-dimensional heterogeneous composite sensing acquisition unit according to the preset trajectory to ensure that the detection covers all parts of the metal repair area. The adaptive coupling contact mechanism of the elastic compensation structure can automatically adjust the probe pressure according to the unevenness of the metal surface, ensuring good contact between the probe and the detection surface and improving the quality of the detection signal. The auxiliary support positioning structure maintains the stability of the entire detection system through multi-point support and dynamic balancing technology, reducing the impact of external vibration and interference on the detection results. The six-axis precision robotic arm platform and the multi-dimensional heterogeneous composite sensing acquisition unit are fixedly connected by rigid connecting components to ensure that the positional relationship between the two is accurate and stable.

[0051] The workflow of this unit is as follows: First, the motion trajectory of the six-axis precision robotic arm platform is planned according to the three-dimensional model and detection requirements of the metal repair area. During the detection process, the robotic arm moves precisely according to the planned trajectory, driving the multi-dimensional heterogeneous composite sensing acquisition unit to scan the metal surface. The adaptive coupling contact mechanism senses the contact state between the probe and the metal surface in real time, and adjusts the pressure through the elastic compensation structure to ensure that the probe and the surface maintain a good coupling state. The auxiliary support positioning structure continuously monitors the vibration and stability of the system, and offsets external interference by dynamically adjusting the position and strength of the support points to maintain stable operation of the system. Throughout the process, the rigid connection between the six-axis precision robotic arm platform and the multi-dimensional heterogeneous composite sensing acquisition unit ensures the accuracy and repeatability of the detection position, providing a reliable guarantee for high-quality detection results.

[0052] The intelligent signal processing and analysis central unit consists of a signal preprocessing subnet based on an optimized generative adversarial network and Transformer architecture, a defect feature mining network with multi-scale feature extraction capabilities, and a decision-making synthesis network that integrates multimodal information. This unit is connected to the multi-dimensional heterogeneous composite sensor acquisition unit and the subsequent visualization output unit via a high-speed data bus;

[0053] Specifically, the intelligent signal processing and analysis central unit implements in-depth processing and analysis of multimodal detection data based on an optimized generative adversarial network and Transformer architecture. The signal preprocessing subnet uses an optimized generative adversarial network to enhance and denoise the original acquired signal, while leveraging the sequence processing capabilities of the Transformer model to extract temporal features and local dependencies from the signal, improving signal quality and feature expression. The defect feature mining network utilizes the Transformer's multi-head attention mechanism to extract multi-scale, multi-level defect features from the preprocessed signal. It can simultaneously focus on the signal's global context and local details, effectively capturing the characteristic patterns of different defect types. The decision-making synthesis network fuses features from different modalities, such as optical and ultrasonic, and through the collaborative work of optimized generative adversarial networks and Transformer models, accurately determines the defect type, size, and location. This unit is connected to the multi-dimensional heterogeneous composite sensing acquisition unit and the subsequent visualization output unit via a high-speed data bus, ensuring efficient and real-time data transmission.

[0054] The implementation process is as follows: First, the signal preprocessing subnet receives the original signal from the multi-dimensional heterogeneous composite sensor acquisition unit, and performs operations such as filtering, enhancement and normalization on the signal. At the same time, it uses the optimized generative adversarial network to generate a clearer signal representation, and the Transformer model mines the potential feature relationships in the signal. Next, the defect feature mining network extracts features from the preprocessed signal, and processes feature information of different scales in parallel through a multi-head attention mechanism to generate a comprehensive and representative defect feature vector. Finally, the decision synthesis network fuses the feature vectors of different modalities, uses the optimized generative adversarial network to learn the nonlinear mapping relationship between features, and the Transformer model performs classification and regression analysis on the fused features to output the final defect detection results. The entire processing process is carried out on a high-speed data bus to ensure that large amounts of data can be transmitted and processed quickly and accurately, providing support for real-time online detection.

