Fire extinguisher tank welding seam quality intelligent monitoring and control method and system based on multi-source sensing

Abstract

The invention provides a fire extinguisher tank welding seam quality intelligent monitoring and control method and system based on multi-source sensing. According to the method, when a circular seam and a longitudinal seam of a fire extinguisher tank are welded, an electromagnetic eddy current phase deviation signal and an optical vision depth-of-field signal of a welding seam area are synchronously collected through a welding gun support integrated flexible sensing unit; fusing the two types of signals to generate a chromatography fusion matrix for quantifying the fusion state and defect distribution; dynamically adjusting the detection focal length and the scanning track of the eddy current and the optical probe according to the matrix characteristic distribution, and outputting a topological map representing interlayer bonding defects; performing eddy current phase decoupling on the atlas to separate surface layer and deep layer electromagnetic response, and constructing a defect distribution vector; and a graph convolutional neural network processing vector is adopted to generate a welding parameter correction instruction, and the welding parameter correction instruction is fed back to a welding power supply to adjust power. Defect dynamic suppression of the weld fusion state is achieved, the welding defect rate is reduced, and manual intervention is not needed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of quality intelligent monitoring and control, and particularly relates to a kind of intelligent monitoring and control method and system for welding seam quality of fire extinguisher tank based on multi-source sensing. BACKGROUND

[0002] As a pressure safety component, the welding seam quality of fire extinguisher tank is directly related to the sealing strength and burst pressure reliability of the product. In the automatic production line, the welding seam of the tank is subject to multiple process disturbances: first, material property fluctuations cause unstable dynamic behavior of the molten pool, which easily leads to incomplete penetration or undercut defects; second, the complexity of the geometric structure limits the field of view of the sensor, making it difficult to capture the root penetration and micro-porosity in real time; third, the multi-physical field coupling effect exacerbates the risk of fatigue crack initiation. Traditional manual detection or single-sensor systems cannot simultaneously perceive the thermal, mechanical and acoustic multi-modal signals during welding, and there is an urgent need for an intelligent monitoring method that can integrate multi-source sensing data, dynamically identify hidden defects and provide real-time closed-loop control to ensure that the welding seam meets the strict safety standards of pressure vessels throughout its life cycle.

[0003] The current targeted solution is an intelligent control system based on machine vision and ultrasonic waves: the system captures the molten pool morphology and welding seam formation characteristics in real time through a high-frame-rate camera, and simultaneously integrates ultrasonic sensors to collect reflected wave signals from internal defects in the welding seam; after feature extraction, the multi-source data is input into a defect recognition model, which outputs the defect category and confidence level; a control module dynamically adjusts the welding parameters based on the recognition results to achieve online correction and process optimization of the welding seam quality. However, the core deficiency is that the real-time performance of multi-sensor data fusion is limited by algorithmic computation delays, and the response sensitivity to microscopic defects is insufficient, making it difficult to achieve closed-loop control in high-speed production lines. SUMMARY

[0004] The present application provides an intelligent monitoring and control method and system for welding seam quality of fire extinguisher tank based on multi-source sensing, which addresses the issues of high defect detection rate, isolated multi-source sensing information and lagging welding quality control in existing technologies.

[0005] In a first aspect, the present application provides an intelligent monitoring and control method for welding seam quality of fire extinguisher tank based on multi-source sensing, comprising: During the welding process of the fire extinguisher tank girth seam and longitudinal seam, a flexible sensing unit is integrated on the welding gun holder to simultaneously collect electromagnetic eddy current phase shift signals and optical vision depth signals from the welding seam area of the fire extinguisher tank; The electromagnetic eddy current phase shift signals and optical vision depth signals are fused to generate a tomographic fusion matrix, which is used to quantify the fusion state and potential defect distribution between different welding passes in the welding seam area of the fire extinguisher tank; Obtaining a feature distribution of the tomographic fusion matrix, and dynamically adjusting a detection focal length and a scanning track of the eddy current probe and the optical probe in the flexible sensing unit according to the feature distribution to output a topological atlas characterizing the interlayer bonding defects of the weld; Performing eddy current phase decoupling processing on the topological atlas to separate surface electromagnetic responses and deep electromagnetic responses, and constructing a defect distribution vector according to the surface electromagnetic responses and the deep electromagnetic responses; Processing the defect distribution vector by using a graph convolutional neural network to generate a welding parameter correction instruction, and feeding back the welding parameter correction instruction to a welding power source to adjust welding power.

[0006] Optionally, during the welding process of the annular seam and the longitudinal seam of the fire extinguisher tank body, a flexible sensing unit is integrated on the welding gun support to synchronously collect electromagnetic eddy current phase shift signals and optical visual depth of field signals of the fire extinguisher tank body weld seam area, comprising: A flexible sensing unit is installed at the front end of the welding gun support, the flexible sensing unit comprising side-by-side arranged eddy current probes and optical probes, the detection ends of the eddy current probes and the optical probes being directed towards the fire extinguisher tank body weld seam area; A synchronous trigger pulse is generated synchronously when the welding gun starts welding, the synchronous trigger pulse simultaneously starting the eddy current probes in the flexible sensing unit to emit electromagnetic wave beams and the optical probes to emit laser beams; The eddy current probes receive electromagnetic waves reflected by the fire extinguisher tank body weld seam area and measure the phase shift amount of the electromagnetic waves, while the optical probes receive laser beams reflected by the fire extinguisher tank body weld seam area and measure the depth of field change amount of the laser beams; The phase shift amount is converted into an electromagnetic eddy current phase shift signal, and the depth of field change amount is converted into the optical visual depth of field signal.

[0007] Optionally, the electromagnetic eddy current phase shift signal and the optical visual depth of field signal are fused to generate a tomographic fusion matrix, the tomographic fusion matrix being used to quantify the fusion state and potential defect distribution between different welds in the fire extinguisher tank body weld seam area, comprising: Aligning the electromagnetic eddy current phase shift signal and the optical visual depth of field signal according to the same time stamp and spatial coordinates; Extracting phase shift feature values of the aligned electromagnetic eddy current phase shift signal, the phase shift feature values describing the degree of change of electromagnetic characteristics of the fire extinguisher tank body weld seam area; Extracting depth of field change feature values of the aligned optical visual depth of field signal, the depth of field change feature values describing the degree of fluctuation of the surface topography of the fire extinguisher tank body weld seam area; Weighted fusion of the phase shift feature values and the depth of field change feature values according to a preset weight ratio to generate a fusion feature value of each spatial position point; The fusion feature values of all spatial position points are arranged according to a preset spatial coordinate grid to form a tomographic fusion matrix quantifying the fusion state and defect distribution of the weld.

[0008] Optionally, a feature distribution of the tomographic fusion matrix is acquired, and a detection focal length and a scanning track of the eddy current probe and the optical probe in the flexible sensing unit are dynamically adjusted according to the feature distribution to output a topological atlas representing the interlayer bonding defects of the weld, including: The fusion feature value of each spatial position point is extracted from the tomographic fusion matrix, and when the fusion feature value exceeds a preset defect feature threshold, a spatial coordinate marker of the spatial position point is generated; A target focusing position of the eddy current probe and the optical probe is calculated according to the spatial coordinate marker, and the target focusing position is converted into an elevation angle adjustment amount and a horizontal angle adjustment amount of the eddy current probe and the optical probe; The mechanical structure of the flexible sensing unit is adjusted according to the elevation angle adjustment amount and the horizontal angle adjustment amount, so that the detection focal length of the eddy current probe and the optical probe is aligned with the weld seam area of the fire extinguisher tank, and the scanning track of the eddy current probe and the optical probe is planned according to the distribution path of the spatial coordinate marker; Weld seam area data are collected under the adjusted detection focal length and scanning track, and the weld seam area data are converted into a topological atlas representing the interlayer bonding defects of the weld.

