Modeling method and modeling system for bicycle shock absorber

Abstract

The invention discloses a modeling method and system for a bicycle shock absorber, and belongs to the technical field of image modeling, and the method comprises the steps: obtaining shock absorber images of different types and different working states of the bicycle shock absorber; marking a key part of the shock absorber image, and processing the marked shock absorber image to obtain a preprocessed image data set; constructing an acting point activation analysis model based on a convolutional neural network, and inputting the preprocessed image data set into the acting point activation analysis model for calculation to obtain a shock absorber stress high activation area vector; and constructing a three-dimensional model of the bicycle shock absorber based on the shock absorber stress high activation area vector and the shock absorber parameters. According to the method, accurate position information can be provided through the stress point vectors obtained through the CNN, and the stress distribution conditions of the shock absorber under different conditions can be accurately known.

Description

Technical Field

[0001] The invention belongs to the technical field of image modeling, and in particular relates to a modeling method and a modeling system for a bicycle shock absorber. Background Art

[0002] Bicycle shock absorbers are mainly used to absorb the impact and vibration caused by uneven roads, improving riding comfort and control. Common shock absorbers include: Spring shock absorbers: use a combination of metal springs and dampers, the damper is used to control the rebound of the spring. Gas shock absorbers: use the compressible properties of gas and oil to absorb shock.

[0003] Currently, the modeling of bicycle shock absorbers mainly involves inputting the shock absorber morphological parameters and mechanical parameters into the software for modeling. It is impossible to accurately locate and identify the force points of the bicycle shock absorber. Since standard parameter modeling is used in the modeling process, the response of the shock absorber cannot be adjusted during the manufacturing process to adapt to different road conditions. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a modeling method and a modeling system for a bicycle shock absorber to solve the above problems in the prior art.

[0005] To achieve the above object, the present invention provides a modeling method for a bicycle shock absorber, comprising:

[0006] Obtain shock absorber images of different types and working states of bicycle shock absorbers;

[0007] Annotating key parts of the shock absorber image, and processing the annotated shock absorber image to obtain a preprocessed image data set;

[0008] Building a force point activation analysis model based on a convolutional neural network, inputting the preprocessed image data set into the force point activation analysis model for calculation, and obtaining a high force activation area vector of the shock absorber;

[0009] A three-dimensional model of a bicycle shock absorber is constructed based on the shock absorber high-force activation area vector and shock absorber parameters.

[0010] Preferably, the process of acquiring shock absorber images of different types and different working states of bicycle shock absorbers includes:

[0011] Using a depth camera to shoot different working states of different types of bicycle shock absorbers to obtain shooting images;

[0012] The different types of bicycle shock absorbers include but are not limited to spring shock absorbers and gas shock absorbers;

[0013] The different working states include but are not limited to a compressed state and an uncompressed state.

[0014] Preferably, the process of marking the key parts of the shock absorber image includes:

[0015] Acquire the coordinates of the shock absorber's application point in the shock absorber image to generate the shock absorber's application point coordinates;

[0016] Converting the shock absorber application point coordinates into an application point vector, and decomposing the application point vector into components along the shock absorber working direction and in a vertical direction;

[0017] The components along the working direction and the vertical direction of the shock absorber are annotated into the shock absorber image to complete the annotation of the key parts of the shock absorber image.

[0018] Preferably, the process of processing the annotated shock absorber image to obtain a preprocessed image data set includes:

[0019] Acquire the shock absorber morphological features, and generate a pixel threshold range based on the shock absorber morphological features;

[0020] Eliminate the pixel points that do not meet the pixel point threshold range in each pixel point of the annotated shock absorber image to generate a shock absorber feature image;

[0021] All shock absorber feature images are resized to generate the preprocessed image data set.

