Search

CN-122023750-A - Thermos cup coating defect detection method and system

CN122023750ACN 122023750 ACN122023750 ACN 122023750ACN-122023750-A

Abstract

The invention relates to a vacuum cup coating defect detection method and system, which consists of a multi-angle image acquisition unit, a precise rotary positioning unit and an intelligent recognition host. The detection flow adopts a rotation synchronous control technology, and after the vacuum cup is clamped by the pneumatic clamp, the rotating mechanism is driven to rotate, and the multi-camera acquires 360-degree surface images and inputs the images into the preprocessing module. The image processing module pre-processes the original image signal through an industrial optimization flow, the processed image is subjected to an innovative 'double-flow fusion multi-mode perception YOLO deep learning model', the apparent characteristic flow and the structural characteristic flow are fused, the recognition rate is greatly improved, the capabilities of scratch enhancement recognition, highlight point elimination and the like are realized, meanwhile, multi-category defects are recognized, and the output module sends the recognition result to the report generating unit. And storing the defect identification result picture to provide a database for intelligent quality control.

Inventors

  • QIU XUKE

Assignees

  • 光知多(宁波)精密仪器有限公司

Dates

Publication Date
20260512
Application Date
20251231

Claims (10)

  1. 1. The vacuum cup coating defect detection method is characterized by comprising the following steps of: Step 1, the vacuum cup is self-adaptively clamped and rotated to be started, and a cup body positioning basis and a rotation angle theta are obtained; Step 2, synchronously acquiring multi-mode images based on the rotation positioning basis and the rotation angle theta in the step 1 to obtain an original multi-illumination image set and position metadata; step 3, preprocessing the image based on the original multi-illumination image set and the position metadata in the step2 to obtain an apparent feature map and a structural feature map; step 4, obtaining a single-frame defect prediction result based on the apparent feature map and the structural feature map of the step 3 and the position metadata of the step 2 through double-flow and multi-mode fusion reasoning; step 5, based on the single-frame prediction result in the step 4, multi-frame fusion and quantization analysis are carried out to obtain accurate defect quantization data; and 6, generating a report and a judging result based on the defect quantification data in the step 5, and completing detection closed loop.
  2. 2. The vacuum cup coating defect detection method according to claim 1 is characterized in that the vacuum cup is placed on a special pneumatic clamp, the clamp adaptively clamps vacuum cups of different sizes, an intelligent host machine sends an instruction through a CAN bus, a high-precision stepping motor is started to drive a rotating mechanism of the clamp, the clamp drives the vacuum cup to rotate at a constant angular speed at a constant speed, and an angle encoder feeds back the rotation angle theta of the vacuum cup in real time.
  3. 3. The thermos cup coating defect detection method according to claim 1 is characterized in that step 2 is specifically that based on a thermos cup rotation angle theta obtained in step 1, when the theta enters a preset acquisition start angle, a system triggers a synchronous signal, a plurality of groups of LED light sources in different directions are controlled to be lightened at high speed in sequence, high-resolution industrial cameras arranged at different heights synchronously trigger exposure, after the thermos cup rotates for one circle, a plurality of images at angle positions are acquired, a plurality of original images are obtained at each position, and meanwhile, the rotation angle theta of the thermos cup corresponding to each image is recorded as a position metadata label, so that an original multi-illumination image set and position metadata combination are formed.
  4. 4. The vacuum cup coating defect detection method according to claim 1, wherein step 3 specifically comprises the following steps of, based on the original multi-illumination image set collected in step 2 and the vacuum cup position metadata, executing a preprocessing procedure: step 3.1, ROI extraction, namely converting an image into an HSV color space, dividing a cup body region according to saturation and brightness thresholds of the cup body color of the vacuum cup, and removing interference of a background and a clamp; step 3.2, image enhancement and denoising, namely performing self-adaptive histogram equalization on the ROI region to enhance contrast, and removing noise through Gaussian filtering; And 3.3, generating multi-mode characteristics, namely obtaining a specular inhibition RGB image which is an apparent characteristic image inhibiting specular reflection through pixel weighted fusion on a plurality of images at the same position, calculating a surface normal image through a photometric stereo method, and obtaining a structural characteristic image which is a surface gradient image and is used for representing geometric deformation of the surface of the vacuum cup.
  5. 5. The thermos cup coating defect detection method according to claim 1, wherein step 4 specifically comprises loading a dual-flow YOLO deep learning model subjected to pruning and quantization optimization based on the apparent feature map and the structural feature map obtained in step 3 and the thermos cup position metadata θ recorded in step 2: Step 4.1, double-flow feature extraction, namely respectively inputting an apparent feature map and a structural feature map into an independent CNN backbone network, mapping position metadata theta into position feature vectors through an embedded layer, and modulating the feature map through conditional normalization to enhance the feature sensitivity of a defect high-incidence area of the vacuum cup; step 4.2, cross-modal fusion, namely carrying out BiFPN weighted fusion on a multi-scale feature pyramid output by a dual-backbone network, and automatically learning importance weights of different modal features by the network; and 4.3, single frame detection, namely inputting the fused characteristics into YOLOv detection heads, and predicting the boundary frame, confidence and class probability of the surface defects of the vacuum cup to obtain a single frame defect prediction result.
