A method for determining parameters of plastic spiral bevel gears with variable shrinkage

Abstract

The present invention relates to a parameter determination method for a plastic spiral bevel gear with a variable shrinkage rate, aiming to solve the influence of the difference in shrinkage rate on the meshing accuracy of the plastic spiral bevel gear. The method comprises the following steps: first, the expected structural data and application environment data of the plastic spiral bevel gear are obtained, the latter including the expected performance data under different working conditions. Subsequently, a preliminary geometric model and preset shrinkage rate data are constructed using a deep learning convolutional neural network (CNN). Then, these data are input into a deep learning generative adversarial network (GAN), and the preliminary model is iteratively optimized for multiple times to take into account the shrinkage rates at different positions, and finally the parameter data of the optimized target geometric model is obtained and outputted. The meshing accuracy of the plastic spiral bevel gear is effectively improved, and the technical challenges brought by the difference in tooth surface shrinkage rates are overcome.

Description

Technical Field

[0001] The invention relates to the technical field of data processing, and in particular to a method for determining parameters of a plastic spiral bevel gear with a variable shrinkage rate. Background Art

[0002] In the field of smart home, although spiral bevel gears are rarely used, they are still widely used in some devices, such as water flosser gearboxes, due to their compact space structure and the need for high transmission ratio and efficiency. Spiral gears can meet these requirements, but as the requirements of smart home users for experience continue to increase, the choice of materials has become crucial. In order to reduce noise, control costs, and achieve lightweight, plastic gears have gradually replaced traditional metal gears. However, the biggest difference between plastic gears and metal gears is the shrinkage characteristics of the material, which brings new challenges to the design and manufacture of gears.

[0003] Among plastic gears, spur gears have relatively simple shrinkage control due to the consistency of their dimensions in the tooth width direction. However, the dimensions of spiral bevel gears in the tooth width direction change gradually, so their shrinkage rate also changes continuously. During the injection molding process, tooth surfaces of different thicknesses will lead to slight differences in material shrinkage, which in turn affects the degree of fit between the actual size of the gear and the designed size. Taking the commonly used POM material as an example, the spiral bevel gear has a significant difference in width between the inside and outside. If the gear is scaled with the same shrinkage rate, the accuracy of the tooth shape will decrease, which will affect the smoothness of the gear meshing and the transmission efficiency.

[0004] At present, most spiral bevel gear injection molding designs are processed with a uniform shrinkage rate. This method is not a significant problem in small gears, but in gears with large modules and large numbers of teeth, the error value will increase significantly as the size increases. Due to the difference in shrinkage rate in the tooth width direction, the tooth surface size may not match the design, which will eventually affect the gear meshing accuracy and wear resistance, causing increased transmission noise and reduced gear life. There is a technical problem in the prior art that the difference in shrinkage rate of the tooth surface of plastic spiral bevel gears affects the meshing accuracy, and no solution has been found so far. Summary of the invention

[0005] The main purpose of the present invention is to provide a method for determining parameters of a plastic spiral bevel gear with a variable shrinkage rate, so as to solve the technical problem in the prior art that the difference in the shrinkage rate of the tooth surface of the plastic spiral bevel gear affects the meshing accuracy.

[0006] To achieve the above-mentioned objectives, the present invention provides a method for determining parameters of a plastic spiral bevel gear with a variable shrinkage rate, comprising the following steps: obtaining expected structural data of the plastic spiral bevel gear and application environment data of the plastic spiral bevel gear in a water flosser gear box, wherein the application environment data at least includes: expected performance data of the plastic spiral bevel gear under multiple working conditions; based on the expected structural data and the application environment data, using a convolutional neural network based on deep learning to construct a preliminary geometric model and preset shrinkage rate data of the plastic spiral bevel gear; inputting the preliminary geometric model and preset shrinkage rate data of the plastic spiral bevel gear into a generative adversarial network based on deep learning, and using the generative adversarial network to iteratively optimize the preliminary geometric model multiple times based on the preset shrinkage rate data; obtaining a target geometric model optimized according to the shrinkage rate at different positions in the tooth width direction, and outputting the parameter data of the target geometric model.

[0007] The present invention also provides a parameter determination device for a plastic spiral bevel gear with a variable shrinkage rate, comprising: an acquisition unit, used to acquire expected structural data of the plastic spiral bevel gear, and application environment data of the plastic spiral bevel gear in a water flosser gear box, wherein the application environment data at least includes: expected performance data of the plastic spiral bevel gear under multiple working conditions; a construction unit, used to construct a preliminary geometric model and preset shrinkage rate data of the plastic spiral bevel gear based on the expected structural data and the application environment data, using a convolutional neural network based on deep learning; an iteration unit, used to input the preliminary geometric model and preset shrinkage rate data of the plastic spiral bevel gear into a generative adversarial network based on deep learning, and use the generative adversarial network to perform multiple iterative optimizations on the preliminary geometric model based on the preset shrinkage rate data; an output unit, used to obtain a target geometric model optimized according to the shrinkage rate at different positions in the tooth width direction, and output the parameter data of the target geometric model.

