Defect identification method and system for gearbox shell of electro-tricycle

Abstract

The invention relates to the technical field of gearbox shell defect identification, and discloses an electro-tricycle gearbox shell defect identification method and system. According to the method, image data of the gearbox shell are acquired in real time through an industrial camera, and whether the gearbox shell has defects or not is judged by adopting a defect recognition model trained by an improved yolk v8 algorithm through data preprocessing and image super-resolution processing. Wherein the picture super-resolution model is obtained through training of an SRFormer algorithm, the picture resolution can be effectively improved, and detail features can be enhanced. The improved yolk v8 algorithm adopts Swin Transform to extract multi-scale features, contextual information is fused, and the position, the category and the confidence coefficient of the gearbox shell defect are accurately predicted. And when the defects are recognized, the gearbox shell with the defects is sorted out from the production line through sorting equipment. According to the method, the accuracy and efficiency of defect identification of the gearbox shell are improved, powerful support is provided for quality control of the gearbox shell of the electro-tricycle, and improvement of production efficiency is promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of gearbox housing defect recognition, and in particular to a gearbox housing defect recognition method and system for an electric tricycle. Background Art

[0002] In the manufacturing process of electric tricycles, the quality of the gearbox housing directly affects the performance and safety of the vehicle. Traditional manual inspection methods have problems such as low efficiency and large errors, which are difficult to meet the needs of large-scale production lines. Therefore, it is particularly important to develop an automated and high-precision gearbox housing defect identification method. Although existing automatic inspection technologies have been applied to many fields, there are still problems in the identification of gearbox housing defects, such as insufficient capture of detailed features and the need to improve recognition accuracy. Therefore, there is an urgent need for a method that can effectively identify subtle defects in the gearbox housing and provide real-time feedback on the results to improve production efficiency and product quality. Summary of the invention

[0003] In view of the above-mentioned technical deficiencies, the purpose of the present invention is to provide a method and system for identifying defects in a gearbox housing of an electric tricycle, so as to solve the problem of low recognition accuracy in the prior art.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a method for identifying defects in a gearbox housing of an electric tricycle, the method comprising: Step S100: acquiring first gearbox housing image data in real time through an industrial camera deployed in a housing defect recognition area of ​​a gearbox housing production line; Step S200: preprocessing the first gearbox housing image data to obtain second gearbox housing image data; Step S300: obtaining third gearbox housing image data from the second gearbox housing image data through an image super-resolution model; Step S400: inputting the third gearbox housing image data into a defect recognition model to determine whether the gearbox housing has defects; Step S500: When there is a defect in the gearbox housing, the defective gearbox housing is sorted out from the gearbox housing production line by a sorting device.

[0005] Preferably, in a possible implementation manner of the first aspect, the data preprocessing includes grayscale processing, noise removal and image enhancement.

[0006] Preferably, in a possible implementation manner of the first aspect, the image super-resolution model is trained by using an SRFormer algorithm, and the image super-resolution dataset uses a first gearbox housing image dataset.

[0007] Preferably, in a possible implementation manner of the first aspect, the SRFormer algorithm structure includes: The encoder module extracts low-level features of the input image and generates a query vector; The attention module is used to transform the query vector through the self-attention mechanism to capture the high-resolution features in the image; The decoder module combines the output of the attention module with low-level features to generate a high-resolution output image.

[0008] Preferably, in a possible implementation manner of the first aspect, the process of constructing the first gearbox housing image dataset includes: The first gearbox housing image in history is selected as a high-resolution image, and bicubic interpolation is used to downsample the first gearbox housing image in history to generate a corresponding low-resolution image; The high-resolution image and the corresponding low-resolution image constitute an image pair, and all image pairs constitute a first gearbox housing image dataset.

[0009] Preferably, in a possible implementation manner of the first aspect, the defect recognition model is trained using an improved yolo v8 algorithm, and the defect recognition data set uses a second gearbox housing image data set.

[0010] Preferably, in a possible implementation of the first aspect, the improved yolo v8 algorithm structure includes: The backbone network module uses Swin Transformer to extract multi-scale features of the input image; The feature fusion module is used to fuse the multi-scale features extracted by the backbone network module to generate a feature map with context information; The prediction head module is used to determine the location, category and confidence of the gearbox housing defect based on the fused feature map.