[0009] Optionally, the topological atlas is subjected to eddy current phase decoupling processing to separate the surface electromagnetic response and the deep electromagnetic response, and a defect distribution vector is constructed according to the surface electromagnetic response and the deep electromagnetic response, including: The eddy current phase raw data of the electromagnetic eddy current phase shift signal of each spatial position point in the topological atlas are extracted, and the eddy current phase raw data are subjected to frequency domain decomposition processing to separate the surface electromagnetic response and the deep electromagnetic response; The signal intensity value of the surface electromagnetic response and the signal intensity value of the deep electromagnetic response are calculated, and the difference between the signal intensity values is taken as a response difference amount of the spatial position point; The response difference amount is bound with the coordinates of the corresponding spatial position point to form a defect feature data unit; All spatial position point defect feature data units are integrated, and the defect feature data units are arranged in spatial grid order to form a defect distribution vector.

[0010] Optionally, the eddy current phase raw data of the electromagnetic eddy current phase shift signal of each spatial position point in the topological atlas are extracted, and the eddy current phase raw data are subjected to frequency domain decomposition processing to separate the surface electromagnetic response and the deep electromagnetic response, including: The original eddy current phase data of the electromagnetic eddy current phase shift signal at each spatial location point is extracted from the topology map. The original eddy current phase data includes the original phase response value of the electromagnetic wave in the weld material. The original phase response value is converted into a frequency domain spectrum distribution, and the spectrum range of the frequency domain spectrum distribution is divided by a preset first cutoff frequency and a second cutoff frequency. Extract the spectral components above the first cutoff frequency, and convert the spectral components into a time-domain signal as the surface electromagnetic response; Extract the spectral components below the second cutoff frequency, convert the spectral components into time-domain signals as deep electromagnetic responses, and output the surface electromagnetic responses and deep electromagnetic responses according to spatial location points.

[0011] Optionally, a graph convolutional neural network is used to process the defect distribution vector to generate welding parameter correction instructions, and the welding parameter correction instructions are fed back to the welding power source to adjust the welding power, including: The defect feature data units in the defect distribution vector are written into the node feature extraction layer of the graph convolutional neural network to extract the defect feature vector of each spatial location point. Obtain the spatial coordinates of the defect distribution vector, construct an adjacency relation matrix of a graph structure based on the spatial coordinates, and connect adjacent spatial position points of the adjacency relation matrix to form a graph node network; The defect feature vectors of adjacent nodes in the graph node network are aggregated through the message passing mechanism of the graph convolutional neural network to generate a fused defect feature quantity. The fusion defect feature quantity is mapped to a welding parameter correction quantity, which includes a welding power adjustment value and a welding speed adjustment value; The welding parameter correction amount is encapsulated into a welding parameter correction command and sent to the welding power source. The welding power source adjusts its output power and travel speed according to the welding parameter correction command.

[0012] Secondly, this application provides an intelligent monitoring and control system for the weld quality of fire extinguisher tanks based on multi-source sensing, including: The acquisition module is used to simultaneously acquire electromagnetic eddy current phase shift signals and optical visual depth signals of the weld seam area of ​​the fire extinguisher tank by integrating a flexible sensing unit on the welding torch bracket during the welding process of the circumferential and longitudinal seams of the fire extinguisher tank. The fusion module is used to fuse the electromagnetic eddy current phase shift signal with the optical visual depth signal to generate a tomographic fusion matrix. The tomographic fusion matrix is ​​used to quantify the fusion state and potential defect distribution between different welds in the weld area of ​​the fire extinguisher tank. An adjusting module is configured to obtain a feature distribution of the tomographic fusion matrix, and dynamically adjust a detection focal length and a scanning track of the eddy current probe and the optical probe in the flexible sensing unit according to the feature distribution, so as to output a topological atlas representing the interlayer bonding defects of the weld; A constructing module is configured to perform eddy current phase decoupling processing on the topological atlas to separate a surface electromagnetic response and a deep electromagnetic response, and construct a defect distribution vector according to the surface electromagnetic response and the deep electromagnetic response. A feedback module is configured to process the defect distribution vector by using a graph convolutional neural network to generate a welding parameter correction instruction, and feed back the welding parameter correction instruction to a welding power source to adjust a welding power.

[0013] In a third aspect, the present application provides a computing device, comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component, to realize the intelligent monitoring and control method of the welding quality of the fire extinguisher tank body based on multi-source sensing as described in the first aspect.

[0014] In a fourth aspect, the present application provides a computer storage medium, which stores a computer program; when the computer program is executed by a computer, the intelligent monitoring and control method of the welding quality of the fire extinguisher tank body based on multi-source sensing as described in the first aspect is realized.

[0015] The present application realizes the accurate control of the welding quality of the fire extinguisher tank body through multi-modal sensing fusion and intelligent defect analysis technology. Among them, the tomographic fusion based on electromagnetic eddy current and optical vision significantly improves the detection depth and resolution of the weld fusion state and defects; the self-adaptive detection adjustment mechanism ensures the synchronous capture of surface and deep defects; the intelligent analysis of the graph convolutional neural network realizes the dynamic optimization of the welding parameters. This method breaks through the limitations of traditional single detection technology, forms a full closed-loop intelligent regulation and control system from signal acquisition to process adjustment, significantly improves the stability and defect suppression ability of the welding process, and provides a high-reliability welding quality control solution for pressure vessel manufacturing.

[0016] Further, through the multi-modal sensing synchronous acquisition technology, the welding process is monitored in all directions in real time. Among them, the flexible integrated eddy current and optical probe significantly improves the synchronous capture ability of the electromagnetic and topographic features of the weld area; the synchronous trigger pulse mechanism ensures the time sequence consistency of multi-source signal acquisition; the accurate conversion of phase shift and depth of field effectively quantifies the weld fusion state. This method breaks through the limitations of traditional step-by-step detection, provides a high-precision, high-response in-situ monitoring means for welding quality control, and significantly improves the real-time and reliability of defect identification.

[0017] These aspects or other aspects of the present application will be more apparent in the following description of the embodiments. Attached Figure Description

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

[0019] Figure 1 A flowchart of an intelligent monitoring and control method for the quality of fire extinguisher tank welds based on multi-source sensing, provided in this application, is shown. Figure 2 A schematic diagram of the structure of an intelligent monitoring and control system for the weld quality of fire extinguisher tanks based on multi-source sensing provided in this application is shown. Figure 3 A schematic diagram of the structure of a computing device provided in this application is shown. Detailed Implementation

[0020] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0021] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0022] The technology for monitoring the quality of weld seams in fire extinguisher tanks faces a fundamental bottleneck: while existing solutions based on the synergy of machine vision and ultrasound can achieve multi-source data acquisition, their insufficient real-time fusion and lack of sensitivity to microscopic defects lead to closed-loop control failure. Specifically, algorithm calculation delays cause welding parameter adjustments to lag behind the dynamic changes in the molten pool, and the response sensitivity to microscopic porosity and root incomplete penetration is insufficient, making it difficult to meet the real-time control requirements of high-speed production lines. This contradiction stems from the insufficient ability to decouple the spatiotemporal synchronization of multimodal sensor signals from deep defects, necessitating the construction of an intelligent fusion monitoring architecture with high-frequency response and deep analysis.

[0023] In view of the above challenges, the present application proposes a multi-source sensing-based intelligent monitoring and control method for welding seam quality of fire extinguisher tank, which is innovative in breaking through the response bottleneck of traditional collaborative systems through multi-dimensional fusion of eddy current phase and optical depth of field and deep learning driving. Specifically: the flexible sensing unit synchronously collects electromagnetic eddy current phase shift signals and optical visual depth signals in the welding seam area of the fire extinguisher tank, generates a tomographic fusion matrix to quantify the fusion state and defect distribution; based on the matrix characteristics, the probe focal length and scanning trajectory are dynamically adjusted, and the topological atlas of the interlayer bonding defects of the welding bead is output; the surface and deep electromagnetic responses are separated through eddy current phase decoupling to construct a defect distribution vector; a graph convolutional neural network is used to process the vector to generate welding parameter correction instructions, which are fed back to the welding power supply to adjust the power. This method overturns the traditional multi-sensing paradigm: the tomographic fusion matrix first realizes the cross-dimensional dynamic correlation of micron-level defects; the topological atlas is optimized through adaptive adjustment of the probe parameters, and the detection blind area under complex geometric structures is overcome; the graph convolutional network converts the defect distribution vector into high-precision control instructions, forming an intelligent control chain of "signal fusion-graph generation-vector analysis-power correction", which provides full-link protection from multi-modal perception to power precise control for high-pressure container welding seams.

