Bearing defect detection system and method based on multi-combination lightsource and multi-view multi-channel YOLO network structure detection algorithm
A multi-view 3D appearance defect detection system using a 2D camera and turntable with a YOLO network algorithm addresses the cost and inefficiency of existing systems, providing accurate and robust 3D defect detection for bearings.
Patent Information
- Application Number
- PCT/CN2024/083812
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2024-03-26
- Publication Date
- 2025-10-02
AI Technical Summary
Existing 2D bearing detection systems fail to accurately detect three-dimensional defects due to their cost and inefficiency, while 3D cameras are expensive and not economically viable for many enterprises.
A multi-view 3D appearance defect detection system using a 2D camera and turntable, combined with a multi-channel YOLO network structure detection algorithm, including image acquisition, pre-processing, and deep learning modules for enhanced defect detection.
The system achieves high detection accuracy and robustness for 3D defects at a lower cost, making it economically viable for enterprises.
Smart Images

Figure CN2024083812_02102025_PF_FP_ABST
Abstract
Description
Bearing defect detection system and method based on multi-combination lightsource and multi-view multi-channel YOLO network structure detection algorithm
[0001] Technology
[0002] The present application relates to a visual detection system using a 2D camera for bearing 3D defects and a multi-angle combination light source indirectly obtained through a rotating platform for the detection system, which belongs to the technical field of machine visual defect detection.
[0003] Background technology
[0004] Bearings are an important basic component of all kinds of machinery and equipment, and their accuracy, performance, life and reliability play a decisive role in the accuracy, performance, life and reliability of the host. In mechanical products, bearings belong to high-precision products, not only need mathematics, physics and many other disciplines of the theory of comprehensive support, but also need materials science, heat treatment technology, precision processing and measurement technology, numerical control technology and effective numerical methods and powerful computer technology and many other disciplines to serve, so bearings are a representative of the country's scientific and technological strength of the product.
[0005] In particular, small bearings, its appearance defect detection usually exists from a single direction when the light can not be extracted, especially scratches, pits, burrs and other three-dimensional defects, it is difficult to judge from one direction, resulting in the existing 2D bearing testing equipment or systems can not fully detect the three-dimensional defects of bearings; In view of this type of defect, most of the use of 3D cameras, but because 3D cameras rely on imports and expensive can not meet the actual application needs of enterprises, the present invention proposes a bearing based on 2D cameras and turntables multi-view 3D defect detection system and methods, can be low-cost premise, to solve the existing bearing manufacturers face three-dimensional defect detection Problems in technology.
[0006] The content of the invention
[0007] In order to solve the defects existing in the above-used technology, the present invention proposes a bearing defect detection system and method based on multi-combination lightsource and multi-view multi-channel YOLO network structure detection algorithm.
[0008] 1) Technical issues to resolve
[0009] The purpose of the present invention is to provide a low-cost and efficient bearing 3D appearance defect detection system and method based on 2D camera for the detection of surface 3D defects of existing bearings, the existence of 2D camera scheme detection accuracy and efficiency, 3D sensor scheme is expensive, etc.
[0010] 2) Technical solutions
[0011] The invention adopts the followingtechnical scheme:
[0012] A three-dimensional appearance defect detection method based on a multi-view view 3D appearance defect detection system based on a 2D camera and turntable is implemented, including a 2Dimage acquisition module, animage pre-processing module, an image defect detection module, and a 2D image acquisition module including a combined light source, a stepper motor rotation platform, and a stepper motor rotation platform 2D industrial cameras and lenses; the multi-angle bearing images captured by the 2D image acquisition module are transmitted to the image pre-processing module; the image pre-processing module is one or more of the filtering, character template positioning, channel separation and fusion, splicing and matching algorithms to obtain preprocessed images for defect detection; and the image defect detection module includes a multi-channel network and pre-training and adjustable deep learning models.
[0013] Preferably, Preferably, it also includes a deep learning model generation and optimization adjustment module, the deep learning model generation module includes a training multi-channel image input module, an image high-level feature extraction module, ahigh-level feature image differential module, adefect detection module and a attention mechanism learning module; to extract the image high-level features, output the high-level feature image differential module to make the defect bearing's high-rise feature map with the standard template as the same high-level feature map, get the defect high-level feature map, the defect detection module input the differential result feature map into the detection network to detect defects, the attention mechanism learning module includes spatial attention mechanism and channel attention mechanism and adaptive selection mechanism of multiple branches, Realize the regression detection network model for increasing the weight of the obtained defect site.