  6. 6. The thermos cup coating defect detection method according to claim 1, wherein the step 5 specifically comprises the following steps of: step 5.1, single frame redundancy elimination, namely eliminating overlapped prediction frames through non-maximum value inhibition; Step 5.2, cross-frame correlation, namely correlating detection results of the same defect in continuous multi-frames according to the position and the type of the predicted frame and the rotation speed of the vacuum cup to form a defect track; Step 5.3, confidence coefficient fusion, namely carrying out weighted average on multi-frame confidence coefficient of the same defect to obtain final confidence coefficient; And 5.4, quantitatively analyzing, namely reconstructing coordinates of the endpoints of scratches on the surface of the vacuum cup in a 3D space, calculating the actual length, and converting the equivalent diameter according to the pixel area of the black point and the calibration coefficient to finally obtain quantized data containing the defect type, the position, the confidence coefficient and the accurate size of the vacuum cup.
  7. 7. The thermos cup coating defect detection method according to claim 1 is characterized in that step 6 is specifically characterized in that based on the accurate defect quantification data obtained in step 5, a system generates a structural detection report, wherein the position of a defect in a thermos cup expansion chart is visually marked on a touch display screen, defect types, sizes and other information are synchronously displayed, the report and original image data are automatically stored in a database for quality tracing, whether a thermos cup is qualified or not is judged according to preset rules, a sorting mechanism is triggered to send the thermos cup to a corresponding flow direction, and finally the whole flow detection of the thermos cup coating defect is completed.
  8. 8. A vacuum cup coating defect detection system, which is used for applying the vacuum cup coating defect detection method according to any one of claims 1-7, and is characterized by comprising a hardware component and control software, wherein the hardware component realizes the functions of vacuum cup positioning, image acquisition, data processing and post detection execution, and the control software completes the whole flow control from equipment control, data processing to result output.
  9. 9. The thermos cup coating defect detection system of claim 8, wherein the hardware assembly comprises: The multi-angle structured light image acquisition unit is responsible for 360-degree dead angle-free image acquisition of the whole surface of the vacuum cup, optimizes the imaging effect of the high-reflection coating through multi-illumination and multi-spectrum imaging, and provides high-quality original image data for subsequent defect identification; the precise rotary positioning unit realizes self-adaptive stable clamping and precise uniform rotation of the vacuum cup, feeds back the rotation angle in real time, ensures the synchronism of image acquisition and cup rotation, and provides a position reference for defect positioning; the intelligent recognition host computer bears data processing tasks such as image preprocessing acceleration, deep learning model reasoning and the like, and realizes visual interaction of detection results and cooperative control of all hardware units; and the auxiliary execution assembly automatically completes qualified/unqualified sorting of the thermos cup according to the detection and judgment result, links the detection flow and the subsequent links of the production line, and ensures automatic closed loop of detection.
  10. 10. The thermos cup coating defect detection system of claim 8, wherein the control software comprises: the equipment control module is used for comprehensively controlling the clamp, the rotating mechanism, the camera, the light source and other hardware equipment, generating synchronous trigger signals, monitoring key parameters such as clamping pressure and the like, and ensuring accurate cooperative work of all hardware units; the image processing module is used for carrying out interference elimination, enhancement, feature extraction and other preprocessing on the original image to generate an apparent and structural bimodal feature map for adapting to defect recognition, so that the accuracy and the efficiency of subsequent recognition are improved; The defect identification module is used for completing bimodal feature fusion and defect reasoning by relying on the optimized deep learning model, realizing type identification, position location and multi-frame result optimization of the defects of the vacuum cup coating, and guaranteeing identification accuracy; the result report module integrates the defect identification data, generates a structured detection report, visually displays defect information, provides visual data support for quality analysis, and supports report export and viewing; the storage module is used for storing full-flow information such as original images, preprocessing data, recognition results and the like, establishing a mapping library of vacuum cup models and detection parameters, and providing data storage for quality tracing and model iterative training.

