Search

CN-121982020-A - Real-time identification method for surface crack defects of complex component

CN121982020ACN 121982020 ACN121982020 ACN 121982020ACN-121982020-A

Abstract

The invention discloses a real-time identification method of a surface crack defect of a complex component, which relates to the technical field of image detection and comprises the following steps of obtaining image frames and acquisition quality metadata synchronously acquired by multiple cameras, establishing a mapping relation between the image frames and surface units based on reference geometry, selecting high-resolution image blocks according to a process risk map, reconnaissance uncertainty and budget constraint, performing crack candidate reasoning on the high-resolution image blocks, mapping a result into surface unit evidence, performing quality gating and state updating, aggregating and confirming state surface units to form crack objects, outputting judging results and evidence packages, and feeding back uncertain state surface units as priority objects of subsequent acquisition groups. According to the invention, crack positioning consistency can be improved under the condition of a complex curved surface, bad frame interference is reduced, detection beat and rechecking tracing are considered, multi-view evidence merging stability is improved, and repeated alarm and cold region missing detection risks are restrained.

Inventors

  • LU YEJUN
  • LI JUNWEI

Assignees

  • 陕西广大重型机械有限公司

Dates

Publication Date
20260505
Application Date
20260401

Claims (10)

  1. 1. The real-time identification method of the surface crack defect of the complex component comprises the steps of obtaining multi-view and/or multi-time image frames of a workpiece to be tested, Dispersing the surface of a workpiece to be measured into a plurality of surface units based on a reference geometry, establishing projection corresponding relation between each image frame and the surface units, and simultaneously acquiring acquisition quality metadata corresponding to each image frame; performing low-resolution reconnaissance on each image frame, and selecting a corresponding high-resolution image block according to the risk prior and reconnaissance uncertainty of the surface unit under the constraint of high-resolution pixel budget and delay budget; Performing crack candidate reasoning on the high-resolution image block, mapping the obtained candidate crack evidence to a corresponding surface unit, and performing gating fusion and state update according to the acquired quality metadata; And forming a crack object according to the adjacent surface units in the confirmation state, outputting a judging result and an evidence packet, and feeding the surface units in the uncertain state back to be a priority object of a subsequent acquisition group.
  2. 2. The method for identifying the surface crack defects of the complex component in real time according to claim 1, wherein the method comprises the following steps: The acquisition quality metadata comprises a time stamp, an action delay mark, a frame triggering loss mark, a frame starting over-triggering mark, a link retransmission count, an ambiguity, saturation and a registration residual error, and the edge computing equipment carries out low-reliability input marks on each image frame according to the acquisition quality metadata and sends the image frames into a weight-reducing path when the comprehensive quality is lower than a low-reliability threshold value and sends the image frames into a candidate queue for supplement acquisition when the comprehensive quality is lower than an abnormal threshold value.
  3. 3. The method for identifying the surface crack defects of the complex component in real time according to claim 2, wherein the method comprises the following steps: The edge computing device performs global registration and local registration on each image frame to generate a corresponding set of visible surface elements and projection correspondence between the image frame and the surface elements.
  4. 4. The method for identifying the surface crack defects of the complex component in real time according to claim 3, wherein the method comprises the following steps: The edge computing device firstly generates a low-resolution reconnaissance image for each image frame, then maps the rough crack response and reconnaissance uncertainty to surface units to form a surface unit level reconnaissance result, and merges the surface units which are spatially adjacent and compatible in projection into a surface unit cluster according to continuous suspicious responses of the adjacent surface units to generate a corresponding high-resolution image block.
  5. 5. The method for identifying the surface crack defects of the complex component in real time according to claim 4, which is characterized in that: The selection of the high-resolution image blocks is simultaneously constrained by the high-resolution pixel budget, the maximum image block number and the residual delay budget, and after the priority selection is completed, the edge computing equipment supplements the high-resolution image blocks from the low-risk surface unit according to the bottom-protection exploration proportion, so that the high-resolution image block request set covers the positions to be verified in the high-risk area and the low-risk area simultaneously.
  6. 6. The method for identifying the surface crack defects of the complex component in real time according to claim 5, which is characterized in that: After performing crack candidate reasoning on the high-resolution image block, the edge computing equipment extracts a crack mask, a framework, a local width and a local confidence, and reflects the crack mask, the framework and the local width to corresponding surface units according to projection corresponding relations between the image frames and the surface units so as to form surface unit evidence bound with the source image frames.
  7. 7. The method for identifying the surface crack defects of the complex component in real time according to claim 6, wherein the method comprises the following steps: And the edge computing equipment executes three-section quality gating on each surface unit according to the acquired quality metadata and the surface unit evidence, wherein the surface unit evidence with the quality higher than a second threshold enters a direct fusion path, the surface unit evidence with the quality between the first threshold and the second threshold enters a low-weight slow fusion path, the surface unit evidence with the quality lower than the first threshold enters a complementary acquisition judging path, and the surface unit evidence is transferred into an uncertain path when the residual delay budget is insufficient.
  8. 8. The method for identifying the surface crack defects of the complex component in real time according to claim 7, wherein the method comprises the following steps: The edge computing device maintains candidate, validation, uncertainty, and suppression states for each surface unit, updates the surface unit cluster to the validation state only when the same surface unit cluster satisfies the independent support number, adjacency continuity, and width smoothness conditions simultaneously, maintains the candidate state for surface units supported by only a single low quality view, and transitions to the uncertainty state when the remaining delay budget is insufficient.
  9. 9. The method for identifying the surface crack defects of the complex component in real time according to claim 8, wherein the method comprises the following steps: The edge computing equipment aggregates surface units which are in a confirmation state and are continuous in topology into crack objects, calculates the length, the average width, the maximum width and the branch number according to the skeleton length and the local width corresponding to the crack objects, and simultaneously generates an evidence packet, wherein the evidence packet comprises a support image frame index, a corresponding surface unit position, a crack local image block, a quality metadata track and a quantification result.
  10. 10. The method for identifying the surface crack defects of the complex component in real time according to claim 9, wherein the method comprises the following steps: If a certain camera continuously generates abnormal events in delayed action, lost frame triggering and over triggering, the evidence weight of the camera is reduced, and adjacent visual angles are preferentially called in the subsequent acquisition group to carry out compensation acquisition on the corresponding region.