[0055] The multimodal data fusion and visualization presentation unit includes an image fusion processing module for B / C scanning image fusion display, a defect stereo modeling component based on a 3D reconstruction algorithm, and a document generation device capable of automatically generating quality rating reports. It interacts with the intelligent signal processing and analysis central unit through a data interface.

[0056] Specifically, the multimodal data fusion and visualization presentation unit integrates and intuitively displays the data obtained by different detection methods. The image fusion processing module adopts a Transformer-based image fusion algorithm to fuse the B / C scanning images of electromagnetic ultrasonic detection with the surface images collected by industrial vision, retaining the complementary information of the two images and generating a more comprehensive and clearer defect image. The defect stereo modeling component converts multimodal data into a three-dimensional point cloud based on a three-dimensional reconstruction algorithm. Through steps such as point cloud optimization, denoising and model building, it generates a three-dimensional model of the defect, which intuitively displays the spatial position and geometric shape of the defect. The document generation device automatically generates a quality rating report containing information such as the defect location, size, type and severity based on the detection results and the three-dimensional model, providing a basis for subsequent repair decisions. This unit interacts with the intelligent signal processing and analysis central unit through a data interface to ensure that the processed data can be fused and visualized in a timely and accurate manner.

[0057] The working method of this unit is as follows: First, the image fusion processing module receives different modal image data from the intelligent signal processing and analysis central unit, uses the Transformer model to learn the correlation between images, and fuses the internal structure information of the electromagnetic ultrasonic image with the surface detail information of the industrial visual image to generate a fused image with a higher amount of information. Then, the defect three-dimensional modeling component combines the fused image data with the surface morphology data obtained by the laser interferometry module to construct a three-dimensional point cloud model, and improves the accuracy and quality of the model through point cloud optimization and denoising. Finally, the document generation device automatically generates a quality rating report based on the three-dimensional model and defect analysis results according to the preset report template, which contains detailed information and visual charts of the defects. The entire process realizes an automated process from data fusion to visual presentation, providing technicians with intuitive and comprehensive detection results.

[0058] The system operation control and parameter adjustment unit includes a central control system for controlling the coordinated operation of each unit, a parameter adjustment module for real-time adjustment of detection parameters, and a system status monitoring and fault diagnosis device. It is connected to other units through control cables and communication protocols.

[0059] Specifically, the system operation control and parameter adjustment unit is responsible for the coordinated operation and parameter optimization of the entire detection system. The central control system is connected to other units such as the multi-dimensional heterogeneous composite sensing acquisition unit and the precision dynamic motion adaptation unit through dedicated control cables and communication protocols to achieve unified scheduling and coordinated control of each unit. The parameter adjustment module monitors the system's operating status and detection results in real time, and automatically adjusts the detection parameters of the sensing unit, the motion parameters of the robotic arm, and the parameters of the signal processing algorithm according to preset algorithms and rules to adapt to different detection scenarios and needs. The system status monitoring and fault diagnosis device continuously monitors the working status and performance indicators of each unit, detects in real time whether there are any abnormal conditions in the system, and promptly issues an alarm when a fault is found, while providing fault location and diagnostic information for rapid repair and recovery.

[0060] The implementation steps of this unit are as follows: First, the central control system initializes the operating parameters of each unit based on the inspection task and the characteristics of the metal component and sends a startup command to each unit. During the inspection process, the parameter adjustment module collects real-time operating data and test results from each unit. By analyzing this data, it dynamically adjusts inspection parameters, such as the excitation frequency of the electromagnetic ultrasonic probe and the exposure time of the industrial vision camera, to improve inspection accuracy and efficiency. The system status monitoring and fault diagnosis device continuously monitors key indicators such as voltage, current, and temperature of each unit through sensors and monitoring software. If an anomaly is detected, it immediately performs fault diagnosis and sends the diagnosis results to the central control system. The central control system then takes appropriate measures based on the fault condition, such as suspending inspection, adjusting the operating mode, or issuing an alarm. This real-time monitoring and dynamic adjustment mechanism ensures that the system always operates in optimal conditions, improving the reliability and stability of inspections.