[0024] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0025] Figure 1 A flowchart of a multi-source sensing-based intelligent monitoring and control method for welding seam quality of fire extinguisher tank is provided for the embodiments of the present application, as shown in Figure 1 The method comprises the following steps. 101、In the welding process of the girth seam and the longitudinal seam of the fire extinguisher tank, a flexible sensing unit is integrated on the welding gun support to synchronously collect electromagnetic eddy current phase shift signals and optical visual depth signals in the welding seam area of the fire extinguisher tank, wherein the electromagnetic eddy current phase shift signals reflect the changes in electromagnetic characteristics inside the welding seam, and the optical visual depth signals represent the three-dimensional topography of the welding seam surface.

[0026] Optionally, step 101 can specifically include the following steps: 1011、A flexible sensing unit is installed at the front end of the welding gun support, wherein the flexible sensing unit comprises eddy current probes and optical probes arranged side by side, and the detection ends of the eddy current probes and the optical probes are directed to the welding seam area of the fire extinguisher tank.

[0027] 1012. A synchronization trigger pulse is generated simultaneously when the welding torch starts welding, which simultaneously starts the electromagnetic wave beam emitted by the eddy current probe and the laser beam emitted by the optical probe in the flexible sensing unit.

[0028] 1013. The eddy current probe receives the electromagnetic waves reflected by the weld area of the fire extinguisher tank body and measures the phase shift of the electromagnetic waves, while the optical probe receives the laser beam reflected by the weld area of the fire extinguisher tank body and measures the change in depth of field of the laser beam.

[0029] 1014. The phase shift is converted into an electromagnetic eddy current phase shift signal, and the change in depth of field is converted into an optical vision depth of field signal.

[0030] In the above scheme, the flexible sensing unit refers to a flexible sensor unit. The electromagnetic eddy current phase shift signal refers to the phase shift signal of electromagnetic eddy current. The optical vision depth of field signal refers to the depth of field signal of optical vision. The change in internal electromagnetic characteristics of the weld refers to the change in internal electromagnetic characteristics of the weld. The three-dimensional topography of the weld surface refers to the three-dimensional shape of the weld surface. The welding torch support refers to the support that supports the welding torch. The eddy current probe refers to the probe that detects eddy current. The optical probe refers to the probe that detects optical. The detection end refers to the detection end of the probe. The synchronization trigger pulse refers to the pulse signal of synchronization trigger. The electromagnetic wave beam refers to the beam of electromagnetic waves. The phase shift refers to the amount of phase shift. The change in depth of field refers to the amount of change in depth of field.

[0031] In the embodiments of the present application, first, by step 1011, a flexible sensing unit is installed at the front end of the welding torch support, which is integrated on the welding torch movement mechanism by mechanical clamping or bolt fixation, and contains an eddy current probe and an optical probe arranged side by side inside. The eddy current probe is usually designed with a coil type electromagnetic sensor, which can emit and receive electromagnetic waves of a specific frequency; the optical probe uses a laser triangulation or structured light projection sensor. The detection ends of the two probes are precisely calibrated during installation to ensure that their detection fields point to the weld area of the fire extinguisher tank body, providing a spatial alignment basis for subsequent synchronous data acquisition.

[0032] Subsequently, by step 1012, when the welding torch power supply starts the welding operation, its control system will generate a high-precision synchronization trigger pulse signal simultaneously. The pulse signal is sent to the eddy current probe drive circuit and the optical probe control module in the flexible sensing unit through cable or wireless transmission. The eddy current probe receives the pulse and immediately excites its internal oscillation circuit to generate an electromagnetic wave beam of a specific frequency and emit it to the weld area; the optical probe synchronously activates its laser emitter to project a laser beam to the same area. This strict synchronization mechanism ensures that the two signal acquisitions are completely consistent in time, eliminating the problem of asynchronous data caused by timing deviation.

[0033] Then, through step 1013, the eddy current probe continuously receives the electromagnetic waves reflected back from the weld seam area of the fire extinguisher tank body, and uses its internal phase detection circuit (usually based on the lock-in amplification technology) to measure the phase shift of the reflected electromagnetic waves relative to the transmitted waves, which reflects the changes in the electromagnetic properties (such as electrical conductivity, magnetic permeability) of the internal material of the weld seam, which can be used to infer the internal defect condition. At the same time, the optical probe receives the laser beam reflected back from the weld seam surface through its image sensor or photodetector, and calculates the propagation time or spot displacement of the laser beam based on the principle of laser triangulation or time-of-flight, thereby accurately measuring the change in the depth of field of the laser beam, which directly represents the three-dimensional topographic features (such as height, depression, etc.) of the weld seam surface.

[0034] Finally, through step 1014, the signal processing circuit built-in the eddy current probe converts the measured raw phase shift into a digitized electromagnetic eddy current phase shift signal after analog-to-digital conversion and signal conditioning (such as amplification, filtering); Similarly, the optical probe converts its measured depth of field change into a digitized optical vision depth signal through internal processing (which may involve geometric optical transformation and calibration algorithms). These two signals are finally output to the host computer system, providing fused data sources for subsequent weld quality analysis.

[0035] In practical applications, on the automatic welding production line of fire extinguisher tanks, a set of intelligent monitoring and control method based on multi-source sensing is implemented for the precise welding process of girth seam and longitudinal seam. In specific applications, a flexible sensing unit is integrated at the front end of the welding gun support, which includes side-by-side arranged eddy current probe and optical probe, whose detection ends are accurately pointed to the to-be-welded area of the fire extinguisher tank. When the welding gun starts welding, the system generates a synchronous trigger pulse simultaneously, which activates the eddy current probe to emit electromagnetic wave beam and the optical probe to emit laser beam. The eddy current probe receives the reflected electromagnetic waves from the weld seam area in real time and measures the phase shift, which reflects the changes in the electromagnetic properties of the weld seam; At the same time, the optical probe receives the reflected laser beam and measures the change in the depth of field to represent the three-dimensional topographic features of the weld seam surface. Subsequently, the system converts the measured phase shift into an electromagnetic eddy current phase shift signal and converts the depth of field change into an optical vision depth signal. These signals are transmitted in real time to the central processing system, and through fusion analysis, the trajectory and parameters of the welding gun are dynamically adjusted to ensure real-time dual monitoring of internal defects and surface topography during welding, thereby significantly improving the sealing and strength consistency of the weld seam of the fire extinguisher tank.

[0036] The scheme of step 101 realizes the synchronous acquisition and accurate characterization of the multi-modal signals of the weld area in the welding process. By integrating a flexible sensing unit on the welding gun support, a dual detection system of electromagnetic eddy current and optical vision is innovatively constructed. The technology adopts a synchronous trigger pulse mechanism to ensure the consistency of the eddy current phase shift signal and the optical depth of field signal in space and time. Through the synchronous acquisition of the electromagnetic wave phase measurement and the laser depth of field change, the comprehensive monitoring of the electromagnetic characteristics and surface topography inside the weld is realized. This multi-probe cooperative working mode provides a rich source of data for the welding quality evaluation, significantly improving the comprehensiveness and accuracy of the welding process monitoring.

[0037] 102. Fusing the electromagnetic eddy current phase shift signal and the optical vision depth of field signal to generate a tomographic fusion matrix, the tomographic fusion matrix being used to quantify the fusion state and potential defect distribution between different weld beads in the weld area of the fire extinguisher tank body.

[0038] Optionally, step 102 can specifically include the following steps: 1021. Aligning the electromagnetic eddy current phase shift signal and the optical vision depth of field signal according to the same time stamp and spatial coordinates.

[0039] 1022. Extracting phase shift characteristic values of the aligned electromagnetic eddy current phase shift signal, the phase shift characteristic values describing the degree of change of the electromagnetic characteristics of the weld area of the fire extinguisher tank body.

[0040] 1023. Extracting depth of field change characteristic values of the aligned optical vision depth of field signal, the depth of field change characteristic values describing the degree of fluctuation of the surface topography of the weld area of the fire extinguisher tank body.