[0022] Preferably, the process of constructing a focus point activation analysis model based on a convolutional neural network includes:

[0023] Obtain a sample set of bicycle shock absorbers, and divide the sample set into a training set and a test set;

[0024] Build an attention mechanism module based on the Transformer module;

[0025] Constructing a convolutional neural network, importing the attention mechanism module into the convolutional neural network, and generating an improved convolutional neural network;

[0026] Training the improved convolutional neural network using the training set to obtain a training model;

[0027] Training the training model using the test set to generate training results;

[0028] Fine-tune the parameters of the training model based on the training results to generate the focus point activation analysis model;

[0029] The sample set of the bicycle shock absorber includes force point analysis information.

[0030] Preferably, the process of inputting the preprocessed image data set into the force point activation analysis model for calculation to obtain the shock absorber high force activation area vector comprises:

[0031] Extracting local force features of the preprocessed image data set through the attention mechanism module of the force point activation analysis model to obtain a force attention score;

[0032] Obtaining a focus point distribution vector based on the convolution module of the force attention score and the focus point activation analysis model;

[0033] The correlation of the force point distribution vector is calculated, and the force point distribution vector is divided into regions based on the correlation to obtain a high-force activation region vector of the shock absorber.

[0034] Preferably, the expression for obtaining the stressed attention score is:

[0035]

[0036] Where Score(f x ,f y ) represents the attention score, f x The component representing the working direction of the shock absorber, f y represents the vertical component of the shock absorber, T represents the matrix transpose, Represents the square root of the component dimension in the vertical direction of the shock absorber.

[0037] Preferably, the process of constructing a bicycle shock absorber three-dimensional model based on the shock absorber stress high activation area vector and shock absorber parameters includes:

[0038] constructing a shock absorber model based on the shock absorber parameters;

[0039] Importing the shock absorber stress high activation area vector as the shock absorber stress attribute into the shock absorber model to generate the shock absorber stress model;

[0040] The structural strength and function of the shock absorber stress model are verified by a finite element analysis method to obtain a three-dimensional model of the bicycle shock absorber.

[0041] To achieve the above object, the present invention also provides a modeling system for a bicycle shock absorber, comprising:

[0042] A data acquisition subsystem, used for acquiring shock absorber images of different types and different working states of the bicycle shock absorber;

[0043] A data processing subsystem, used for marking key parts of the shock absorber image, processing the marked shock absorber image, and obtaining a preprocessed image data set;

[0044] A computing subsystem, configured to construct a force point activation analysis model based on a convolutional neural network, input the preprocessed image data set into the force point activation analysis model for calculation, and obtain a vector of a high force activation area of ​​the shock absorber;

[0045] The modeling subsystem is used to construct a three-dimensional model of the bicycle shock absorber based on the shock absorber high-force activation area vector and shock absorber parameters.

[0046] Compared with the prior art, the present invention has the following advantages and technical effects:

[0047] The force point vector obtained by CNN in the present invention can provide accurate position information, which is helpful to accurately understand the force distribution of the shock absorber under different conditions, and can better analyze the stress and strain of the shock absorber structure in the shock absorber three-dimensional model, and calculate the stress distribution under various loading conditions, which effectively avoids over-design or material waste, and ensures the reliability and durability of the shock absorber. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0049] Figure 1 A flow chart of a modeling method for a bicycle shock absorber according to an embodiment of the present invention;

[0050] Figure 2 Schematic diagram of a modeling system for a bicycle shock absorber according to an embodiment of the present invention. DETAILED DESCRIPTION

[0051] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0052] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0053] Embodiment 1

[0054] like Figure 1 As shown, this embodiment provides a modeling method for a bicycle shock absorber, comprising:

[0055] Obtain shock absorber images of different types and working states of bicycle shock absorbers;

[0056] Annotating key parts of the shock absorber image, and processing the annotated shock absorber image to obtain a preprocessed image data set;

[0057] Building a force point activation analysis model based on a convolutional neural network, inputting the preprocessed image data set into the force point activation analysis model for calculation, and obtaining a high force activation area vector of the shock absorber;

[0058] A three-dimensional model of a bicycle shock absorber is constructed based on the shock absorber high-force activation area vector and shock absorber parameters.