[0061] The environmental adaptation and auxiliary function unit includes an oil / water circulation filtration system that maintains a constant oil / water film layer between the probe and the repair area, a mobile oil / water blocking mechanism to prevent water flow interference, and a temperature regulation and protection device to ensure stable operation of the system. It has physical and functional connections with the multi-dimensional heterogeneous composite sensing acquisition unit and the precision dynamic motion adaptation unit.

[0062] Specifically, the environmental adaptation and auxiliary function unit provides stable environmental conditions and necessary support for the inspection process. The oil / water circulation and filtration system, through pipes and nozzles, creates an oil / water film of a specific thickness between the probe and the metal repair area. This film not only serves as a coupling medium for ultrasonic testing, improving the propagation efficiency of ultrasonic waves, but also cools and cleans the inspection surface, meeting the surface lubrication and corrosion protection requirements of some parts. The movable oil / water baffle is made of flexible material and automatically adjusts its position with the probe's movement, preventing the oil / water flow from spreading outside the inspection area and interfering with the normal operation of other inspection equipment. The temperature control unit, through a heat exchanger and temperature control system, maintains the inspection environment temperature within a set range, minimizing the impact of temperature fluctuations on inspection results. The protective device, designed with high-strength materials, effectively prevents damage from external dust, moisture, and mechanical impact, ensuring stable operation. This unit is physically and functionally connected to the multi-dimensional heterogeneous composite sensing acquisition unit and the precision dynamic motion adaptation unit, providing them with the necessary operating conditions and protection.

[0063] The operating process is as follows: Before inspection begins, the oil / water circulation and filtration system activates, transferring clean oil / water between the probe and the metal surface to form a stable oil / water film. A mobile oil / water baffle automatically adjusts based on the probe's position to ensure the integrity and stability of the oil / water film. A temperature control device monitors the ambient temperature in real time and maintains it within a preset range through heating or cooling systems. The protective device remains operational throughout system operation, providing comprehensive protection. During inspection, the oil / water circulation and filtration system continuously circulates and filters the oil / water, maintaining a clean and uniform oil / water film. The mobile oil / water baffle adjusts its position in real time to prevent interference with the oil / water flow. The temperature control device continuously monitors temperature changes to ensure a stable ambient temperature. The protective device provides real-time protection against external interference and damage, ensuring safe system operation. Through these measures, the environmental adaptation and auxiliary function unit provides a stable and reliable operating environment for online metal repair defect detection.

[0064] 2. The metal repair defect online detection system according to claim 1 is characterized in that the signal preprocessing subnet in the intelligent signal processing and analysis central unit adopts a signal enhancement model based on an optimized generative adversarial network, and the formula is:

[0065]

[0066] Among them, X enhanced represents the enhanced signal, X original is the original acquisition signal, is the generator in the generative adversarial network, θ Gis the parameter set of the generator, Z is the random noise vector, T γT is the signal transformation function of the Transformer model, θ T is a parameter of the Transformer model. The defect feature mining network uses the Transformer's multi-head attention mechanism to build a feature extraction model. The formula is as follows:

[0067] F=Concat(head1, head2,..., head h )W O

[0068]

[0069] Among them, F is the extracted defect feature vector, head i is the output of the i-th attention head, h is the number of attention heads, W O is the output weight matrix, Q, K, V are query, key and value matrices respectively, d k is the dimension of the key vector.

