CN-121978201-A - Weld defect real-time and off-line detection system and method

CN121978201ACN 121978201 ACN121978201 ACN 121978201ACN-121978201-A

Abstract

The invention discloses a weld defect real-time and off-line detection system and method. The device comprises a planar electromagnetic chromatography sensor array, a multichannel excitation and acquisition unit, a data processing and imaging unit, a prototype guided segmentation network unit and an information processing platform, wherein the data processing and imaging unit is in communication connection with the prototype guided segmentation network unit and can input a preprocessed image into the prototype guided segmentation network unit, a coder extracts characteristics, a decoder recovers spatial resolution step by step, a prototype similarity graph is calculated by utilizing prototype branches, then the prototype similarity graph is fused with the characteristics of the decoder, a defect probability graph P is output, and the automatic judgment of weld defects is realized after the processing of the information processing platform. The method has the advantages that high robustness, high sensitivity and low omission ratio can be realized on the surface of the complex welding seam, and the technical problems that the sensitivity for detecting the shallow, tiny and inclined defects is insufficient, the real-time imaging and the high-precision off-line reconstruction are difficult to be compatible, the omission ratio is difficult to be reduced and the like are solved.

Inventors

SHE SAIBO
YIN LIJIA
XIONG LEI
YIN WULIANG

Assignees

苏州耀迈测控科技有限公司

Dates

Publication Date: 20260505
Application Date: 20251230

Claims (8)

1. The real-time and offline weld defect detection system is characterized by comprising a planar electromagnetic chromatography sensor array, a multichannel excitation and acquisition unit, a data processing and imaging unit, an information processing platform and a weld defect automatic judging unit, wherein the multichannel excitation and acquisition unit is used for exciting and acquiring the planar electromagnetic chromatography sensor array, the data processing and imaging unit is connected with the multichannel excitation and acquisition unit and used for acquiring reference signals, calculating differential voltages, constructing a sensitivity matrix, reconstructing real-time LBP (local binary pattern), reconstructing offline LW (line-of-sight) and preprocessing reconstructed images, the data processing and imaging unit is provided with an encoder, a decoder, a prototype branch module and a similarity fusion module, the information processing platform is in communication connection with the prototype guide and division network unit, the data processing and imaging unit is in communication connection with the prototype guide and division network unit and can input the preprocessed images into the prototype guide and division network unit, the encoder extracts characteristics, the decoder recovers spatial resolution step by step, the prototype similarity map is calculated by utilizing the branch, and then the prototype similarity map is fused with the decoder characteristics, and the weld defect automatic judging is realized after the processing of the information processing platform.
2. The real-time and offline detection system for weld defects according to claim 1, wherein the planar electromagnetic chromatography sensor array is composed of a plurality of planar electromagnetic coils and a magnetic core which are arranged in a staggered manner in two rows, each coil is wound on the magnetic core, the planar electromagnetic chromatography sensor is arranged above a weld to be detected, the axis of the planar electromagnetic coils is in the vertical direction of the surface of the weld, the multichannel excitation and acquisition unit sequentially gates each coil as an excitation coil, the other coils are used as receiving coils for measurement, and measurement data can be transmitted to the data processing and imaging unit through UDP or a bus.
3. The weld defect real-time and offline detection system according to claim 1 or 2, wherein the data processing and imaging unit comprises a reference signal acquisition module for acquiring a reference voltage vector V ref under the condition of a defect-free weld or a flat plate, a differential voltage calculation module for acquiring a voltage vector V sample under the condition of a detected weld and calculating a differential signal Δv=v sample -V ref , a sensitivity matrix S based on finite element simulation and comprising only adjacent and next-adjacent coil pairs is constructed, the corresponding coil pair spacing is 1 time and 2 times the coil center-to-center spacing, a simplified sensitivity matrix construction module for discarding remote coil pairs, a real-time LBP reconstruction module for mapping differential voltages into a two-dimensional weld cross-section conductivity/permeability disturbance image by adopting a linear back projection algorithm σ=s T Δv, and an offline reconstruction module for obtaining a reconstructed image of higher spatial resolution and contrast by adopting a Landweber iteration algorithm σ (k+1) ＝σ (k) +αS T (ΔV-Sσ (k) ) for stored detection data.
4. The welding seam defect real-time and offline detection system according to claim 3 is characterized in that the encoder extracts multi-scale depth features based on a residual error network, the decoder adopts UNet jump connection to gradually up-sample the multi-scale features and restore spatial resolution, the prototype branching module carries out global aggregation on encoder high-level feature images to obtain prototype vectors of defect types and background types, cosine similarity is adopted to calculate similarity of each pixel position feature and the prototype, prototype similarity images are generated, the prototype is continuously updated in a training and reasoning process by adopting an exponential sliding average (EMA) updating strategy, and the prototype similarity images and decoder output features are spliced in channel dimensions by a similarity fusion module to generate a final defect probability image through 1X 1 convolution fusion.
5. The system for detecting the defects of the welding seam in real time and offline according to claim 1,2 or 4, wherein the information processing platform is further connected with a visualization and interaction unit.
6. The system for real-time and offline detection of weld defects according to claim 5, wherein said visualization and interaction unit is capable of providing a multi-window interface, displaying I/Q waveforms in real time, LBP reconstruction maps in real time, defect probability heatmaps and binary segmentation results and supporting user adjustment of thresholds, displaying historical frame sequences, switching real-time/offline mode operation.
7. A method for detecting weld defects by using the real-time and offline detection system according to any one of claims 1 to 6, comprising the following steps: S1, reference data acquisition Under the condition of a defect-free welding line or a reference test piece, the multichannel excitation and acquisition unit is controlled to sequentially excite each coil, the voltage vectors of all selected coil pairs are acquired, and the reference voltage V ref is calculated and stored. S2, data acquisition and difference of the welded seam to be detected Arranging a planar electromagnetic chromatography sensor array above a welded seam to be detected, repeating the excitation and collection process of the step S1 to obtain a detected voltage vector V sample , and calculating a differential signal DeltaV=V sample -V ref ; S3, real-time LBP imaging based on simplified sensitivity matrix Calculating a two-dimensional disturbance image sigma=S T DeltaV by linear back projection by using a simplified sensitivity matrix S only comprising adjacent and secondary adjacent coil pairs, and mapping the two-dimensional disturbance image sigma=S T DeltaV to a predefined grid to obtain real-time weld joint section imaging; S4, preprocessing the reconstructed image Normalizing and interpolating the real-time reconstructed image to a specified size, and executing lightweight morphological operation and denoising treatment to serve as input of a segmentation network; S5, prototype guided segmentation network reasoning Inputting the preprocessed image into a prototype guide segmentation network unit, extracting features by an encoder, recovering the spatial resolution step by a decoder, calculating a prototype similarity graph by utilizing prototype branches, and then fusing the prototype similarity graph with the features of the decoder to output a defect probability graph P; s6, post-segmentation processing and defect judgment The method comprises the steps of carrying out binarization on a defect probability map by adopting an adjustable threshold tau to obtain a defect region mask, removing isolated noise and filling small holes through morphological opening and closing operation, calculating parameters such as the area, the length, the position and the direction of the defect, and realizing automatic judgment and visual display of the weld defect.
8. The method for detecting the weld defects in real time and off-line according to claim 7, wherein the method further comprises an off-line high-precision reconstruction step after the step S6, wherein in the step, the stored detection data is subjected to off-line high-precision reconstruction by adopting Landweb er iteration algorithm and combining a full-sensitivity matrix according to the requirement, a weld imaging result with higher signal-to-noise ratio and spatial resolution is obtained, and then the weld imaging result is input into a prototype guided segmentation network unit for off-line high-precision segmentation and analysis.