Description

Thermos cup coating defect detection method and system Technical Field The invention relates to the technical field of industrial automatic detection, in particular to a method and a system for detecting defects of a coating of a vacuum cup, which are particularly suitable for automatic quality detection of the quality of a layer in the production process of the vacuum cup. Background Along with the improvement of the automation degree of the vacuum cup manufacturing industry, the surface quality detection of the vacuum cup coating becomes a key link for ensuring the product quality. The traditional detection method has the following defects: 1. the manual detection efficiency is low, the manual visual inspection is relied on, the speed is low, and the fatigue is easy to occur; 2. the subjectivity is strong, and different inspectors judge the defect standard inconsistently; 3. The micro defect omission rate is high, and the defects such as micron-sized scratches, black spots and the like are difficult to find by human eyes; 4. Failing to quantitatively analyze, lack of accurate records of defect size, position and type; 5. The full surface coverage is difficult, and all angles of the coating of the vacuum cup are difficult to detect at one time. An Automatic Optical Inspection (AOI) system in the prior art mainly adopts a fixed camera and a traditional image processing algorithm, and has the problems of poor adaptability, high false inspection rate, incapability of identifying complex defects and the like. The optical detection system in the prior art is not organically integrated with the detection environment, and only the apparent characteristics of the image are detected independently. The input image stream is not collected, and the cross-modal feature fusion identification of the apparent feature stream and the structural feature stream double-flow fusion can not be carried out on the input image stream. Has no sensitivity to the characteristic of the cup body changing along with light. Resulting in limited precision improvement. Disclosure of Invention In summary, in order to solve the defects in the prior art, the invention aims to provide a high-efficiency and accurate thermos cup coating defect detection system and method, and solve the problems that the detection efficiency is low, the omission factor is high, the full surface detection cannot be realized and the like in the prior art. The method specifically comprises the following steps: 1. The high-speed intelligent defect identification capability is that the existing thermos cup surface detection scheme has complex flow or relies on manpower, and the existing identification algorithm cannot identify the defects of the thermos cup coating on an automatic production line and cannot realize the detection. 2. The technical scheme of the invention adopts a plurality of high-definition industrial cameras, and simultaneously integrates the small defect detection rate and the quick recognition capability through a high-efficiency YOLO model and a specific scene recognition optimization algorithm aiming at a vacuum cup. 3. The structural defect detection capability and the multi-mode fusion processing capability are insufficient, and the traditional scheme is only aimed at apparent picture characteristics and has weak structural characteristic processing capability. 4. The scheme can realize accurate recording and quantitative analysis of defect sizes, positions and types, and can be seamlessly integrated with digital quality control of intelligent factories. The technical scheme for solving the technical problems is as follows, the vacuum cup coating defect detection method comprises the following steps: Step 1, the vacuum cup is self-adaptively clamped and rotated to be started, and a cup body positioning basis and a rotation angle theta are obtained; Step 2, synchronously acquiring multi-mode images based on the rotation positioning basis and the rotation angle theta in the step 1 to obtain an original multi-illumination image set and position metadata; step 3, preprocessing the image based on the original multi-illumination image set and the position metadata in the step2 to obtain an apparent feature map and a structural feature map; step 4, obtaining a single-frame defect prediction result based on the apparent feature map and the structural feature map of the step 3 and the position metadata of the step 2 through double-flow and multi-mode fusion reasoning; step 5, based on the single-frame prediction result in the step 4, multi-frame fusion and quantization analysis are carried out to obtain accurate defect quantization data; and 6, generating a report and a judging result based on the defect quantification data in the step 5, and completing detection closed loop. Based on the technical scheme, the invention can be further improved as follows: The method comprises the steps of 1, placing a vacuum cup on a special pneumatic clamp, adaptively c