[0011] Preferably, in a possible implementation of the first aspect, whether the gearbox housing has defects is judged based on the location, category and confidence of the gearbox housing defect output by the defect recognition model; when an identification result with a confidence level greater than a preset threshold appears, it is judged that a corresponding gearbox housing defect appears at that location.

[0012] Preferably, in a possible implementation manner of the first aspect, step S500 specifically includes: When the defect recognition model determines that there is a defect in the gearbox housing, the sorting strategy is triggered; Based on the identification results and preset sorting strategies, the sorting equipment sorts the defective gearbox housings from the production line and transports them to the defective product processing area.

[0013] In a second aspect, the present invention provides a system for identifying defects in a gearbox housing of an electric tricycle, the system comprising: An image acquisition module is used to acquire the first gearbox housing image data in real time through an industrial camera deployed in a housing defect recognition area of ​​a gearbox housing production line; A data preprocessing module, used to perform grayscale processing, noise removal and image enhancement on the first gearbox housing image data to obtain second gearbox housing image data; A super-resolution processing module is used to process the second gearbox housing image data through an image super-resolution model to obtain third gearbox housing image data, wherein the image super-resolution model is trained using the SRFormer algorithm, and the image super-resolution data set uses the first gearbox housing image data set; A defect recognition module is used to input the third gearbox housing image data into a defect recognition model to determine whether the gearbox housing has defects. The defect recognition model is trained using an improved yolo v8 algorithm, and the defect recognition data set uses the second gearbox housing image data set; The sorting control module is used to trigger the sorting strategy when the defect recognition model determines that the gearbox housing has defects, and sort the defective gearbox housing from the gearbox housing production line through the sorting equipment and transport it to the defective product processing area.

[0014] The beneficial effects of the present invention are: by introducing an industrial camera to obtain a picture of the gearbox housing in real time, combined with data preprocessing, picture super-resolution processing and an improved defect recognition model, the accuracy and efficiency of gearbox housing defect recognition can be significantly improved. The picture super-resolution model trained with the SRFormer algorithm can effectively improve the picture resolution, enhance detail features, and provide high-quality input for subsequent defect recognition. At the same time, the improved yolo v8 algorithm uses Swin Transformer to extract multi-scale features and fuse contextual information to achieve accurate prediction of the position, category and confidence of the gearbox housing defect. In addition, combined with sorting equipment, defective gearbox housings can be sorted out from the production line in real time, greatly improving the automation level and product quality of the production line. The application of this method will provide strong support for the quality control of the gearbox housing of electric tricycles and promote the improvement of production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 A flow chart of a method for identifying defects in a gearbox housing of an electric tricycle is provided for this application.

[0017] Figure 2 A structural diagram of an electric tricycle gearbox housing defect identification system is provided for this application.

[0018] Explanation of the accompanying drawings: 1-image acquisition module, 2-data preprocessing module, 3-super-resolution processing module, 4-defect recognition module, 5-sorting control module. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] Embodiment 1: Figure 1 As shown, the present invention provides a method for identifying defects in a gearbox housing of an electric tricycle, comprising: Step S100: acquiring first gearbox housing image data in real time through an industrial camera deployed in a housing defect recognition area of ​​a gearbox housing production line.

[0021] Specifically, the industrial camera is fixedly installed above the shell defect recognition area of ​​the production line, covering the entire shell defect recognition area. When the gearbox housing passes through the recognition area, the industrial camera captures the gearbox housing image, obtains the first gearbox housing picture data, and transmits the first gearbox housing picture data to the control center.

[0022] In this embodiment, the industrial camera uses the acA1300-60gc model industrial camera. When the gearbox housing slowly moves to the identification area along the production line, the industrial camera starts the shooting function, and the captured image data is converted into a digital format, connected to the data communication interface through optical fiber, and transmitted to the control center. Step S200: preprocessing the first gearbox housing image data to obtain second gearbox housing image data.