[0059] The process of further optimizing the scheme to obtain shock absorber images of different types and working states of bicycle shock absorbers includes:

[0060] Based on the depth camera, different types of bicycle shock absorbers in different working states are photographed to obtain photographed images; this embodiment collects images of various bicycle shock absorbers. These images may include shock absorbers of different brands, different types (such as front forks, rear shock absorbers) and different working states (compressed and uncompressed states), which are acquired using a depth camera; the collected images are annotated to mark the key parts of the shock absorber, such as springs, dampers, locking mechanisms, etc. The annotation can be done manually or with the assistance of semi-automatic tools.

[0061] The different types of bicycle shock absorbers include but are not limited to spring shock absorbers and gas shock absorbers;

[0062] The different working states include but are not limited to a compressed state and an uncompressed state.

[0063] Further optimizing the solution, the process of labeling the key parts of the shock absorber image includes:

[0064] Acquire the coordinates of the shock absorber's application point in the shock absorber image to generate the shock absorber's application point coordinates;

[0065] Converting the shock absorber application point coordinates into an application point vector, and decomposing the application point vector into components along the shock absorber working direction and in a vertical direction;

[0066] The components along the working direction and the vertical direction of the shock absorber are annotated into the shock absorber image to complete the annotation of the key parts of the shock absorber image.

[0067] Further optimizing the scheme, the process of processing the labeled shock absorber image to obtain the preprocessed image data set includes:

[0068] Acquire the shock absorber morphological features, and generate a pixel threshold range based on the shock absorber morphological features;

[0069] Eliminate the pixel points that do not meet the pixel point threshold range in each pixel point of the annotated shock absorber image to generate a shock absorber feature image;

[0070] The sizes of all shock absorber feature images are unified to generate the preprocessed image data set, and the image data can be enhanced by methods such as rotation, scaling and color adjustment to improve the generalization ability of the model.

[0071] Performing Mosaic image data enhancement on all shock absorber feature images, randomly reading four original image data, randomly selecting an area as the center area of ​​stitching, stitching the four randomly cropped, randomly scaled, and randomly arranged images according to the center area, and obtaining the preprocessed image;

[0072] The image data after the Mosaic image data enhancement is subjected to Letterbox image data enhancement, the long side of the image data is scaled to a specified size, the short side is filled with black pixels, and the preprocessed image is obtained. The formula for Letterbox image data enhancement is as follows:

[0073]

[0074] Among them, d is the length that needs to be padded on one side, w is the width of the initial image, h is the height of the initial image, w′ is the specified width after the image is scaled, h′ is the specified height after the image is scaled, and the improved YOLOv8 forest fire detection model is downsampled 5 times in total, and the image features are scaled to 1 / 32 of the input image. Therefore, the size of the input image needs to be a multiple of 32. The total padding length is obtained through the mod32 operation, and the padding length of one side is obtained by / 2.

[0075] Further optimizing the scheme, the process of constructing the focus point activation analysis model based on the convolutional neural network includes:

[0076] Obtain a sample set of bicycle shock absorbers, and divide the sample set into a training set and a test set;

[0077] Build an attention mechanism module based on the Transformer module;

[0078] Construct a convolutional neural network, import the attention mechanism module into the convolutional neural network, and generate an improved convolutional neural network; in the early stage of the network, use a small convolution kernel (such as 3x3 or 5x5) and a small step size (such as a step size of 1). This can help capture detailed local features without losing spatial information prematurely, which is very important for accurately locating the focus point. As the depth of the network increases, the size of the convolution kernel can be gradually increased or the number of convolution layers can be increased to capture higher-level abstract features. These features can provide information about the overall structure and shape, which helps to identify specific parts of the shock absorber.