[0070] Specifically, the signal preprocessing subnet in the intelligent signal processing and analysis central unit adopts a signal enhancement model based on an optimized generative adversarial network. This model processes the original collected signal and random noise vector through a generator, and combines the signal transformation function composed of the Transformer model to enhance the signal feature expression capability. The defect feature mining network uses the Transformer's multi-head attention mechanism to construct a feature extraction model, splicing the outputs of multiple attention heads and transforming them through a weight matrix to extract multi-scale

[0071] 3. The metal repair defect online detection system according to claim 2 is characterized in that the decision-making integration network in the intelligent signal processing and analysis central unit constructs a multimodal decision fusion model based on optimized generative adversarial networks and Transformer, and the formula is:

[0072]

[0073] Among them, D final is the final decision result, F opt is the defect feature extracted by the optical channel, F ult is the defect feature extracted by the ultrasonic channel, [F opt , F ult ] means to splice the two together. is the generative adversarial network generator for feature fusion, θ G2 is the control parameter, T θT2 is the Transformer model used for decision transformation, θ T2are the parameters of the Transformer model.

[0074] Specifically, the decision synthesis network constructs a multimodal decision fusion model based on optimized generative adversarial networks and Transformer. It splices the defect features extracted from the optical channel and the ultrasonic channel, first generates fusion features through the generative adversarial network, then uses the Transformer model to perform decision transformation, and finally outputs the classification results through the Softmax function, realizing the effective fusion of multimodal information and accurate judgment of defect types.

[0075] 4. The metal repair defect online detection system according to claim 1 is characterized in that the electromagnetic ultrasonic phased array probe cluster in the multi-dimensional heterogeneous composite sensing acquisition unit adopts a probe excitation optimization model based on an optimized generative adversarial network, and the formula is:

[0076]

[0077] Among them, I excitation is the probe excitation current, P defect is the preset defect parameter vector, including defect type and size information, S env is the environmental parameter vector, including temperature and material properties, is the generative adversarial network generator, γ G3 is the control parameter, and M is the excitation pattern matrix.

[0078] Specifically, the electromagnetic ultrasonic phased array probe cluster in the multi-dimensional heterogeneous composite sensing acquisition unit adopts a probe excitation optimization model based on an optimized generative adversarial network. This model generates excitation current through a generative adversarial network according to the preset defect parameter vector and environmental parameter vector, and combines it with the excitation pattern matrix to achieve adaptive optimization of the probe excitation parameters, thereby improving the detection sensitivity of different types of defects.

[0079] 5. The metal repair defect online detection system according to claim 1 is characterized in that the image fusion processing module in the multimodal data fusion and visualization presentation unit uses a Transformer-based image fusion model, and the formula is:

[0080] I fused =T θT3 ([I opt , I ult ])

[0081] Among them, I fused is the fused image, I opt is the optically collected image, I ult is the image detected by ultrasound, [I opt , I ult] represents image stitching, T θT3 is the Transformer model used for image fusion transformation, θ T3 are the parameters of the Transformer model.

[0082] Specifically, the image fusion processing module in the multimodal data fusion and visualization presentation unit uses a Transformer-based image fusion model to splice optically collected images and ultrasonically detected images. By learning the correlation between images through the Transformer model, it realizes the effective fusion of multimodal images and generates more comprehensive and clearer defect images.

[0083] 6. The metal repair defect online detection system according to claim 1 is characterized in that the intelligent signal processing and analysis central unit has an adaptive learning and updating mechanism based on optimized generative adversarial networks and Transformer, and the model formula is:

[0084]

[0085] Among them, θ new is the updated model parameter, θ old is the original model parameter, α is the learning rate, is the gradient of the loss function with respect to the parameter θ, is the generative adversarial network generator used to generate new data, θ G4 For the control parameters, is the Transformer model used to process new data, θ T4 is the parameter of the Transformer model, X new For newly collected data.

[0086] Specifically, the intelligent signal processing and analysis central unit has an adaptive learning and updating mechanism based on optimized generative adversarial networks and Transformers. According to the newly collected data, new training samples are generated through generative adversarial networks, and these samples are processed using the Transformer model. The gradient of the loss function with respect to the model parameters is calculated, and the model parameters are updated in combination with the learning rate, so that the system can continuously adapt to new detection environments and defect types.