[0041] 1024. Weighted fusing the phase shift characteristic values and the depth of field change characteristic values according to a preset weight ratio to generate a fusion characteristic value of each spatial position point.

[0042] 1025. Arranging the fusion characteristic values of all spatial position points according to a preset spatial coordinate grid to form a tomographic fusion matrix quantifying the fusion state and defect distribution of the weld bead.

[0043] In the above scheme, the tomographic fusion matrix refers to the data matrix of tomographic fusion. The fusion state refers to the state of welding fusion. The potential defect distribution refers to the distribution of potential defects. The phase shift characteristic value refers to the characteristic value of phase shift. The depth of field change characteristic value refers to the characteristic value of depth of field change. The fusion characteristic value refers to the characteristic value after fusion. The preset weight ratio refers to the preset weight ratio. The spatial coordinate grid refers to the grid of spatial coordinates. The weld bead fusion state refers to the state of weld bead fusion. The defect distribution refers to the distribution of defects.

[0044] In the embodiments of the present application, first, through step 1021, the system aligns the electromagnetic eddy current phase shift signal and the optical visual depth of field signal according to the same time stamp and spatial coordinates. The time synchronization module uses a high-precision clock source to mark the two types of signals with millisecond-level time stamps, ensuring the instantaneous consistency of data acquisition. The spatial registration engine performs geometric calibration on the detection fields of view of the electromagnetic eddy current probe and the optical probe through a coordinate mapping algorithm (such as affine transformation or feature point matching), so that each spatial coordinate point (such as a two-dimensional grid position) corresponds completely in the two types of signals. The aligned data is stored in a buffer queue to provide a unified input source in time and space for subsequent feature extraction.

[0045] Subsequently, through step 1022, the system extracts a phase shift feature value from the aligned electromagnetic eddy current phase shift signal. The phase demodulator uses a phase-locked amplification technique or a Hilbert transform algorithm to analyze the phase component of the signal and calculate the offset thereof relative to a reference benchmark. The feature quantization unit normalizes the offset to a scalar value, i.e., the phase shift feature value, which directly reflects the phase disturbance intensity caused by the change in the electromagnetic properties (such as electrical conductivity and magnetic permeability) of the weld area of the fire extinguisher tank body, and is used to quantify the electromagnetic response of internal defects (such as pores and inclusions).

[0046] Next, through step 1023, the system extracts a depth of field change feature value from the aligned optical visual depth of field signal. The topography analyzer calculates the height deviation of the surface points based on the laser triangulation or structured light projection principle. The gradient calculation module processes the height data through spatial difference operation (such as Sobel operator or Gaussian derivative) to generate a depth of field change feature value (such as the height change rate per unit distance) describing the degree of surface relief. This value represents the physical convex-concave features (such as weld bead protrusions, depressions or cracks) of the weld surface topography.

[0047] Then, through step 1024, the system performs weighted fusion of the phase shift feature value and the depth of field change feature value according to a preset weight ratio to generate a fusion feature value for each spatial position point. The weight distributor sets the fusion ratio of the two types of features according to prior knowledge such as the importance weight of the defect type, and the fusion engine executes a weighted sum algorithm to output the fusion feature value, which integrates the internal electromagnetic properties and surface topography information to form a composite description of the weld fusion state. For example, a high fusion value may indicate an unfused or porosity defect.

[0048] Finally, through step 1025, the system arranges the fusion feature values of all spatial position points according to a preset spatial coordinate grid to form a tomographic fusion matrix. The grid generator divides the two-dimensional grid elements according to the physical size of the detection area, and the matrix construction module fills the fusion feature values in each grid element into a two-dimensional array in row and column order to generate the tomographic fusion matrix. The row and column indices correspond to the spatial position elements, and the value quantifies the fusion quality of the position. The defect distribution and severity can be directly displayed through visualization techniques such as heat maps.

[0049] In practical applications, during the girth welding process of the fire extinguisher tank body, after the flexible sensing unit integrated in the welding gun support synchronously collects the electromagnetic eddy current phase shift signal and the optical visual depth of field signal, the system immediately starts the multi-source information fusion process to generate a tomographic fusion matrix. First, the system accurately aligns the electromagnetic eddy current phase shift signal and the optical visual depth of field signal according to the same time stamp and spatial coordinates, ensuring the consistency of the data in the time and space dimensions. Then, the algorithm extracts the phase shift characteristic value from the aligned electromagnetic eddy current phase shift signal, which quantitatively describes the degree of change in the electromagnetic characteristics inside the weld area; at the same time, it extracts the depth of field change characteristic value from the aligned optical visual depth of field signal, which quantifies the degree of fluctuation of the three-dimensional topography of the weld surface. Next, the system weights and fuses the phase shift characteristic value and the depth of field change characteristic value according to the preset weight ratio to generate a fusion characteristic value for each spatial location point, which reflects the correlation between the internal material properties and the surface morphology. Finally, the fusion characteristic values of all spatial location points are arranged according to the preset spatial coordinate grid system to form a tomographic fusion matrix, which quantitatively reflects the fusion state between different welds and the distribution of potential defects such as pores or incomplete fusion, providing a structured data foundation for subsequent intelligent judgment.

[0050] The above-mentioned scheme of step 102 realizes intelligent fusion of multi-source sensing signals and quantitative evaluation of weld quality. Through the alignment and weighted fusion of electromagnetic signals and optical signals, a tomographic fusion matrix reflecting the fusion state of the weld is innovatively constructed. This technology uses a combination of feature value extraction and weight distribution to effectively integrate multi-modal information. The innovative matrix construction algorithm converts discrete sensing data into a systematic quality evaluation index, providing a quantitative basis for weld defect identification. This data fusion method significantly improves the accuracy and reliability of weld quality evaluation.

[0051] 103、Obtain the feature distribution of the tomographic fusion matrix, and dynamically adjust the detection focal length and scanning trajectory of the eddy current probe and the optical probe in the flexible sensing unit according to the feature distribution to output a topological map representing the interlayer bonding defects of the weld.

[0052] Optionally, step 103 can specifically include the following steps: 1031、Extract the fusion characteristic value of each spatial location point from the tomographic fusion matrix, and generate a spatial coordinate marker for the spatial location point when the fusion characteristic value exceeds a preset defect characteristic threshold.

[0053] 1032、Calculate the target focusing position of the eddy current probe and the optical probe according to the spatial coordinate marker, and convert the target focusing position into the pitch angle adjustment amount and the horizontal angle adjustment amount of the eddy current probe and the optical probe.

[0054] 1033、adjusting the mechanical structure of the flexible sensing unit according to the pitch angle adjustment amount and the horizontal angle adjustment amount, so that the detection focal length of the eddy current probe and the optical probe is aligned with the weld seam area of the fire extinguisher tank body, and the scanning trajectory of the eddy current probe and the optical probe is planned according to the distribution path marked by the spatial coordinates.

[0055] 1034、collecting weld seam area data under the adjusted detection focal length and scanning trajectory, and converting the weld seam area data into a topological atlas representing the inter-pass bond defects of the weld bead.

[0056] In the above scheme, the feature distribution refers to the distribution of the features. The detection focal length refers to the focal length of the detection. The scanning trajectory refers to the path of the scanning. The inter-pass bond defects of the weld bead refer to the defects between the layers of the weld bead. The topological atlas refers to the atlas reflecting the topological structure. The defect feature threshold refers to the critical value for judging the defect features. The spatial coordinate marker refers to the marker of the spatial coordinates. The target focusing position refers to the position of the target focusing. The pitch angle adjustment amount refers to the adjustment amount of the pitch angle. The horizontal angle adjustment amount refers to the adjustment amount of the horizontal angle. The mechanical structure refers to the composition structure of the machine. The distribution path refers to the distribution path of the defects. The weld seam area data refers to the data of the weld seam area.

[0057] In the embodiments of the present application, first, the fusion feature value of each spatial position point is extracted from the tomographic fusion matrix through step 1031, and when the fusion feature value exceeds the preset defect feature threshold, the spatial coordinate marker of the spatial position point is generated. This process uses a feature extraction algorithm to scan the feature intensity of each position in the tomographic fusion matrix, and marks the abnormal points exceeding the threshold as potential defect areas in real time, and records their three-dimensional spatial coordinates; these spatial coordinate markers provide target positioning data for subsequent probe focusing and scanning, forming a preliminary spatial distribution map of the defect area.