[0079] This embodiment uses maximum pooling to reduce the size of each feature map while retaining the most significant features, which is very helpful for highlighting important features (such as the focus area). However, the location and frequency of the pooling layer should be carefully selected to avoid excessive pooling that causes loss of important spatial information. Therefore, this embodiment uses an adaptive pooling layer at the end of the network, such as adaptive average pooling or adaptive maximum pooling, to ensure that the output feature map has a fixed size, which is particularly useful for subsequent fully connected layer processing and helps maintain performance consistency on input images of different sizes.

[0080] Training the improved convolutional neural network using the training set to obtain a training model;

[0081] Training the training model using the test set to generate training results;

[0082] Fine-tune the parameters of the training model based on the training results to generate the focus point activation analysis model;

[0083] The sample set of bicycle shock absorbers includes force point analysis information. In this embodiment, the collected sample set of bicycle shock absorbers is re-labeled with force points.

[0084] The process of marking the focus point includes:

[0085] Identify points of force: Identify which points are subject to the greatest forces, these are usually points of connection or fixing.

[0086] Decompose Force Vector: Decomposes the force vector into components perpendicular and parallel to the surface of interest.

[0087] Calculate stresses: Calculate the normal and shear stresses at each point using the formulas above.

[0088] Strain Calculation: Based on material properties such as Young's modulus and shear modulus, the strains caused by these stresses are calculated.

[0089] Further optimizing the scheme, the process of inputting the preprocessed image data set into the force point activation analysis model for calculation, and obtaining the shock absorber high force activation area vector includes:

[0090] Extracting local force features of the preprocessed image data set through the attention mechanism module of the force point activation analysis model to obtain a force attention score;

[0091] Obtaining a focus point distribution vector based on the convolution module of the force attention score and the focus point activation analysis model;

[0092] The correlation of the force point distribution vector is calculated, and the force point distribution vector is divided into regions based on the correlation to obtain a high-force activation region vector of the shock absorber.

[0093] Further optimizing the scheme, the expression for obtaining the attention score is:

[0094]

[0095] Where Score(f x ,f y ) represents the attention score, f x The component representing the working direction of the shock absorber, f y represents the vertical component of the shock absorber, T represents the matrix transpose, Represents the square root of the component dimension in the vertical direction of the shock absorber.

[0096] The Transformer structure calculates the force attention score. The global features of the bicycle shock absorber are obtained based on the Transformer module. Its input is a feature map whose size is reduced by the average pooling layer to reduce the computational cost. The specific process is that each pixel in the feature map is mapped to a high-dimensional embedding vector, each pixel is regarded as a token, and its features are represented by the embedding vector. Then the token embedding operation is performed. Since the Transformer algorithm itself does not contain the position information of the sequence, it is necessary to add positional encodings to represent the position information of each pixel. The positional encoding is a fixed vector whose value varies according to the position of the pixel. By combining positional encoding with token embedding, it helps the model understand the different position information in the sequence.

[0097] The Transformer structure calculates the dot product of the query vector and the key vector to obtain the attention score. This embodiment replaces the query vector and the key vector with the components in the working direction of the shock absorber and the components in the vertical direction of the shock absorber to better scale the attention score, and applies the softmax function to obtain the attention weight. The self-attention mechanism is applied to the global information of the sequence to capture long-distance dependencies. Then, a convolutional neural network can be used to process local patterns and feature extraction. These two structures can be connected in series or in parallel.

[0098] It is also possible to consider adding residual connections to the network, which helps train deeper network structures without causing gradient vanishing or exploding. Residual connections also help maintain the integrity of input information in deeper layers, which is very important for maintaining accurate prediction of the focus point location.