[0023] Specifically, data preprocessing includes grayscale processing, noise removal and image enhancement, and the first gearbox housing image data is obtained after data preprocessing to obtain the second gearbox housing image data.

[0024] In this embodiment, first read the image of the gearbox housing, call the cv2.imread function, which receives an image path as a parameter and returns the image data. Next, grayscale the read image, use the cv2.cvtColor function to grayscale, which receives the original image and a color space conversion code as parameters, and returns the converted grayscale image. Then, the grayscale image is subjected to noise removal, which is implemented by calling the cv2.GaussianBlur function, which receives the grayscale image, the size and standard deviation of the Gaussian kernel as parameters, and returns the image after noise removal. After that, the image after noise removal is enhanced, and the image enhancement is performed by using the histogram equalization method, which is implemented by calling the cv2.equalizeHist function, which receives the image after noise removal as a parameter and returns the enhanced image. Finally, the preprocessed image is saved, which is implemented by calling the cv2.imwrite function, which receives a file path and the image to be saved as parameters.

[0025] Step S300: The second gearbox housing image data is obtained through an image super-resolution model to obtain third gearbox housing image data.

[0026] Specifically, the image super-resolution model is trained using the SRFormer algorithm, and the image super-resolution dataset uses the first gearbox housing image dataset.

[0027] The SRFormer algorithm structure includes: an encoder module, which is used to extract low-level features of the input image and generate a query vector; an attention module, which is used to transform the query vector through the self-attention mechanism to capture the high-resolution features in the image; and a decoder module, which is used to combine the output of the attention module with the low-level features to generate a high-resolution output image.

[0028] The construction process of the first gearbox housing image dataset includes: selecting the historical first gearbox housing image as the high-resolution image, downsampling the historical first gearbox housing image using bicubic interpolation, and generating the corresponding low-resolution image; the high-resolution image and the corresponding low-resolution image constitute an image pair, and all image pairs constitute the first gearbox housing image dataset.

[0029] In this embodiment, the image super-resolution model is trained using the SRFormer algorithm, and the data set uses the first gearbox housing image data set, which is divided into a training set and a validation set at a ratio of 8:2. The feature encoder includes multiple encoder layers based on permuted self-attention. These layers model the feature map through the self-attention mechanism and gradually extract deeper features. A 3x3 convolution is used to implement pixel embedding, mapping the input low-resolution image to the feature space to generate the initial feature map.

[0030] The feature encoder consists of multiple permuted self-attention groups (PAB groups), each of which contains multiple permuted self-attention blocks (PABs). The PAB consists of two parts: the permuted self-attention layer (PSA) and the convolutional feed-forward network (ConvFFN). In PSA, the query (Q), key (K), and value (V) matrices are first obtained through three linear layers. In order to reduce the amount of computation and maintain the expressiveness of the attention map generated by each attention head, the number of channels of K and V is compressed, and part of the spatial information is transferred to the channel dimension through a permutation operation. In this way, the number of channels is reduced, but there is no loss of spatial information. Subsequently, self-attention calculations are performed using Q and the permuted K and V to generate an attention map, and the output is obtained by weighted summing of the attention map. ConvFFN adds a local deep convolution branch between two linear layers to help encode more details and compensate for the loss of high-frequency information caused by self-attention.

[0031] High-resolution image reconstruction consists of a 3x3 convolution and a Pixel Shuffle operation. Pixel Shuffle implements the upsampling function, which changes the dimension of the feature map from Convert to , and obtain high-resolution images.

[0032] The second gearbox housing image data is input into the trained image super-resolution model, and the third gearbox housing image data after super-resolution reconstruction is output.

[0033] Step S400: The third gearbox housing image data is input into a defect recognition model to determine whether the gearbox housing has defects.

[0034] Specifically, the defect recognition model is trained using the improved YOLO V8 algorithm, and the defect recognition data set uses the second gearbox housing image data set. The position, category, and confidence level of the gearbox housing defect output by the defect recognition model are used to determine whether the gearbox housing has defects. When a recognition result with a confidence level greater than a preset threshold appears, it is determined that the corresponding gearbox housing defect occurs at that position.