Description

Real-time identification method for surface crack defects of complex component Technical Field The invention relates to the technical field of image detection, in particular to a video real-time identification method for surface crack defects of a complex component. Background For die-casting aluminum shells, stamping forming parts, welding parts, blades and other complex components, on an actual production line, a robot hand-held camera is adopted to conduct circular scanning on the surface of a workpiece or synchronously pick up images of all angles along a multi-camera, defect identification and result output are completed through edge calculation, and the conventional flow comprises image acquisition, scaling or blocking, crack extraction, defect judgment and manual checking, so that on-line screening of fine cracks on the surface is completed within a specified beat. In US12039441B2, a method and a system for detecting cracks based on a full convolution network are proposed, in which a video camera and a scanning mechanism are provided, the camera is moved along a surface to be detected by using the scanning mechanism to shoot and continuously shoot, so as to obtain continuous frame video, an overlapping area exists between the continuous frames, and the same crack can repeatedly appear in the continuous frames. And analyzing at least a part of video frames based on a full convolution network of the processor to obtain a video frame crack score map, and combining the frame scores under a space-time coordinate system by using a parameterized data fusion scheme to obtain a crack identification result. It is also mentioned that the method can be used for robotic inspection, full high definition and higher resolution video. On the other hand, the foreign patent document WO2024226041A1 proposes an "improved method for detecting surface features of parts manufactured on a production line", and for visual inspection of parts manufactured on a production line, one or more cameras are used to acquire images of a workpiece and to complete surface feature detection using composite image training, image block extraction, multi-field reasoning and result fusion, the specification also states that the resolution of industrial inspection images is high, model training and reasoning can be completed by image block extraction in different modes, and the US patent document US12347038B2 proposes a method for completing crack evaluation and visualization using a combination of two-dimensional crack recognition results and a three-dimensional grid model. The technology can realize surface crack or surface feature detection in a suitable scene, but has limitations in a complex curved surface metal component high-beat online detection scene. Because the fusion object of US12039441B2 is based on continuous video frames and space-time score graphs, when the object to be detected is a complex curved surface component with curvature change, shielding and strong reflection, the projection position, length appearance, contrast and the like of the same crack at different visual angles or different scanning positions can change along with the time, and the cross-frame evidence is unstable, and for the high-resolution detection route of WO2024226041A1 based on image blocks, if the whole detection route is enlarged for adapting to model input, the sub-millimeter crack can be weakened due to insufficient effective pixels when the original image is integrally scaled, and the time cost and the calculation power consumption can be increased when a large-size image is subjected to a large amount of fixed image block reasoning. Meanwhile, the multi-camera synchronous acquisition is also influenced by network transmission, trigger time sequence and imaging quality fluctuation, and Basler documents show that PTP synchronization accuracy depends on network hardware and configuration, setting of frame transmission delay and inter-packet delay can influence transmission congestion and frame rate, and events such as too late action, frame trigger loss, excessive frame initiation trigger and the like (which can be caused by improper setting) can occur in the acquisition process. After the above factors are overlapped, missing detection, false detection, repeated alarm, heavy manual rechecking burden and limited production line beat may be caused. Therefore, in the multi-camera high-resolution on-line detection scene of the metal component with the complex curved surface, how to improve the stability and reliability of recognition of the tiny cracks while guaranteeing the detection beat becomes the technical problem to be solved currently. Disclosure of Invention (One) solving the technical problems Aiming at the defects of the prior art, the invention provides a real-time identification method for surface crack defects of a complex component, which comprises the steps of selecting a high-resolution image block according to a process risk graph, re