[0014] The technical scheme of the present invention can solve the technical problems such as the high price of the system, the low detection accuracy and the poor robustness of the system in the surface 3D defect detection technology of the existing bearing production enterprise, and the technical scheme of the present invention can be realized under the premise that the system acceptable to the enterprise is lower cost, with high detection accuracy and robustness.
[0015] Any skilled technical personnel familiar with the technical field of the present invention revealed within the scope of the present invention, according to the technical scheme of the present invention and its improved ideas to be equivalent to replace or change, shall be covered within the scope of protection of the present invention.
[0016] The content of the invention
[0017] The present invention is described in further detail in the following combination of drawings and embodiments:
[0018] Figure 1 is a rendering diagram of the overall structure of a bearing defect detection system and method based on multi-combination lightsource and multi-view multi-channel YOLO network structure detection algorithm;
[0019] Figure 2 shows the overall implementation flowchart of the detection system of the present invention;
[0020] Figure 3 shows the specific composition of the modules of the detection system of the present invention;
[0021] Figure 4 is a flowchart of the detection algorithm of the detection system of the present invention;
[0022] Figure 5 is a multi-angle image obtained by the detection target small bearing through the platform with a step length of 90degrees and a rotation for one week by coaxial light and ring-type light source;
[0023] Figure 6 is an image of the detection results of the detection system for three-dimensional defects in the bearing surface of the present invention.
Claims
1.a bearing defect detection system and method based on multi-combination lightsource and multi-view multi-channel YOLO network structure detection algorithm, characterized by:S1: For bearings of different sizes, choose a different light source combination and use the turntring to set out the out-of-steplength (360x / N) , (N≥3) , complete shooting the bearing surface at N angles under the light source to obtain N sheets Bearing image to be tested;S2: Pre-processing the sequence of N sheets to be detected obtained;S3: the N image obtained through the turntable at different angles of the light source is entered into the pre-trained N-channel deep learning model and the result is dedjuaned;S4: Store the judgment results and the N-sheet detection image for the optimization and modification of the deep learning model.2.According to claim 1, a three-dimensional defect detection method based on a turntable and a 2D camera is characterized by one or more methods of filtering, character template positioning, channel separation and fusion, stitching and matching used in the pre-processing;3.According to claim 1, a multi-view multi-view 3D defect detection method based on 2D camera and turntable is characterized by S0: training the improved YOLO multi-channel neural network;The S0 includes:S0-1: Using the combination of the light source, through the step setting of the turntable, the N bearing surface image of the bearing at different angles of the light source is obtained;S0-2: the captured images at N angles are pre-processed and multiple detected preprocessed images are obtained;S0-3: The detection of the N detected preprocessed images is divided into the N channel processing of the above, and the processing image or subset of the data of the N-detecting is obtained;S0-4: Using the proposed N-channel fitting to improve the YOLO network detection method, set a number of deep learning model generation algorithms with different defect detection task objectives;S0-5: Using N-channel import of the training set, using the image feature extraction algorithm to parameter fit the sample image set, seeking to fit the image set with the largest set of parameter combinations with the contrast of the background area, obtaining the multi-channel washing and defining it as the optimal digital combination, and making the fitted image set into a supervised data set;S0-6: The feature set is imported into the deep learning model with different defect detection task objectives, and multiple deep learning models are obtainedS0-7: Themodel accuracy verification is carried out according to the deep learning test results, and the training parameters of the deep learning model are adjusted.
Citation Information
Patent Citations
Appearance defect detection method and system based on multi-light-source fusion
CN110473178A
Bearing three-dimensional defect detection method and system
CN113111828A
Three-dimensional defect detection system and method for bearing
CN116519700A
A 2D image acquisition device and a bearing three-dimensional defect detection device
CN215066100U
Method for defect detection for rolling elements
US20230408421A1
Cited By
Thrust bearing surface damage large model detection method and system
CN121616593A
A thrust bearing surface damage large model detection method and system
CN121616593B