[0035] The improved YOLO V8 algorithm structure includes: a backbone network module, which uses Swin Transformer to extract multi-scale features of the input image; a feature fusion module, which is used to fuse the multi-scale features extracted by the backbone network module to generate a feature map with context information; a prediction head module, which is used to determine the location, category and confidence of the gearbox housing defect based on the fused feature map.

[0036] In this embodiment, the defect recognition model is trained based on the improved yolo v8 algorithm, the data set uses the second gearbox housing picture data set, the second gearbox housing picture data set is annotated using LabelImg based on the historical second gearbox housing pictures, and the annotated pictures are divided into training set and verification set according to the ratio of 8:2, and the historical second gearbox housing pictures are obtained from the control center.

[0037] For the improved YOLO v8 algorithm, firstly, the backbone network module of the algorithm adopts the Swin Transformer structure, which can efficiently extract the multi-scale features of the input image. Compared with the traditional convolutional neural network, Swin Transformer has stronger feature extraction capabilities and better generalization performance, which is crucial for identifying complex and diverse defects on the gearbox housing. Secondly, the feature fusion module fuses the multi-scale features extracted by the backbone network to generate a feature map with rich contextual information, and uses the feature information at different scales to capture more detailed defect features. Finally, the prediction head module uses regression and classification methods to determine the location, category and confidence of the gearbox housing defect based on the fused feature map.

[0038] The third gearbox housing image data is input into the trained defect recognition model, and the position, category and confidence of the gearbox housing defect are output. When a detection frame with a confidence level greater than 0.6 appears, it is determined that the gearbox housing has a defect of this category, and the position of the gearbox housing defect is output.

[0039] Step S500: When there is a defect in the gearbox housing, the defective gearbox housing is sorted out from the gearbox housing production line by a sorting device.

[0040] Specifically, when the defect recognition model determines that the gearbox housing has defects, the sorting strategy is triggered; the sorting equipment sorts the defective gearbox housing from the production line and transports it to the defective product processing area based on the recognition results and the preset sorting strategy.

[0041] In this embodiment, when the defect recognition model determines that there is a defect in the gearbox housing, that is, when it recognizes that there is a defect with a confidence level higher than 0.6 at a certain position on the gearbox housing, a preset sorting strategy is triggered.

[0042] The sorting equipment used on the gearbox housing production line integrates high-precision sensors, robotic arms and conveying systems, which can accurately identify and grab the gearbox housing on the production line. When the defect recognition model sends a defect signal, the sorting equipment receives the sorting instruction and determines the best grabbing point and sorting path based on the defect location information provided in the recognition result. The sorting equipment is then started, and its robotic arm moves to the top of the defective gearbox housing under precise control. After confirming the grabbing point, the robotic arm clamps the gearbox housing and sorts it out from the production line along the preset conveying path. The sorted defective gearbox housing is transported to the defective product processing area.

[0043] Embodiment 2: Figure 2 As shown, the present invention provides a gearbox housing defect recognition system for an electric tricycle, comprising: An image acquisition module 1 is used to acquire first gearbox housing image data in real time through an industrial camera deployed in a housing defect recognition area of ​​a gearbox housing production line; A data preprocessing module 2 is used to perform grayscale processing, noise removal and image enhancement on the first gearbox housing image data to obtain second gearbox housing image data; A super-resolution processing module 3 is used to process the second gearbox housing image data through an image super-resolution model to obtain third gearbox housing image data, wherein the image super-resolution model is trained using the SRFormer algorithm, and the image super-resolution data set uses the first gearbox housing image data set; Defect recognition module 4, used for inputting the third gearbox housing image data into a defect recognition model to determine whether the gearbox housing has defects, the defect recognition model is trained by using an improved yolo v8 algorithm, and the defect recognition data set uses the second gearbox housing image data set; The sorting control module 5 is used to trigger the sorting strategy when the defect recognition model determines that the gearbox housing has defects, and sort the defective gearbox housing from the gearbox housing production line through the sorting equipment and transport it to the defective product processing area.