[0087] 7. The metal repair defect online detection system according to claim 1 is characterized in that the signal preprocessing subnet in the intelligent signal processing and analysis central unit includes: a signal noise reduction unit for removing noise interference in the collected signal; a signal normalization unit for normalizing signals of different scales; a preliminary signal feature extraction unit for extracting the basic features of the signal; and a signal quality evaluation unit for evaluating the quality of the preprocessed signal.

[0088] Specifically, the signal preprocessing subnet in the intelligent signal processing and analysis central unit contains a signal noise reduction unit that removes noise interference in the collected signal through filtering and other operations; the signal normalization unit normalizes signals of different scales to make them have the same dimension and distribution; the preliminary signal feature extraction unit extracts the basic features of the signal to provide a basis for subsequent deep feature mining; the signal quality assessment unit assesses the quality of the preprocessed signal to ensure that the signal input into the subsequent network has high quality.

[0089] 8. The metal repair defect online detection system according to claim 1 is characterized in that the defect three-dimensional modeling component in the multimodal data fusion and visualization presentation unit includes: a three-dimensional point cloud generation unit, which converts multimodal data into a three-dimensional point cloud; a point cloud optimization unit, which smoothes and denoises the generated point cloud; a three-dimensional model construction unit, which constructs a three-dimensional defect model based on the optimized point cloud; and a model rendering unit, which renders and observes the three-dimensional model.

[0090] Specifically, the defect three-dimensional modeling component in the multimodal data fusion and visualization presentation unit includes a three-dimensional point cloud generation unit that converts multimodal data into a three-dimensional point cloud, a point cloud optimization unit that smoothes and denoises the generated point cloud to improve the point cloud quality, a three-dimensional model construction unit that constructs a defect three-dimensional model based on the optimized point cloud, and a model rendering unit that renders the three-dimensional model to make it easier to observe and provide technicians with intuitive defect morphology information.

[0091] 9. The metal repair defect online detection system according to claim 1 is characterized in that the parameter adjustment module in the system operation control and parameter adjustment unit includes: a detection parameter automatic adjustment unit, which automatically adjusts the sensor unit parameters according to the detection situation; a motion parameter adjustment unit, which adjusts the motion parameters of the precision dynamic motion adaptation unit; an algorithm parameter optimization unit, which optimizes the algorithm parameters in the intelligent signal processing and analysis central unit; and a system performance adjustment unit, which comprehensively adjusts the overall performance parameters of the system.

[0092] Specifically, the parameter adjustment module in the system operation control and parameter adjustment unit is equipped with a detection parameter automatic adjustment unit that automatically adjusts the sensor unit parameters according to the detection situation, the motion parameter adjustment unit adjusts the motion parameters of the precision dynamic motion adaptation unit, the algorithm parameter optimization unit optimizes the algorithm parameters in the intelligent signal processing and analysis central unit, and the system performance adjustment unit comprehensively adjusts the overall performance parameters of the system. Through the coordinated work of each adjustment unit, it is ensured that the system can achieve the best operating state in different detection scenarios.

[0093] like Figure 2 As shown, a metal repair defect online detection method includes the following steps:

[0094] Step S1: Install the multi-dimensional heterogeneous composite sensing acquisition unit on the six-axis precision robotic arm platform of the precision dynamic motion adaptation unit, fix each sensing component through a dedicated connection component, and enable the oil / water circulation filtration system to build an oil / water film layer of preset thickness between the probe and the repair area, while activating the mobile oil / water blocking mechanism;

[0095] Step S2: Controlling the precision dynamic motion adaptation unit to drive the multi-dimensional heterogeneous composite sensing acquisition unit at a preset speed to scan the metal repair area. During the scanning process, the electromagnetic ultrasonic phased array probe cluster, the pulsed eddy current induction detection module, the industrial vision acquisition device, and the laser interferometry measurement module are synchronously triggered to collect data;