[0008] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0009] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0010] The parameter determination method and device of the variable shrinkage plastic spiral bevel gear provided by the present invention are aimed at solving the influence of the shrinkage rate difference of the plastic spiral bevel gear tooth surface on the meshing accuracy in the prior art. The method includes several key steps: first, the expected structural data of the plastic spiral bevel gear and the application environment data in the tooth flosser gear box are obtained, and the latter includes the expected performance data under different working conditions. Next, the preliminary geometric model and shrinkage rate preset data of the plastic spiral bevel gear are constructed using a convolutional neural network (CNN) based on deep learning. Then, these preliminary models and data are input into the generative adversarial network (GAN) based on deep learning, and the preliminary geometric model is iteratively optimized multiple times to take into account the shrinkage rate at different positions, so as to obtain the parameter data of the target geometric model optimized in the tooth width direction, and output processing is performed. The meshing accuracy of the plastic spiral bevel gear is effectively improved, and the technical problems caused by the difference in tooth surface shrinkage rate are solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 Schematic diagram of a method for determining parameters of a plastic spiral bevel gear with a variable shrinkage ratio in one embodiment of the present invention;

[0012] Figure 2 is a schematic diagram of a gear structure of a conventional oral flushing device in one embodiment of the present invention;

[0013] Figure 3 is a schematic diagram of a spiral bevel gear structure of a water flosser according to an embodiment of the present invention;

[0014] Figure 4 1.02 is a schematic diagram of a shrinkage rate model of a spiral bevel gear structure in one embodiment of the present invention;

[0015] Figure 5 Schematic diagram of a multi-shrinkage model of a spiral bevel gear structure in one embodiment of the present invention;

[0016] Figure 6 It is a framework diagram of a device for determining parameters of a plastic spiral bevel gear with a variable shrinkage ratio in one embodiment of the present invention;

[0017] Figure 7 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0018] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0020] Reference Figure 1 The present invention provides a method for determining parameters of a variable shrinkage plastic spiral bevel gear, comprising the following steps:

[0021] S1, obtaining expected structural data of the plastic spiral bevel gear and application environment data of the plastic spiral bevel gear in a water flosser gear box, wherein the application environment data at least includes: expected performance data of the plastic spiral bevel gear under multiple working conditions.

[0022] S2, based on the expected structure data and the application environment data, a preliminary geometric model and preset shrinkage rate data of the plastic spiral bevel gear are constructed using a convolutional neural network based on deep learning.

[0023] S3, inputting the preliminary geometric model and the preset shrinkage rate data of the plastic spiral bevel gear into a generative adversarial network based on deep learning, and using the generative adversarial network to perform multiple iterative optimizations on the preliminary geometric model based on the preset shrinkage rate data.

[0024] S4, obtaining a target geometric model optimized according to the shrinkage rates at different positions in the tooth width direction, and outputting parameter data of the target geometric model.

[0025] That is, in step S1, obtaining the expected structural data of the plastic spiral bevel gear is the basis for building the entire model. The expected structural data includes the design geometry of the gear, material properties, and its performance under various working conditions. These data are closely related to the application environment data of the plastic spiral bevel gear in the tooth flushing device gearbox, where the application environment data covers the performance of the gear in actual operation, such as working performance under different loads, speeds, temperatures, etc. These expected performance data provide necessary reference standards for subsequent geometric model optimization.

[0026] Next, based on the expected structural data and application environment data obtained, a convolutional neural network (CNN) is used in step S2 to construct a preliminary geometric model. When dealing with complex structural features and environmental variables, the convolutional neural network can effectively extract the local and global geometric features of the plastic spiral bevel gear. By extracting and analyzing these features, a preliminary geometric model is established, and shrinkage preset data is generated. These shrinkage data are predicted based on the possible dimensional changes of the gear under different working conditions. The multi-level structure of the convolutional neural network can ensure that geometric features of different scales and complexities are effectively captured, thus providing a solid foundation for the next iterative optimization.

[0027] In step S3, the preliminary geometric model and the preset shrinkage rate data are input into the deep learning-based generative adversarial network (GAN) and optimized through multiple iterations. In the working mechanism of the generative adversarial network, the generator and the judge use adversarial learning to enable the generator to gradually optimize the preliminary geometric model, especially to accurately adjust the shrinkage rate of the gear at different positions. The judge compares the model before and after optimization to evaluate its smoothness and consistency in the tooth width direction, thereby providing feedback to the generator. After multiple iterations, the generator will gradually adjust the local features of the model to make it more in line with the actual requirements in the expected performance data.

[0028] Finally, in step S4, the parameter data of the target geometric model obtained through multiple iterations of optimization is precisely adjusted in the tooth width direction according to the shrinkage rate at different positions. This model can not only reflect the actual performance of the gear under different working conditions, but also predict and compensate for the geometric changes caused by material shrinkage. The final target geometric model is output and processed and can be directly used in the manufacturing process of the gear to ensure the reliability and accuracy of the plastic spiral bevel gear in actual use.