[0044] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for identifying defects in a gearbox housing of an electric tricycle, characterized in that: The method comprises: Step S100: acquiring first gearbox housing image data in real time through an industrial camera deployed in a housing defect recognition area of ​​a gearbox housing production line; Step S200: preprocessing the first gearbox housing image data to obtain second gearbox housing image data; Step S300: obtaining third gearbox housing image data from the second gearbox housing image data through an image super-resolution model; Step S400: inputting the third gearbox housing image data into a defect recognition model to determine whether the gearbox housing has defects; Step S500: When there is a defect in the gearbox housing, the defective gearbox housing is sorted out from the gearbox housing production line by a sorting device.

2. The method for identifying defects in a gearbox housing of an electric tricycle as claimed in claim 1, characterized in that: The data preprocessing includes grayscale processing, noise removal and image enhancement.

3. The electric tricycle gearbox housing defect identification method according to claim 1, characterized in that: The image super-resolution model is trained by using the SRFormer algorithm, and the image super-resolution dataset uses the first gearbox housing image dataset.

4. The method for identifying defects in a gearbox housing of an electric tricycle as claimed in claim 3, characterized in that: The SRFormer algorithm structure includes: The encoder module extracts low-level features of the input image and generates a query vector; The attention module is used to transform the query vector through the self-attention mechanism to capture the high-resolution features in the image; The decoder module combines the output of the attention module with low-level features to generate a high-resolution output image.

5. The method for identifying defects in the gearbox housing of an electric tricycle as claimed in claim 3, characterized in that: The process of constructing the first gearbox housing image dataset includes: The first gearbox housing image in history is selected as a high-resolution image, and bicubic interpolation is used to downsample the first gearbox housing image in history to generate a corresponding low-resolution image; The high-resolution image and the corresponding low-resolution image constitute an image pair, and all image pairs constitute a first gearbox housing image dataset.

6. The electric tricycle gearbox housing defect identification method according to claim 1, characterized in that: The defect recognition model is trained using an improved YOLO V8 algorithm, and the defect recognition data set uses a second gearbox housing image data set.

7. The method for identifying defects in a gearbox housing of an electric tricycle as claimed in claim 6, characterized in that: The improved YOLO V8 algorithm structure includes: The backbone network module uses Swin Transformer to extract multi-scale features of the input image; The feature fusion module is used to fuse the multi-scale features extracted by the backbone network module to generate a feature map with context information; The prediction head module is used to determine the location, category and confidence of the gearbox housing defect based on the fused feature map.

8. The method for identifying defects in a gearbox housing of an electric tricycle as claimed in claim 7, characterized in that: The position, category and confidence level of the gearbox housing defect output by the defect recognition model are used to determine whether the gearbox housing has defects. When a recognition result with a confidence level greater than a preset threshold appears, it is determined that a corresponding gearbox housing defect exists at that position.

9. The method for identifying defects in a gearbox housing of an electric tricycle as claimed in claim 1, characterized in that: The step S500 specifically includes: When the defect recognition model determines that there is a defect in the gearbox housing, the sorting strategy is triggered; Based on the identification results and preset sorting strategies, the sorting equipment sorts the defective gearbox housings from the production line and transports them to the defective product processing area.

10. An electric tricycle gearbox housing defect recognition system, characterized in that: The system comprises: An image acquisition module is used to acquire the first gearbox housing image data in real time through an industrial camera deployed in a housing defect recognition area of ​​a gearbox housing production line; A data preprocessing module, used to perform grayscale processing, noise removal and image enhancement on the first gearbox housing image data to obtain second gearbox housing image data; A super-resolution processing module is used to process the second gearbox housing image data through an image super-resolution model to obtain third gearbox housing image data, wherein the image super-resolution model is trained using the SRFormer algorithm, and the image super-resolution data set uses the first gearbox housing image data set; A defect recognition module is used to input the third gearbox housing image data into a defect recognition model to determine whether the gearbox housing has defects. The defect recognition model is trained using an improved yolo v8 algorithm, and the defect recognition data set uses the second gearbox housing image data set; The sorting control module is used to trigger the sorting strategy when the defect recognition model determines that the gearbox housing has defects, and sort the defective gearbox housing from the gearbox housing production line through the sorting equipment and transport it to the defective product processing area.

Images