[0058] Subsequently, the target focusing position of the eddy current probe and the optical probe is calculated according to the spatial coordinate marker through step 1032, and the target focusing position is converted into the pitch angle adjustment amount and the horizontal angle adjustment amount of the eddy current probe and the optical probe. Based on the three-dimensional information of the spatial coordinate marker, the system calculates the orientation and distance of each defect point relative to the probe base through a geometric projection model, and then calculates the pitch angle and horizontal angle adjustment amount required by the probe; this conversion process is realized through an embedded motion control algorithm, which converts the position coordinates into precise mechanical driving instructions, so that the probe can accurately aim at the target area.

[0059] Then, the mechanical structure of the flexible sensing unit is adjusted according to the pitch angle adjustment amount and the horizontal angle adjustment amount in step 1033 to align the detection focal length of the eddy current probe and the optical probe with the weld seam area of the fire extinguisher tank body, and to plan the scanning trajectory of the eddy current probe and the optical probe according to the distribution path marked by the spatial coordinates. After receiving the angle adjustment amount, the multi-axis servo system of the flexible sensing unit drives the probe support to pitch and horizontally rotate, so that the probe focal length dynamically focuses on the weld defect point; at the same time, the path planning algorithm generates a continuous scanning trajectory according to the distribution sequence marked by the spatial coordinates, to ensure that the probe covers all defect areas along the weld direction, and to realize adaptive tracking scanning.

[0060] Finally, the weld area data is collected under the adjusted detection focal length and scanning trajectory in step 1034, and the weld area data is converted into a topological atlas representing the interlayer bonding defects of the weld. The probe synchronously collects eddy current and optical data under the optimized focal length and trajectory, and generates a high-resolution interlayer feature image through signal fusion technology; the image processing algorithm maps the feature data into a gray or pseudo-color topological atlas, in which the brightness or color change directly represents the defect type, size and depth distribution, and finally outputs a digital defect atlas for quality evaluation.

[0061] In actual application, in the automatic welding and quality detection scene of the fire extinguisher tank body, the system further performs adaptive monitoring control based on the previously generated tomographic fusion matrix. Specifically, the system first extracts the fusion feature value of each spatial position point from the tomographic fusion matrix, and when the fusion feature value exceeds the preset defect feature threshold, a spatial coordinate marker of the spatial position point is automatically generated, thereby locating the potential defect area. Subsequently, the algorithm calculates the target focusing position of the eddy current probe and the optical probe according to the spatial coordinate marker, and converts the target focusing position into the pitch angle adjustment amount and the horizontal angle adjustment amount of the eddy current probe and the optical probe. By driving the mechanical structure of the flexible sensing unit, the system dynamically adjusts the pitch angle and the horizontal angle of the probe, so that the detection focal length of the eddy current probe and the optical probe accurately aligns with the high-risk position of the weld seam area of the fire extinguisher tank body, and the scanning trajectory is planned in real time according to the distribution path marked by the spatial coordinates, to ensure that all defect feature areas are covered. Under the adjusted detection focal length and scanning trajectory, the system re-collects the electromagnetic characteristic and three-dimensional topography data of the weld area, and finally converts the weld area data into a high-resolution topological atlas, which clearly represents the morphology and distribution of the interlayer bonding defects (such as incomplete fusion or cracks) of the weld, and provides an intuitive basis for subsequent process optimization.

[0062] The scheme of step 103 realizes the dynamic optimization of the sensing system parameters and the generation of the defect topology atlas. Based on the feature analysis of the tomographic fusion matrix, the adaptive adjustment scheme of the probe focal length and the scanning trajectory is innovatively designed. The technology adopts the method of combining defect threshold judgment and coordinate marking to realize the accurate adjustment of the detection parameters. The innovative mechanical structure adjustment mechanism ensures the alignment accuracy of the probe, and the high-resolution defect topology atlas is generated through the optimized scanning path. This intelligent adjustment technology significantly improves the pertinence and effectiveness of defect detection.

[0063] 104. performing eddy current phase decoupling processing on the topology atlas to separate the surface electromagnetic response and the deep electromagnetic response, and constructing a defect distribution vector according to the surface electromagnetic response and the deep electromagnetic response.

[0064] Optionally, step 104 can specifically include the following steps: 1041. extracting the eddy current phase raw data of the electromagnetic eddy current phase shift signal of each spatial position point in the topology atlas, and performing frequency domain decomposition processing on the eddy current phase raw data to separate the surface electromagnetic response and the deep electromagnetic response.

[0065] In step 1041, the eddy current phase raw data of the electromagnetic eddy current phase shift signal of each spatial position point in the topology atlas is extracted, and the eddy current phase raw data contains the original phase response value of the electromagnetic wave in the weld material. The original phase response value is converted into a frequency domain spectrum distribution, and the frequency spectrum distribution is divided by a preset first cutoff frequency and a second cutoff frequency. The frequency spectrum components above the first cutoff frequency are extracted, and the frequency spectrum components are converted into time domain signals as the surface electromagnetic response. The frequency spectrum components below the second cutoff frequency are extracted, and the frequency spectrum components are converted into time domain signals as the deep electromagnetic response. The surface electromagnetic response and the deep electromagnetic response are output according to the spatial position points.

[0066] 1042. calculating the signal intensity values of the surface electromagnetic response and the deep electromagnetic response, and taking the difference value of the signal intensity values as the response difference value of the spatial position point.

[0067] 1043. binding the response difference value with the coordinates of the corresponding spatial position point to form a defect feature data unit.

[0068] 1044. integrating the defect feature data units of all spatial position points, and arranging the defect feature data units in a spatial grid sequence to form a defect distribution vector.

[0069] In the above scheme, the eddy current phase decoupling processing refers to an eddy current phase decoupling processing method. The surface electromagnetic response refers to an electromagnetic response of a surface layer. The deep electromagnetic response refers to an electromagnetic response of a deep layer. The defect distribution vector refers to a vector of a defect distribution. The eddy current phase raw data refers to raw data of an eddy current phase. The frequency domain decomposition processing refers to a frequency domain decomposition processing method. The original phase response value refers to an original phase response value. The frequency domain spectrum distribution refers to a frequency domain spectrum distribution. The first cutoff frequency refers to a first cutoff frequency. The second cutoff frequency refers to a second cutoff frequency. The spectrum range refers to a frequency range of a spectrum. The spectrum component refers to a component of a spectrum. The time domain signal refers to a time domain signal. The signal intensity value refers to a signal intensity value. The response difference value refers to a response difference value. The defect feature data unit refers to a defect feature data unit. The spatial grid sequence refers to a spatial grid arrangement sequence.

[0070] In the embodiment of the application, first, the eddy current phase raw data of the electromagnetic eddy current phase offset signal of each spatial position point in the topological atlas is extracted through step 1041, and the eddy current phase raw data is subjected to frequency domain decomposition processing to separate the surface electromagnetic response and the deep electromagnetic response. The process reads the original phase response value of each spatial position point from the topological atlas, and the value represents the propagation phase characteristic of the electromagnetic wave in the weld material; then the original phase response value is converted from the time domain to the frequency domain through fast Fourier transform to form a frequency domain spectrum distribution; the preset first cutoff frequency and second cutoff frequency are used as threshold boundaries to separate the high-frequency component (above the first cutoff frequency) and inverse transform it into a time domain signal, which represents the rapid attenuation electromagnetic response of the surface layer of the material, and separate the low-frequency component (below the second cutoff frequency) and inverse transform it into a time domain signal, which represents the slow attenuation electromagnetic response of the deep layer of the material; finally, two independent response signals are output according to the spatial position points.

[0071] Then, the signal intensity value of the surface electromagnetic response and the signal intensity value of the deep electromagnetic response are calculated through step 1042, and the difference value of the signal intensity value is taken as the response difference value of the spatial position point. The system respectively performs amplitude integration or root mean square calculation on the surface and deep time domain signals of each spatial position point to quantify the signal intensity value; the absolute difference value of the two intensity values is obtained through subtraction operation, and the difference value reflects the attenuation gradient of the electromagnetic characteristics of the same position surface and deep layer, thereby defining the response difference value, which is used to indicate the possibility of the existence of defects.