[0096] Step S3: The collected multimodal data is transmitted to the intelligent signal processing and analysis central unit, and processed in sequence by the signal preprocessing subnet, the defect feature mining network, and the decision synthesis network to identify and analyze the defects;

[0097] Step S4: The results processed by the intelligent signal processing and analysis central unit are transmitted to the multimodal data fusion and visualization presentation unit, and data fusion, three-dimensional modeling and quality rating report generation are performed through the image fusion processing module, defect three-dimensional modeling component and document generation device;

[0098] Step S5: The system operation control and parameter adjustment unit monitors the operation status of each unit in real time and adjusts the system parameters accordingly based on the monitoring results;

[0099] Step S6: If an excessive defect is detected, the system operation control and parameter adjustment unit triggers the corresponding alarm device, and uploads the defect location coordinates and detection data to the external management system.

[0100] This invention addresses the shortcomings of traditional technologies, such as incomplete information and insufficient data processing capabilities from single-sensor detection, and achieves breakthroughs through multi-faceted innovation. At the sensing level, the multi-dimensional heterogeneous composite sensing acquisition unit integrates an electromagnetic ultrasonic phased array probe cluster, a pulsed eddy current induction detection module, an industrial vision acquisition device, and a laser interferometer measurement module to collect data from multiple dimensions, including electromagnetic properties, surface morphology, and internal structure. Different sensing components cover different detection dimensions. For example, the electromagnetic ultrasonic phased array can detect deep defects, and the industrial vision acquisition device can capture surface microscopic features. Multi-source data complements and verifies each other, greatly improving the detection capability of complex structures and hidden defects compared to traditional single sensing methods, and effectively reducing the probability of missed detection or false detection.

[0101] In terms of data processing and analysis, the intelligent signal processing and analysis central unit plays a core role. Based on the signal preprocessing subnet optimized for generative adversarial networks and the Transformer architecture, the generative adversarial network is used to enhance the signal and suppress noise, while the Transformer model performs signal transformation to improve signal quality. The defect feature mining network, relying on the Transformer's multi-head attention mechanism, extracts multi-scale features from multimodal data and accurately captures defect information. The decision-making synthesis network integrates multimodal features and makes decisions by optimizing the generative adversarial network and the Transformer model, enabling the system to adapt to metal detection of different materials and repair processes. At the same time, the adaptive learning update mechanism can dynamically optimize model parameters based on newly collected data, completely changing the situation where traditional fixed algorithms cannot flexibly adapt to detection needs, and significantly improving detection efficiency and accuracy.

[0102] In addition, the multimodal data fusion and visualization unit fuses and displays inspection images from different modalities, such as electromagnetic ultrasound and industrial vision. It also constructs a 3D defect model based on a 3D reconstruction algorithm. Combined with the automatically generated quality rating report, it presents inspection results in an intuitive and comprehensive manner, enabling technicians to quickly and accurately determine defect conditions and addressing the limitations of traditional inspection results, which often lack a single visualization format. The system operation control and parameter adjustment unit monitors the operating status of each unit in real time and adjusts parameters. The environmental adaptation and auxiliary function unit ensures a stable inspection environment. These units operate in coordination, providing an efficient, reliable, and complete solution for metal repair defect detection.

