Search

CN-122007594-A - Aluminum alloy friction stir welding defect identification and repair method based on artificial intelligence

CN122007594ACN 122007594 ACN122007594 ACN 122007594ACN-122007594-A

Abstract

The invention relates to the technical field of industrial vision detection, and discloses an aluminum alloy friction stir welding defect identification and repair method based on artificial intelligence, wherein multi-source alignment data are constructed under a unified welding seam path coordinate system by collecting time sequence data, apparent-geometric data of a welding seam and internal defect response data, and the welding seam defect is identified and positioned by utilizing a multi-mode feature extraction and fusion reasoning model; and after obtaining the defect position, boundary and depth information, generating a repair decision and a repair track, and forming closed loop control of recognition, repair and quality verification through a rechecking mechanism. The invention realizes the linkage control of the identification and repair of the welding defects, and improves the accuracy of defect identification and the intelligent level of welding quality control under complex welding working conditions.

Inventors

  • LIU JIE
  • YAN ZIHAO

Assignees

  • 长春工程学院

Dates

Publication Date
20260512
Application Date
20260413

Claims (7)

  1. 1. The aluminum alloy friction stir welding defect identification and repair method based on artificial intelligence is characterized by comprising the following steps of: Step S1, establishing a workpiece coordinate system and a task queue; Step S2, performing welding according to a welding line path in the task queue to form a welding process time sequence tensor; Step S3, performing surface scanning imaging according to a welding seam path to obtain a welding seam apparent image, and collecting three-dimensional contour data of the weld seam residual height, the recess, the groove and the appearance of the weld toe; S4, acquiring internal defect response data, processing the internal defect response data according to a workpiece coordinate system, enabling the internal defect response data to establish a corresponding relation with weld apparent-geometric joint data and a welding process time sequence tensor, and constructing three-source alignment data; S5, performing fusion reasoning on the three-source alignment data by adopting a multi-mode feature extraction network, and executing weld defect recognition and outputting a defect candidate set, wherein the multi-mode feature extraction network comprises a time sequence coding branch, an apparent-geometric coding branch, an internal response coding branch and a defect reasoning branch; step 6, grading decision is carried out on the defect candidate set, and repair decision is output to carry out repair planning; and step S7, after the repair is completed, repeatedly executing the step S3 and the step S4, obtaining weld apparent-geometric combined data and internal defect response data again, generating repaired reinspection data, inputting the repaired reinspection data into a multi-mode feature extraction network for reinspection reasoning, and executing quality judgment.
  2. 2. The artificial intelligence based aluminum alloy friction stir welding defect identification and repair method according to claim 1, wherein the method comprises the following steps: the time sequence coding branch adopts a time sequence convolution network to code a welding process time sequence tensor to obtain a process state embedding; Performing defect region detection, boundary positioning and visual feature extraction on weld apparent-geometric joint data by using a YOLO detection network in an apparent-geometric coding branch to obtain surface morphology embedding; the internal response coding branch adopts a U-Net segmentation network to segment and code the characteristic of an abnormal response region in the internal defect response data, so as to obtain an internal response embedding; The method comprises the steps of obtaining a multi-modal fusion characteristic by aligning and fusing process state embedding, surface morphology embedding and internal response embedding through a cross-modal attention mechanism, introducing Jacobian regular constraint for alignment in the learning process of the multi-modal fusion characteristic, restraining excessive response of the multi-modal fusion characteristic to micro disturbance by limiting Jacobian sensitivity of a fusion model in the defect reasoning branch in an anti-rising direction generated by inner-layer disturbance optimization, obtaining a robust fusion characteristic, constructing context state embedding according to the robust fusion characteristic, performing adaptive filtering estimation on the robust fusion characteristic, obtaining defect state characterization, and generating a defect candidate set based on the defect state characterization.
  3. 3. The artificial intelligence based aluminum alloy friction stir welding defect identification and repair method according to claim 2 is characterized in that the robust fusion characteristic acquisition process comprises the following steps: R1, constructing an inner-layer disturbance optimization process based on multi-mode fusion characteristics, presetting a disturbance constraint domain, introducing disturbance variables in the disturbance constraint domain, defining an inner-layer loss function as a disturbance optimization target, and gradually iterating the disturbance variables in a projection gradient rising mode under the limitation of the disturbance constraint domain to form an anti-disturbance track; r2, extracting gradient vectors of an inner-layer disturbance optimization process relative to disturbance variables along an anti-disturbance track, performing normalization processing on the gradient vectors to obtain an anti-ascending direction, and establishing a track alignment relation; R3, calculating the Jacobian direction amplification amount of the fusion model in the corresponding countermeasure ascending direction on the basis of the track alignment relation, and constructing a countermeasure alignment Jacobian regular term; R4, merging the alignment-resisting Jacobian regular term into an outer layer training target of the fusion model, and carrying out joint weighting with an inner layer loss function to form a total loss function; and R5, acquiring an updated fusion model after parameter updating is completed, carrying out reasoning calculation on the updated fusion model, and outputting robust fusion characteristics.
  4. 4. The method for identifying and repairing an aluminum alloy friction stir welding defect according to claim 3, wherein the inner layer loss function is composed of one or more of a defect position prediction loss, a defect boundary segmentation loss, a defect depth loss, and a defect type judgment loss.
  5. 5. The method for identifying and repairing an aluminum alloy friction stir welding defect based on artificial intelligence according to claim 3, wherein an alignment-resistant Jacobian regularization term is used for measuring a degree of directional sensitivity amplification of the fusion model in an ascending-resistant direction on an anti-disturbance trajectory.
  6. 6. The method for identifying and repairing the defects of the friction stir welding of the aluminum alloy based on the artificial intelligence according to claim 2, wherein the acquisition process of the candidate set of defects is characterized by comprising the following steps: the method comprises the steps of B1, obtaining a current robust fusion feature, a last moment defect state estimation result and a state covariance matrix, constructing a prediction state sigma point set, and obtaining a prediction observation sigma point set through a defect observation mapping process; Step B2, performing linear projection, layer normalization and nonlinear mapping on the observation residual sequence, and performing recursive encoding to obtain context state embedding; Step B3, embedding the context state into an input strategy network, generating an unscented mean weight parameter vector and a covariance weight parameter vector through a context projection and weight synthesis head, and carrying out Softmax normalization constraint to obtain a sigma point mean weight set and a covariance weight set; Step B4, weighting calculation is carried out on the prediction state sigma point set and the prediction observation sigma point set by utilizing the sigma point mean weight set and the covariance weight set to obtain a prediction state mean value, a prediction state covariance matrix, a prediction observation mean value and a prediction observation covariance matrix; Step B5, performing adaptive filtering update by using Kalman gain and the current robust fusion characteristic to obtain stable defect state representation; and step B6, performing spatial position decoding, boundary range recovery and geometric parameter inversion on the defect state characterization to form a type judgment result and generating a defect candidate set.
  7. 7. The method for identifying and repairing aluminum alloy friction stir welding defects according to claim 6, wherein the candidate set of defects comprises defect locations, defect boundaries, defect depths, defect types, and defect confidence.