[0072] Then, the response difference quantity is bound with the coordinate of the corresponding spatial position point through step 1043 to form a defect feature data unit. The response difference quantity of each spatial position point is associated with its three-dimensional spatial coordinates (including lateral, longitudinal and depth coordinates) through a data mapping algorithm to generate a structured data unit containing spatial attributes and intensity difference values; the data unit records the position identification and the difference quantity value at the same time, and constitutes the minimum data representing the defect feature of a single point.

[0073] Finally, the defect feature data units of all spatial position points are integrated through step 1044, and the defect feature data units are arranged in sequence according to the spatial grid to form a defect distribution vector. The system traverses the defect feature data units of all spatial position points, and interpolates or sorts them in the pre-defined spatial grid template according to their coordinate information; the discrete units are concatenated into a high-dimensional vector in the order of grid index through a data fusion algorithm, wherein the vector dimension is consistent with the number of spatial grids, and each dimension value corresponds to the response difference quantity of a specific grid; the finally generated defect distribution vector is used as the digital expression of the global defect feature, which is used for subsequent graph convolutional neural network processing.

[0074] In actual application, in the intelligent monitoring process of the ring seam welding quality of the fire extinguisher tank body, the system further performs eddy current phase decoupling processing based on the topological atlas representing the interlayer bonding defects of the weld bead generated in the early stage to accurately separate the defect signals. In specific implementation, the system first extracts the eddy current phase original data of the electromagnetic eddy current phase shift signal of each spatial position point in the topological atlas, and the original data includes the original phase response value of the electromagnetic wave during propagation in the weld seam material of the fire extinguisher tank body; then, the original phase response value is converted into a frequency domain spectrum distribution, and the frequency spectrum distribution is divided into a frequency spectrum range by a pre-set first cut-off frequency and a second cut-off frequency; the system extracts a high-frequency spectrum component above the first cut-off frequency, and converts the spectrum component into a time domain signal as a surface electromagnetic response, which is mainly used for identifying small cracks or pores near the surface of the weld seam; at the same time, a low-frequency spectrum component below the second cut-off frequency is extracted, and the spectrum component is converted into a time domain signal as a deep electromagnetic response, which is used for detecting hidden defects such as incomplete fusion or slag inclusion in the weld seam. After separation, the system calculates the signal intensity value of the surface electromagnetic response and the signal intensity value of the deep electromagnetic response of each spatial position point, and the difference between the two signal intensity values is taken as the response difference quantity of the spatial position point, which quantifies the relative significance of the surface and deep defect signals; then, the response difference quantity is bound with the coordinate of the corresponding spatial position point to form a defect feature data unit, each unit accurately corresponds to a physical position in the weld seam area; finally, the defect feature data units of all spatial position points are integrated, and are arranged in sequence according to the spatial grid to form a defect distribution vector, which comprehensively represents the spatial distribution and severity of various defects in the ring seam area of the fire extinguisher tank body, and provides a structured data basis for subsequent quality judgment and process adjustment.

[0075] The scheme of step 104 realizes the deep analysis of the weld defect and the construction of the distribution vector. Through the frequency domain decoupling processing of the eddy current phase signal, the separation and extraction of the surface and deep electromagnetic response are realized. The technology adopts the method of frequency domain decomposition and signal strength analysis to realize the deep analysis of the defect characteristics. The innovative response difference calculation and spatial coordinate binding mechanism form an accurate defect distribution vector, which provides structured data input for subsequent intelligent processing. This layered analysis method significantly improves the depth and accuracy of defect identification.

[0076] 105, processing the defect distribution vector by using a graph convolutional neural network to generate a welding parameter correction instruction, and feeding back the welding parameter correction instruction to a welding power source to adjust the welding power.

[0077] Optionally, step 105 can specifically include the following steps: 1051, write the defect feature data unit in the defect distribution vector into the node feature extraction layer of the graph convolutional neural network to extract the defect feature vector of each spatial position point.

[0078] 1052, obtain the spatial coordinates of the defect distribution vector, and construct an adjacency relationship matrix of the graph structure according to the spatial coordinates, and connect the adjacent spatial position points of the adjacency relationship matrix to form a graph node network.

[0079] 1053, aggregate the defect feature vectors of adjacent nodes in the graph node network by using the message passing mechanism of the graph convolutional neural network to generate a fused defect feature quantity.

[0080] 1054, map the fused defect feature quantity to a welding parameter correction quantity, and the welding parameter correction quantity includes a welding power adjustment value and a welding speed adjustment value.

[0081] 1055, encapsulate the welding parameter correction quantity as a welding parameter correction instruction and send it to the welding power source, and the welding power source adjusts the output power and the travel speed according to the welding parameter correction instruction.

[0082] In the above scheme, the graph convolutional neural network refers to a convolutional neural network for processing graph data. The welding parameter correction instruction refers to an instruction for correcting the welding parameter. The welding power source refers to a device for providing welding power. The node feature extraction layer refers to a network layer for extracting node features. The defect feature vector refers to a feature vector of defect features. The adjacency relationship matrix refers to a matrix representing the adjacency relationship. The graph node network refers to a node network of a graph structure. The message passing mechanism refers to a message passing manner of a graph neural network. The fused defect feature quantity refers to a fused defect feature quantity. The welding parameter correction quantity refers to a correction quantity of the welding parameter. The welding power adjustment value refers to an adjustment value of the welding power. The welding speed adjustment value refers to an adjustment value of the welding speed. The output power refers to the output power of the welding power source. The travel speed refers to the travel speed of the welding.

[0083] In the embodiment of the present application, first, the defect feature data units in the defect distribution vector are written into the node feature extraction layer of the graph convolutional neural network through step 1051 to extract the defect feature vector of each spatial position point. This process utilizes the inherent node feature extraction capability of the graph convolutional neural network to convert the original data of the defect type, size, and morphology corresponding to each spatial position point in the defect distribution vector into a high-dimensional numerical form of the defect feature vector through feature encoding and normalization processing, thereby providing standardized node feature input for subsequent graph structure processing.

[0084] Subsequently, the spatial coordinates of the defect distribution vector are obtained through step 1052, and an adjacency relationship matrix of the graph structure is constructed according to the spatial coordinates to connect adjacent spatial position points of the adjacency relationship matrix to form a graph node network. Based on the actual physical position information of the defect points, the system calculates the Euclidean distance or relative orientation between points, and if the distance between two points is below a certain threshold, it is determined to be adjacent, thereby generating a symmetric adjacency relationship matrix to represent the connection relationship between nodes; this matrix and the defect feature vector extracted in step 1051 jointly define the structure of the graph node network, where the nodes represent spatial position points and the edges represent the adjacency relationship between positions, thereby embedding the defect distribution into a non-Euclidean graph structure.

[0085] Then, the defect feature vectors of adjacent nodes in the graph node network are aggregated using the message passing mechanism of the graph convolutional neural network through step 1053 to generate a fused defect feature quantity. According to the adjacency relationship matrix, the graph convolutional layer iteratively performs aggregation operations on the defect feature vectors of each node, i.e., each node receives feature information from its direct neighbor nodes and fuses them through weighted summation or attention mechanism, thereby updating its own feature representation; through multi-layer stacked graph convolution operations, the features of each node are finally fused with the semantic information of its multi-order neighbors, generating a fused defect feature quantity that can comprehensively reflect the local defect cluster pattern and spatial dependency.

[0086] Then, the fusion defect feature quantity is mapped to the welding parameter correction quantity including the welding power adjustment value and the welding speed adjustment value through step 1054. The system inputs the fusion defect feature quantity output by the graph convolutional neural network into a fully connected layer or a regression layer, which performs nonlinear transformation and dimension reduction on the high-dimensional features through a pre-trained weight matrix, directly decodes continuous numerical output, and corresponds to the adjustment amount of the welding power and the adjustment amount of the welding speed respectively; the mapping process is essentially a data-based regression prediction, which enables the correction quantity to adapt to the spatial pattern of the defect distribution.