[0099] The formula for calculating the grid position of the convolution kernel along the x-axis is as follows:

[0100]

[0101] The formula for calculating the grid position of the convolution kernel along the y-axis is as follows:

[0102]

[0103] Since the Δ offset is in decimal form and the coordinates are in integer form, the bilinear interpolation method is used:

[0104] CK=∑ CK′ B(CK′,CK)·CK′;

[0105] Where CK represents CK in the above formula i±c With CK j±c The decimal part, CK′ represents the enumeration of all integer spaces, and B represents the two-dimensional bilinear interpolation kernel, which is decomposed into two one-dimensional kernels. The formula is as follows:

[0106] B(CK,CK′)=b(CK x ,CK x ′)·b(CK y ,CK y ′);

[0107] The feature map size obtained by bilinear interpolation sampling along the x-axis is [B, C, K*H, W]. After the convolution kernel operation with a size of K×1 and the ReLU activation function, the output feature map size in the x-axis direction is [B, C, H, W].

[0108] Finally, the vector of the central force point of the feature map is extracted, the correlation of the force point distribution vector is calculated, and the force point distribution vector is divided into regions based on the correlation to obtain the shock absorber high force activation region vector;

[0109] Further optimizing the scheme, the process of constructing a bicycle shock absorber three-dimensional model based on the shock absorber high force activation area vector and shock absorber parameters includes:

[0110] constructing a shock absorber model based on the shock absorber parameters;

[0111] The shock absorber stress high activation area vector is imported into the shock absorber model as the shock absorber stress attribute to generate the shock absorber stress model. The structural strength and function of the shock absorber stress model are verified by the finite element analysis method to obtain the bicycle shock absorber three-dimensional model.

[0112] Convolution module: Through a set of convolution kernels, the input data is convolved to extract spatial features. The size, step size, padding and other parameters of the convolution kernel can affect the scale of feature extraction.

[0113] Focus point distribution vector: This vector can represent the "attention" or "importance" of different regions in the image. It can be obtained by the following steps:

[0114] Feature map generation: The feature map output by the convolutional layer contains the local features of the image at different locations.

[0115] Weighted aggregation: By weighting or aggregating the features of different regions, a vector is generated that represents the "focus point distribution" of the entire image.

[0116] Fusing attention scores: If the attention mechanism is used, the obtained attention scores can be combined with the convolutional feature maps to enhance the model's focus on key areas.

[0117] At this point, the shock absorber 3D model is no longer a simple 3D model, but can simulate the stress and strain of the shock absorber at various points through the high-stress activation area vector of the shock absorber; use finite element analysis in CAD or special simulation software to verify the structural strength and function of the shock absorber. And it can simulate actual working conditions such as collision, long-term fatigue, etc. to test the performance of the shock absorber.

[0118] Embodiment 2

[0119] like Figure 2 As shown, in this embodiment, a modeling system for a bicycle shock absorber is provided, comprising:

[0120] A data acquisition subsystem, used for acquiring shock absorber images of different types and different working states of the bicycle shock absorber;

[0121] A data processing subsystem, used for marking key parts of the shock absorber image, processing the marked shock absorber image, and obtaining a preprocessed image data set;

[0122] A computing subsystem, configured to construct a force point activation analysis model based on a convolutional neural network, input the preprocessed image data set into the force point activation analysis model for calculation, and obtain a vector of a high force activation area of ​​the shock absorber;

[0123] The modeling subsystem is used to construct a three-dimensional model of the bicycle shock absorber based on the shock absorber high-force activation area vector and shock absorber parameters.

[0124] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A modeling method for a bicycle shock absorber, characterized in that: The following steps are involved: Obtain shock absorber images of different types and working states of bicycle shock absorbers; Annotating key parts of the shock absorber image, and processing the annotated shock absorber image to obtain a preprocessed image data set; Building a force point activation analysis model based on a convolutional neural network, inputting the preprocessed image data set into the force point activation analysis model for calculation, and obtaining a high force activation area vector of the shock absorber; A three-dimensional model of a bicycle shock absorber is constructed based on the shock absorber high-force activation area vector and shock absorber parameters.

2. The modeling method of a bicycle shock absorber according to claim 1, characterized in that: The process of acquiring shock absorber images of different types and working conditions of bicycle shock absorbers includes: Using a depth camera to shoot different working states of different types of bicycle shock absorbers to obtain shooting images; The different types of bicycle shock absorbers include but are not limited to spring shock absorbers and gas shock absorbers; The different working states include but are not limited to a compressed state and an uncompressed state.