Description

Weld defect real-time and off-line detection system and method Technical Field The invention relates to a weld defect real-time and off-line detection system and method based on a planar electromagnetic tomography (EMT) and deep learning segmentation network, and belongs to the technical field of electromagnetic nondestructive detection and industrial intelligent detection. Background Welds are key connection parts in pressure vessels, pipelines, rail transit and large steel structures, internal or surface defects (such as cracks, unfused, air holes, slag inclusions and the like) of the welds can obviously reduce the bearing capacity of the structures, and catastrophic accidents are caused when serious, so that research and development of nondestructive detection technology capable of realizing high sensitivity and high reliability under the condition of complex weld surfaces has important engineering significance, and the existing X-ray, ultrasonic detection and array eddy current detection methods are widely applied to weld detection, but still have the following defects: (1) Weld toe unevenness and rough surface interference are serious, namely weld wave fluctuation, splashing and polishing traces are usually present on the surface of a weld joint, so that ultrasonic detection is highly sensitive to coupling conditions, and a background signal of array vortex detection is easy to mask shallow or inclined defect signals; (2) The direction sensitivity is insufficient, the conventional array eddy current is usually measured by adopting only adjacent coil pairs, and the crack resolution capability for different orientations is limited; (3) The real-time performance and the precision of the imaging algorithm are difficult to be compatible, although the calculation speed of a non-iterative reconstruction algorithm such as Linear Back Projection (LBP), tikhonov regularization and the like is high, the imaging blurring and noise suppression capability is limited, and the iteration algorithm such as Landweb er (LW) has higher imaging precision, but the calculation amount is large, and is difficult to be used for online real-time detection; (4) The defect area is difficult to accurately divide, namely even if an electromagnetic tomography reconstruction image is obtained, the defect area is difficult to accurately divide under the condition of low contrast and small size defect scene by the traditional threshold value or simple image processing method, and particularly under the condition of limited sample number, the deep learning network is easy to be over-fitted and has high omission ratio. Therefore, an integrated detection scheme capable of realizing real-time coarse screening and high-precision offline analysis is needed to realize high robustness, high sensitivity and low omission factor on the surface of a complex welding seam. Disclosure of Invention The technical problem to be solved by the invention is to provide a real-time and off-line detection system and method for weld defects, which have the capabilities of real-time imaging, accurate segmentation and off-line high-fidelity reconstruction, so that high robustness, high sensitivity and low omission ratio can be realized on the surface of a complex weld. In order to solve the technical problems, the weld defect real-time and offline detection system comprises a planar electromagnetic chromatography sensor array, a multichannel excitation and acquisition unit for exciting and acquiring the planar electromagnetic chromatography sensor array, a data processing and imaging unit connected with the multichannel excitation and acquisition unit and used for carrying out reference signal acquisition, differential voltage calculation, sensitivity matrix construction, real-time LBP reconstruction, offline LW reconstruction and preprocessing on reconstructed images on data of the multichannel excitation and acquisition unit, a prototype guide segmentation network unit provided with an encoder, a decoder, a prototype branch module and a similarity fusion module, and an information processing platform in communication connection with the prototype guide segmentation network unit, wherein the data processing and imaging unit is in communication connection with the prototype guide segmentation network unit and can input the preprocessed images into the prototype guide segmentation network unit, the encoder extracts characteristics and the decoder restore spatial resolution step by step, calculates a prototype similarity map by utilizing the prototype branch, then fuses with the decoder characteristics, and the defect probability map P is output and is processed by the information processing platform to realize automatic determination of the weld defects. The planar electromagnetic chromatography sensor array consists of a plurality of planar electromagnetic coils and magnetic cores which are arranged in a staggered and staggered mode in two rows, each coil is wound