[0087] Finally, the welding parameter correction quantity is packaged into a welding parameter correction instruction through step 1055 and sent to the welding power source, and the welding power source adjusts the output power and the travel speed according to the instruction. The control system packages the welding power adjustment value and the welding speed adjustment value according to a predetermined communication protocol format, generates a digital instruction frame, and transmits it to the welding power source in real time through an industrial bus or Ethernet; after the welding power source analyzes the instruction, it dynamically adjusts the output characteristics of the inverter and the wire feeder speed, thereby realizing online optimization and closed-loop control of the welding process.

[0088] In practical applications, in the intelligent control scene of the ring seam welding quality of the fire extinguisher tank body of the fire extinguishing system, the system generates a defect distribution vector (which quantifies the defect types and densities of different spatial positions) based on the welding defect data collected by multiple source sensors (such as visual sensors and laser scanners). First, the defect feature data units (such as pore density and crack length) in the defect distribution vector are written into the node feature extraction layer of the graph convolutional neural network, and the defect feature vector of each spatial position point is extracted through convolution kernel weight calculation (each vector corresponds to the defect quantification features of a detection point on the tank body surface). Then, the spatial coordinates of the defect distribution vector are obtained (the coordinates map the actual position of the defect on the tank body surface), and the adjacency relationship matrix of the graph structure is constructed according to the spatial coordinates (the matrix elements represent the topological connection relationship between adjacent detection points). The adjacent spatial position points of the adjacency relationship matrix are connected to form a graph node network (the network nodes represent the detection points, and the edges represent the spatial adjacency relationship). Then, the defect feature vectors of the adjacent nodes in the graph node network are aggregated through the message passing mechanism of the graph convolutional neural network (each node receives the feature information of the neighborhood nodes and performs nonlinear fusion), and the fused defect feature quantity is generated (this feature quantity integrates the local and global defect distribution patterns). Then, the fused defect feature quantity is mapped to the welding parameter correction quantity (the conversion from features to parameters is realized through a fully connected layer), which includes the welding power adjustment value and the welding speed adjustment value (the power adjustment value compensates for the pore defects caused by insufficient heat input, and the speed adjustment value suppresses the crack propagation caused by overheating). Finally, the welding parameter correction quantity is packaged as a welding parameter correction instruction and sent to the welding power source, and the welding power source adjusts the output power and the travel speed according to the welding parameter correction instruction (such as reducing the power and increasing the speed to alleviate local overheating). The system optimizes the welding dynamic parameters in real time, significantly improving the sealing and mechanical strength of the ring seam of the fire extinguisher tank body.

[0089] The scheme of the above step 105 realizes intelligent optimization and closed-loop control of welding parameters. Through the innovative application of graph convolutional neural networks, intelligent mapping of defect distribution to welding parameters is realized. This technology uses a combination of graph structure construction and feature aggregation to realize deep learning of spatial defect features. The innovative parameter correction quantity generation mechanism converts defect features into executable welding instructions, realizing the coordinated adjustment of welding power and speed. This intelligent feedback control technology significantly improves the stability of the welding process and the consistency of the weld quality.

[0090] Figure 2 A structure diagram of an intelligent monitoring and control system for fire extinguisher tank welding quality based on multiple source sensors is provided for the embodiments of the present application, as shown in Figure 2 The system comprises: The acquisition module 21 is used for synchronously collecting electromagnetic eddy current phase shift signals and optical visual depth of field signals of the fire extinguisher tank welding seam area on the welding gun support during the welding process of the fire extinguisher tank ring seam and longitudinal seam. The fusion module 22 is used for fusing the electromagnetic eddy current phase shift signals and the optical visual depth of field signals to generate a tomographic fusion matrix, and the tomographic fusion matrix is used for quantifying the fusion state and potential defect distribution between different welding beads in the fire extinguisher tank welding seam area. The adjustment module 23 is used for acquiring the feature distribution of the tomographic fusion matrix, and dynamically adjusting the detection focal length and scanning track of the eddy current probe and the optical probe in the flexible sensing unit according to the feature distribution to output a topological atlas representing the interlayer bonding defects of the welding bead. The construction module 24 is used for carrying out eddy current phase decoupling processing on the topological atlas to separate the surface electromagnetic response and the deep electromagnetic response, and constructing a defect distribution vector according to the surface electromagnetic response and the deep electromagnetic response. The feedback module 25 is used for processing the defect distribution vector by using a graph convolutional neural network to generate a welding parameter correction instruction, and feeding back the welding parameter correction instruction to a welding power source to adjust the welding power.

[0091] Figure 2 The multi-source sensing based fire extinguisher tank welding seam quality intelligent monitoring and control system can perform the following steps Figure 1 The multi-source sensing based fire extinguisher tank welding seam quality intelligent monitoring and control method of the embodiment has the implementation principle and technical effects which will not be repeated. The specific operation modes of each module and unit of the multi-source sensing based fire extinguisher tank welding seam quality intelligent monitoring and control system in the above embodiment have been described in detail in the embodiment related to the method, and will not be described in detail here.

[0092] In one possible design, Figure 2 The multi-source sensing based fire extinguisher tank welding seam quality intelligent monitoring and control system of the embodiment can be implemented as a computing device, such as a computer. Figure 3 As shown, the computing device can include a storage component 31 and a processing component 32. The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32.

[0093] The processing component 32 is used for the above Figure 1 The multi-source sensing based fire extinguisher tank welding seam quality intelligent monitoring and control method of the embodiment.

[0094] The processing component 32 can include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component can also be one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, micro-controllers, microprocessors or other electronic components, for executing the above method.

[0095] The storage component 31 is configured to store various types of data to support the operation of the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0096] Of course, the computing device can also include other components, such as an input / output interface, a display component, a communication component, etc.

[0097] The input / output interface provides an interface between the processing component and peripheral interface modules, which can be output devices, input devices, etc.

[0098] The communication component is configured to facilitate wired or wireless communication between the computing device and other devices, etc.

[0099] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform, and the computing device can be a cloud server, and the processing component, the storage component, etc. can be a basic server resource rented or purchased from the cloud computing platform.

[0100] The embodiment of the present application also provides a computer storage medium, which stores a computer program, and the computer program can implement the above Figure 1 The embodiment shown in the figure is an intelligent monitoring and control method for welding seam quality of fire extinguisher tank based on multi-source sensing.

[0101] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device and unit can refer to the corresponding process in the foregoing method embodiment, which will not be described here.