Description

Aluminum alloy friction stir welding defect identification and repair method based on artificial intelligence Technical Field The invention relates to the technical field of industrial visual inspection, in particular to an aluminum alloy friction stir welding defect identification and repair method based on artificial intelligence. Background The aluminum alloy material has the characteristics of low density, high specific strength, excellent corrosion resistance and the like, and is widely applied to the fields of aerospace, rail transit, ship manufacturing, new energy automobile structural parts and the like. Friction stir welding, a typical solid phase joining technique, generates frictional heat in a welding region by rotating a stirring tool and mechanically stirs plastic materials to form a high quality welded joint, is widely used in the manufacture of aluminum alloy structures. However, in the actual welding process, due to the influence of factors such as fluctuation of welding parameters, non-uniformity of material structure, tool wear, and variation of heat input distribution, various defect types such as tunnel defects, holes, unwelded parts, root defects, and surface grooves may still be generated in the weld joint region. Aiming at the problems, the prior art has started to introduce a machine vision detection method, a nondestructive detection method and a weld defect identification method based on deep learning, and the automatic identification and classification of the weld defects are realized through image analysis or signal analysis. However, most of the existing intelligent detection technologies analyze a single data source, for example, defect identification is performed only based on a visual image or a single nondestructive detection image of a weld surface, and it is difficult to simultaneously perform joint analysis by combining welding process signals, apparent geometric information of the weld and internal defect response information, so that problems of insufficient defect identification stability and limited positioning accuracy are easy to occur under a complex welding working condition. Meanwhile, the prior art is usually focused on the detection or identification process of the weld defects, and lacks a unified data association and automatic planning mechanism between defect identification results and subsequent repair decisions, so that a closed-loop control flow from defect identification, defect positioning to repair process generation and repair result rechecking is difficult to form. Disclosure of Invention The invention provides an aluminum alloy friction stir welding defect identification and repair method based on artificial intelligence, aiming at the problems that the prior intelligent welding detection technology is insufficient in multi-source information utilization, the defect identification stability is affected by disturbance under complex working conditions, the defect identification and repair process lacks linkage control and the like. The method comprises the steps of establishing a unified path coordinate alignment mechanism of welding process time sequence signals, weld apparent-geometric information and internal defect response data, establishing a corresponding relation among welding process state information, weld surface morphology information and weld internal structure information in the same weld path space, establishing a multi-mode feature extraction and fusion reasoning model on the basis, carrying out joint coding on welding process signals, visual images and nondestructive detection images, realizing structure alignment and feature fusion among different data modes through a cross-mode attention mechanism, improving the spatial positioning capability and type determination capability of defect identification, introducing anti-alignment Jacobian regular constraint into the multi-mode fusion model, enabling the model to still maintain stable defect characterization capability under the conditions of welding process fluctuation, imaging disturbance and detection noise by constructing an internal layer disturbance optimization track and inhibiting the local sensitivity amplification effect of the model, further combining with dynamic filtering and state updating of the fusion features based on the self-adaptive state estimation mechanism of non-trace transformation, obtaining stable defect positions, boundaries and depth characterization, automatically generating a defect decision parameter and a structure safety threshold according to the self-adaptive decision mechanism, enabling the defect parameter and structure safety threshold to automatically generate a defect identification and repair and self-control and carrying out intelligent repair process recognition and repair quality control, and complete repair and intelligent repair process identification, and fault quality verification and repair verification. The invention provides an artif