[0103] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "disposed," "installed," "connected," "connected," and "fixed" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; they may refer to mechanical connections or electrical connections; they may refer to direct connections or indirect connections through an intermediate medium; and they may refer to internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0104] While embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and variations may be made to the 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 metal repair defect online detection system, characterized in that: include: A multi-dimensional heterogeneous composite sensing acquisition unit, which integrates a tunable electromagnetic ultrasonic phased array probe cluster with a frequency band of 2-10MHz, a pulsed eddy current induction detection module with adaptive frequency adjustment function, an industrial vision acquisition device with a pixel count of no less than 5 million and equipped with a multispectral imaging component, and a laser interferometry module with a wavelength of 532nm. Each component is connected to the system main control unit via a dedicated signal transmission cable; A precision dynamic motion adaptation unit, comprising a six-axis precision robotic arm platform with a repeatability accuracy of ±0.01mm, an adaptive coupling contact mechanism with an elastic compensation structure, and an auxiliary support positioning structure for maintaining a stable detection environment. The six-axis precision robotic arm platform is fixedly connected to the multi-dimensional heterogeneous composite sensing acquisition unit via rigid connection components. The intelligent signal processing and analysis central unit consists of a signal preprocessing subnet based on an optimized generative adversarial network and Transformer architecture, a defect feature mining network with multi-scale feature extraction capabilities, and a decision-making synthesis network that integrates multimodal information. This unit is connected to the multi-dimensional heterogeneous composite sensor acquisition unit and the subsequent visualization output unit via a high-speed data bus; The multimodal data fusion and visualization presentation unit includes an image fusion processing module for B / C scanning image fusion display, a defect stereo modeling component based on a 3D reconstruction algorithm, and a document generation device capable of automatically generating quality rating reports. This unit interacts with the intelligent signal processing and analysis central unit through a data interface. The system operation control and parameter adjustment unit includes a central control system for controlling the coordinated operation of each unit, a parameter adjustment module for real-time adjustment of detection parameters, and a system status monitoring and fault diagnosis device. It is connected to other units through control cables and communication protocols. The environmental adaptation and auxiliary function unit includes an oil / water circulation filtration system that maintains a constant oil / water film layer between the probe and the repair area, a mobile oil / water blocking mechanism to prevent oil / water flow interference, and a temperature regulation and protection device to ensure stable operation of the system. It has physical and functional connections with the multi-dimensional heterogeneous composite sensing acquisition unit and the precision dynamic motion adaptation unit.

2. The metal repair defect online detection system according to claim 1, characterized in that: The signal preprocessing subnet in the intelligent signal processing and analysis central unit adopts a signal enhancement model based on an optimized generative adversarial network, and the formula is: Among them, X enhanced represents the enhanced signal, X original is the original acquisition signal, is the generator in the generative adversarial network, θ G is the parameter set of the generator, Z is the random noise vector, T θT is the signal transformation function of the Transformer model, θ T is a parameter of the Transformer model. The defect feature mining network uses the Transformer's multi-head attention mechanism to build a feature extraction model. The formula is as follows: F=Concat(head1,head2,…,head h )W O Among them, F is the extracted defect feature vector, head i is the output of the i-th attention head, h is the number of attention heads, W O is the output weight matrix, Q, K, V are query, key and value matrices respectively, d k is the dimension of the key vector.

3. The metal repair defect online detection system according to claim 2, characterized in that: The decision synthesis network in the intelligent signal processing and analysis central unit constructs a multimodal decision fusion model based on optimized generative adversarial networks and Transformer. The formula is: Among them, D final is the final decision result, F opt is the defect feature extracted by the optical channel, F ult is the defect feature extracted by the ultrasonic channel, [F opt , F ult ] means to splice the two together. is the generative adversarial network generator for feature fusion, θ G2 is the control parameter, T θT2 is the Transformer model used for decision transformation, θ T2 are the parameters of the Transformer model.

4. The metal repair defect online detection system according to claim 1, characterized in that: The electromagnetic ultrasonic phased array probe cluster in the multi-dimensional heterogeneous composite sensing acquisition unit adopts a probe excitation optimization model based on an optimized generative adversarial network, and the formula is: Among them, I excitation is the probe excitation current, P defect is the preset defect parameter vector, including defect type and size information, S env is the environmental parameter vector, including temperature and material properties, is the generative adversarial network generator, θ G3 is the control parameter, and M is the excitation pattern matrix.