[0102] The apparatus embodiments described above are merely illustrative, wherein the units described as separated components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purposes of the embodiments according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0103] Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and necessary universal hardware platforms, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of software products, and the computer software products can be stored in a computer readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and include a plurality of instructions to make a computer device (which can be a personal computer, a server, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0104] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A multi-source sensing-based intelligent monitoring and control method for welding seam quality of fire extinguisher tank, characterized in that, The application relates to a method for quantifying the fusion state and potential defect distribution of a welding seam area of a fire extinguisher tank. The method comprises the following steps: In the welding process of the annular seam and the longitudinal seam of the fire extinguisher tank, a flexible sensing unit is integrated on a welding gun support to synchronously collect electromagnetic eddy current phase shift signals and optical visual depth of field signals of the welding seam area of the fire extinguisher tank; The electromagnetic eddy current phase shift signals and the optical visual depth of field signals are fused to generate a tomographic fusion matrix, which is used to quantify the fusion state and potential defect distribution between different welding beads in the welding seam area of the fire extinguisher tank; The feature distribution of the tomographic fusion matrix is obtained, and the detection focal length and scanning trajectory of the eddy current probe and the optical probe in the flexible sensing unit are dynamically adjusted according to the feature distribution to output a topological atlas representing the interlayer bonding defects of the welding bead; The topological atlas is subjected to eddy current phase decoupling processing to separate the surface electromagnetic response and the deep electromagnetic response, and a defect distribution vector is constructed according to the surface electromagnetic response and the deep electromagnetic response; 2. The method of claim 1, wherein, A graph convolutional neural network is used to process the defect distribution vector to generate a welding parameter correction instruction, and the welding parameter correction instruction is fed back to a welding power source to adjust the welding power. In the welding process of the annular seam and the longitudinal seam of the fire extinguisher tank, a flexible sensing unit is integrated on a welding gun support to synchronously collect electromagnetic eddy current phase shift signals and optical visual depth of field signals of the welding seam area of the fire extinguisher tank. The method comprises the following steps: A flexible sensing unit is installed at the front end of the welding gun support, the flexible sensing unit comprises eddy current probes and optical probes arranged side by side, and the detection ends of the eddy current probes and the optical probes are directed towards the welding seam area of the fire extinguisher tank; When the welding gun starts welding, a synchronous trigger pulse is generated synchronously, the synchronous trigger pulse simultaneously starts the eddy current probes in the flexible sensing unit to emit electromagnetic wave beams and the optical probes to emit laser beams; 3. The method of claim 1, wherein, The eddy current probes receive the reflected electromagnetic waves of the welding seam area of the fire extinguisher tank and measure the phase shift amount of the electromagnetic waves, and at the same time, the optical probes receive the reflected laser beams of the welding seam area of the fire extinguisher tank and measure the depth of field change amount of the laser beams; The phase shift amount is converted into electromagnetic eddy current phase shift signals, and the depth of field change amount is converted into optical visual depth of field signals. The electromagnetic eddy current phase shift signals and the optical visual depth of field signals are fused to generate a tomographic fusion matrix, which is used to quantify the fusion state and potential defect distribution between different welding beads in the welding seam area of the fire extinguisher tank. The method comprises the following steps: The electromagnetic eddy current phase shift signals and the optical visual depth of field signals are aligned according to the same time stamp and spatial coordinates; Phase shift characteristic values of the aligned electromagnetic eddy current phase shift signals are extracted, the phase shift characteristic values describe the change degree of the electromagnetic characteristics of the welding seam area of the fire extinguisher tank; Depth of field change characteristic values of the aligned optical visual depth of field signals are extracted, the depth of field change characteristic values describe the fluctuation degree of the surface topography of the welding seam area of the fire extinguisher tank; The phase shift characteristic values and the depth of field change characteristic values are weighted and fused according to a preset weight ratio to generate fusion characteristic values of each spatial position point; The fusion characteristic values of all the spatial position points are arranged according to a preset spatial coordinate grid to form a tomographic fusion matrix quantifying the fusion state and defect distribution of the welding bead.

4. The method of claim 1, wherein, Obtaining a feature distribution of the tomographic fusion matrix, and dynamically adjusting a detection focal length and a scanning track of the eddy current probe and the optical probe in the flexible sensing unit according to the feature distribution to output a topological atlas characterizing the interlayer bonding defects of the weld, including: Extracting a fusion feature value of each spatial position point from the tomographic fusion matrix, and generating a spatial coordinate marker of the spatial position point when the fusion feature value exceeds a preset defect feature threshold value; Calculating a target focusing position of the eddy current probe and the optical probe according to the spatial coordinate marker, and converting the target focusing position into an elevation angle adjustment amount and a horizontal angle adjustment amount of the eddy current probe and the optical probe; Adjusting a mechanical structure of the flexible sensing unit according to the elevation angle adjustment amount and the horizontal angle adjustment amount, so that the detection focal length of the eddy current probe and the optical probe is aligned with the weld seam area of the fire extinguisher tank, and the scanning track of the eddy current probe and the optical probe is planned according to the distribution path of the spatial coordinate marker; Collecting weld area data under the adjusted detection focal length and scanning track, and converting the weld area data into a topological atlas characterizing the interlayer bonding defects of the weld.

5. The method as claimed in claim 1, wherein, Performing eddy current phase decoupling processing on the topological atlas to separate the surface electromagnetic response and the deep electromagnetic response, and constructing a defect distribution vector according to the surface electromagnetic response and the deep electromagnetic response, including: Extracting eddy current phase raw data of an electromagnetic eddy current phase shift signal of each spatial position point in the topological atlas, and performing frequency domain decomposition processing on the eddy current phase raw data to separate the surface electromagnetic response and the deep electromagnetic response; Calculating a signal intensity value of the surface electromagnetic response and a signal intensity value of the deep electromagnetic response, and taking a difference value of the signal intensity values as a response difference amount of the spatial position point; Binding the response difference amount with the coordinates of the corresponding spatial position point to form a defect feature data unit; Integrating the defect feature data units of all spatial position points, and arranging the defect feature data units in a spatial grid sequence to form a defect distribution vector.

6. The method of claim 5, wherein, Extracting eddy current phase raw data of an electromagnetic eddy current phase shift signal of each spatial position point in the topological atlas, and performing frequency domain decomposition processing on the eddy current phase raw data to separate the surface electromagnetic response and the deep electromagnetic response, including: Extracting eddy current phase raw data of an electromagnetic eddy current phase shift signal of each spatial position point from the topological atlas, the eddy current phase raw data containing an original phase response value of an electromagnetic wave in the weld material; Converting the original phase response value into a frequency domain frequency spectrum distribution, and dividing the frequency spectrum range of the frequency domain frequency spectrum distribution by a preset first cutoff frequency and a second cutoff frequency; Extracting a frequency spectrum component above the first cutoff frequency, and converting the frequency spectrum component into a time domain signal as the surface electromagnetic response; Extracting a frequency spectrum component below the second cutoff frequency, and converting the frequency spectrum component into a time domain signal as the deep electromagnetic response, while outputting the surface electromagnetic response and the deep electromagnetic response according to the spatial position points.

7. The method as claimed in claim 1, wherein, adopting a graph convolutional neural network to process the defect distribution vector to generate a welding parameter correction instruction, and feeding back the welding parameter correction instruction to a welding power source to adjust welding power, comprising: writing a defect feature data unit in the defect distribution vector into a node feature extraction layer of a graph convolutional neural network to extract a defect feature vector of each spatial position point; obtaining spatial coordinates of the defect distribution vector, and constructing an adjacency relationship matrix of a graph structure according to the spatial coordinates, and connecting adjacent spatial position points of the adjacency relationship matrix to form a graph node network; aggregating defect feature vectors of adjacent nodes in the graph node network through a message passing mechanism of the graph convolutional neural network to generate a fused defect feature quantity; mapping the fused defect feature quantity into a welding parameter correction quantity, the welding parameter correction quantity including a welding power adjustment value and a welding speed adjustment value; packaging the welding parameter correction quantity into a welding parameter correction instruction and sending it to a welding power source, the welding power source adjusting output power and travel speed according to the welding parameter correction instruction.

8. A multi-source sensing based intelligent monitoring and control system for weld quality of fire extinguisher tank, characterized in that, comprising: a collection module for synchronously collecting electromagnetic eddy current phase shift signals and optical visual depth of field signals of a fire extinguisher tank body weld area on a welding torch support during the welding process of the fire extinguisher tank body ring seam and longitudinal seam; a fusion module for fusing the electromagnetic eddy current phase shift signals and the optical visual depth of field signals to generate a tomographic fusion matrix, the tomographic fusion matrix being used to quantify the fusion state and potential defect distribution between different weld beads in the fire extinguisher tank body weld area; an adjustment module for obtaining a feature distribution of the tomographic fusion matrix, and dynamically adjusting the detection focal length and scanning trajectory of the eddy current probe and optical probe in the flexible sensing unit according to the feature distribution to output a topological atlas representing the interlayer bonding defects of the weld beads; a construction module for performing eddy current phase decoupling processing on the topological atlas to separate surface electromagnetic response and deep electromagnetic response, and constructing a defect distribution vector according to the surface electromagnetic response and the deep electromagnetic response; a feedback module for adopting a graph convolutional neural network to process the defect distribution vector to generate a welding parameter correction instruction, and feeding back the welding parameter correction instruction to a welding power source to adjust welding power.

9. A computing device, comprising: comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement the intelligent monitoring and control method of fire extinguisher tank weld quality based on multi-source sensing according to any one of claims 1 to 7.

10. A computer storage medium, characterized in that, a computer program is stored, and when the computer program is executed by a computer, the intelligent monitoring and control method of fire extinguisher tank weld quality based on multi-source sensing according to any one of claims 1 to 7 is realized.

Images