3. The modeling method of a bicycle shock absorber according to claim 1, characterized in that: The process of labeling the key parts of the shock absorber image includes: Acquire the coordinates of the shock absorber's application point in the shock absorber image to generate the shock absorber's application point coordinates; Converting the shock absorber application point coordinates into an application point vector, and decomposing the application point vector into components along the shock absorber working direction and in a vertical direction; The components along the working direction and the vertical direction of the shock absorber are annotated into the shock absorber image to complete the annotation of the key parts of the shock absorber image.

4. The modeling method of a bicycle shock absorber according to claim 1, characterized in that: The process of processing the labeled shock absorber image to obtain a preprocessed image data set includes: Acquire the shock absorber morphological features, and generate a pixel threshold range based on the shock absorber morphological features; Eliminate the pixel points that do not meet the pixel point threshold range in each pixel point of the annotated shock absorber image to generate a shock absorber feature image; All shock absorber feature images are resized to generate the preprocessed image data set.

5. The modeling method of a bicycle shock absorber according to claim 1, characterized in that: The process of constructing a focus point activation analysis model based on a convolutional neural network includes: Obtain a sample set of bicycle shock absorbers, and divide the sample set into a training set and a test set; Build an attention mechanism module based on the Transformer module; Constructing a convolutional neural network, importing the attention mechanism module into the convolutional neural network, and generating an improved convolutional neural network; Training the improved convolutional neural network using the training set to obtain a training model; Training the training model using the test set to generate training results; Fine-tune the parameters of the training model based on the training results to generate the focus point activation analysis model; The sample set of the bicycle shock absorber includes force point analysis information.

6. The modeling method of a bicycle shock absorber according to claim 5, characterized in that: The process of inputting the preprocessed image data set into the force point activation analysis model for calculation to obtain the shock absorber high force activation area vector includes: Extracting local force features of the preprocessed image data set through the attention mechanism module of the force point activation analysis model to obtain a force attention score; Obtaining a focus point distribution vector based on the force attention score and the convolution module of the focus point activation analysis model; The correlation of the force point distribution vector is calculated, and the force point distribution vector is divided into regions based on the correlation to obtain a high-force activation region vector of the shock absorber.

7. The modeling method of a bicycle shock absorber according to claim 6, characterized in that: The expression for obtaining the attention score is: Among them, Score(f x ,f y ) represents the attention score, f x The component representing the working direction of the shock absorber, f y represents the vertical component of the shock absorber, T represents the matrix transpose, It represents the square root of the component dimension in the vertical direction of the shock absorber, x represents the working direction of the shock absorber, and y represents the shock absorber.

8. The modeling method of a bicycle shock absorber according to claim 1, characterized in that: The process of constructing a bicycle shock absorber three-dimensional model based on the shock absorber stress high activation area vector and shock absorber parameters includes: constructing a shock absorber model based on the shock absorber parameters; Importing the shock absorber stress high activation area vector as the shock absorber stress attribute into the shock absorber model to generate the shock absorber stress model; The structural strength and function of the shock absorber stress model are verified by a finite element analysis method to obtain a three-dimensional model of the bicycle shock absorber.

9. A modeling system for a bicycle shock absorber, characterized in that: include: A data acquisition subsystem, used for acquiring shock absorber images of different types and different working states of the bicycle shock absorber; A data processing subsystem, used for marking key parts of the shock absorber image, processing the marked shock absorber image, and obtaining a preprocessed image data set; A computing subsystem, configured to construct a force point activation analysis model based on a convolutional neural network, input the preprocessed image data set into the force point activation analysis model for calculation, and obtain a vector of a high force activation area of ​​the shock absorber; The modeling subsystem is used to construct a three-dimensional model of the bicycle shock absorber based on the shock absorber high-force activation area vector and shock absorber parameters.

Images