5. The metal repair defect online detection system according to claim 1, characterized in that: The image fusion processing module in the multimodal data fusion and visualization presentation unit uses a Transformer-based image fusion model, and the formula is: I fused =T θT3 ([I opt ,I ult ]) Among them, I fused is the fused image, I opt is the optically collected image, I ult is the image detected by ultrasound, [I opt , I ult ] represents image stitching, T θT3 is the Transformer model used for image fusion transformation, θ T3 are the parameters of the Transformer model.

6. The metal repair defect online detection system according to claim 1, characterized in that: The intelligent signal processing and analysis central unit has an adaptive learning and updating mechanism based on optimized generative adversarial networks and Transformer. The model formula is: Among them, θ new is the updated model parameter, θ old is the original model parameter, α is the learning rate, is the gradient of the loss function with respect to the parameter γ, is the generative adversarial network generator used to generate new data, γ G4 is the control parameter, is the Transformer model used to process new data, γ T4 is the parameter of the Transformer model, X new For newly collected data.

7. The metal repair defect online detection system according to claim 1, characterized in that: The signal preprocessing subnet in the intelligent signal processing and analysis central unit includes: a signal noise reduction unit for removing noise interference in the collected signal; a signal normalization unit for normalizing signals of different scales; a preliminary signal feature extraction unit for extracting the basic features of the signal; and a signal quality assessment unit for assessing the quality of the preprocessed signal.

8. The metal repair defect online detection system according to claim 1, characterized in that: The defect three-dimensional modeling component in the multimodal data fusion and visualization presentation unit includes: a three-dimensional point cloud generation unit, which converts multimodal data into a three-dimensional point cloud; a point cloud optimization unit, which smoothes and denoises the generated point cloud; a three-dimensional model construction unit, which constructs a three-dimensional defect model based on the optimized point cloud; and a model rendering unit, which renders and observes the three-dimensional model.

9. The metal repair defect online detection system according to claim 1, characterized in that: The parameter adjustment module in the system operation control and parameter adjustment unit includes: a detection parameter automatic adjustment unit, which automatically adjusts the sensor unit parameters according to the detection situation; a motion parameter adjustment unit, which adjusts the motion parameters of the precision dynamic motion adaptation unit; an algorithm parameter optimization unit, which optimizes the algorithm parameters in the intelligent signal processing and analysis central unit; and a system performance adjustment unit, which comprehensively adjusts the overall performance parameters of the system.

10. A metal repair defect online detection method, characterized in that: The following steps are involved: Step S1: Install the multi-dimensional heterogeneous composite sensing acquisition unit on the six-axis precision robotic arm platform of the precision dynamic motion adaptation unit, fix each sensing component through a dedicated connection component, and enable the water circulation filtration system to build an oil / water film layer of preset thickness between the probe and the repair area, while activating the mobile oil / water blocking mechanism; Step S2: Controlling the precision dynamic motion adaptation unit to drive the multi-dimensional heterogeneous composite sensing acquisition unit at a preset speed to scan the metal repair area. During the scanning process, the electromagnetic ultrasonic phased array probe cluster, the pulsed eddy current induction detection module, the industrial vision acquisition device, and the laser interferometry measurement module are synchronously triggered to collect data; Step S3: The collected multimodal data is transmitted to the intelligent signal processing and analysis central unit, and processed in sequence by the signal preprocessing subnet, the defect feature mining network, and the decision synthesis network to identify and analyze the defects; Step S4: The results processed by the intelligent signal processing and analysis central unit are transmitted to the multimodal data fusion and visualization presentation unit, and data fusion, three-dimensional modeling and quality rating report generation are performed through the image fusion processing module, defect three-dimensional modeling component and document generation device; Step S5: The system operation control and parameter adjustment unit monitors the operation status of each unit in real time and adjusts the system parameters accordingly based on the monitoring results; Step S6: If an excessive defect is detected, the system operation control and parameter adjustment unit triggers the corresponding alarm device, and uploads the defect location coordinates and detection data to the external management system.

Images