[0029] In one example, after obtaining the target geometric model optimized according to the shrinkage rates at different positions in the tooth width direction, the parameter determination method further includes: simulating and verifying the target geometric model, and predicting the deformation data of the target geometric model under different loads, wherein the deformation data at least includes: dimensional change data in the tooth width direction and dimensional change data in the tooth height direction; based on the deformation data, judging whether the target geometric model passes the expected design requirements; if so, continuing to execute the step of "outputting the target geometric model"; if not, feeding back the deformation data to the generative adversarial network based on deep learning, adjusting the parameters of the generator in the generative adversarial network according to the deformation data, and continuing to execute the step of "using the generative adversarial network to iteratively optimize the preliminary geometric model for multiple times".

[0030] That is, in the further implementation of the gear model determination method, evaluating the performance of the target geometric model under actual working conditions through simulation verification is a key step to ensure that it meets the design requirements. After obtaining the target geometric model that is optimized according to the shrinkage rate at different positions in the tooth width direction, it is first necessary to perform simulation verification. The purpose of simulation verification is to predict the deformation behavior of the gear under different loads, especially the dimensional changes in the tooth width direction and tooth height direction. These deformation data will reflect the real response of the gear when it is subjected to external loads during operation. Through simulation, the performance of the gear under various stress conditions can be effectively predicted, so that potential problems can be identified in advance.

[0031] Specifically, the dimensional change data in the tooth width direction and the dimensional change data in the tooth height direction in the deformation data are key indicators for evaluating whether the gear model meets the expected design requirements. The dimensional change in the tooth width direction is related to the contact surface and load distribution of the gear, while the dimensional change in the tooth height direction affects the meshing efficiency and transmission accuracy of the gear. These data are analyzed in detail during the simulation process to determine the deformation characteristics of the target geometric model under different working conditions. If the simulation results show that the deformation data of the target geometric model meets the expected design requirements, it means that the model can show stability and reliability in practical applications, and can then directly enter the output processing stage for subsequent manufacturing and application.

[0032] However, if the simulation results show that the target geometry model fails to meet the expected design requirements, the model needs to be further adjusted and optimized. In this case, the deformation data will be input into the generative adversarial network as feedback information to guide the adjustment of the generator parameters. Through this feedback mechanism, the generator can learn and adapt to the deformation characteristics of the gear under different load conditions, and then further iterate and optimize the preliminary geometry model. During the adjustment process, the generator will combine the specific performance in the deformation data to optimize the local structure of the gear to better meet the expected design standards.

[0033] After multiple iterations of optimization, the target geometry model will be verified by simulation again to ensure that it ultimately meets the design requirements. This feedback-based optimization cycle not only improves the accuracy of the model, but also enhances its adaptability and robustness under different working conditions. Ultimately, through this series of simulation verification and iterative optimization, the gear model can accurately reflect the performance requirements in actual applications and enter the output processing stage in the best state, ensuring that the expected design effect is achieved during actual manufacturing and application.

[0034] In one example, after obtaining the expected structural data of the plastic spiral bevel gear, the parameter determination method also includes: converting the expected structural data into a point cloud data format to obtain first structural data, wherein each data point in the first structural data includes an XYZ coordinate value; filtering the abnormal values ​​in the first structural data, and deleting the data points far away from the main data set to obtain second structural data; using the DBSCAN algorithm to cluster the data points in the second structural data according to the density values ​​of the data points in the second structural data to obtain multiple data clusters and multiple data points that do not belong to any data cluster; deleting the data points that do not belong to any data cluster as noise, and using three-dimensional spline interpolation technology to fill the missing points in the deleted second structural data to obtain third structural data; and using the third structural data as the expected structural data to execute the step of "based on the expected structural data and the application environment data, using a convolutional neural network based on deep learning to construct a preliminary geometric model and shrinkage rate preset data of the plastic spiral bevel gear".

[0035] That is, in this example of the gear model determination method, after obtaining the expected structural data of the plastic spiral bevel gear, these data are first converted into a point cloud data format. The point cloud data format is a three-dimensional data representation method, in which each data point has an XYZ coordinate value, representing a specific position in space. Through this conversion, the original expected structural data can be represented as a three-dimensional point cloud in a more intuitive and accurate way, and these point cloud data constitute the so-called first structural data.

[0036] Next, the first structure data is filtered for outliers. This process involves identifying and removing outlier data points that may exist in the data set, especially those that are far away from the main data set. Outlier data points may be caused by measurement errors or noise in the data acquisition process, and their presence may affect subsequent analysis and modeling. Therefore, after removing these outlier data points, more accurate second structure data is obtained, which better reflects the true expected structure of the plastic spiral bevel gear.

[0037] After obtaining accurate second structure data, DBSCAN algorithm is used for clustering. By analyzing the density distribution of data points, the areas with higher density are divided into different data clusters. These data clusters represent different structural features in the gear model. At the same time, those data points that do not belong to any data cluster are identified as noise. Usually, these noise data points may be caused by measurement errors or external interference, so they will be deleted in this step.

[0038] After deleting the noise data points, since the deletion operation may cause gaps or missing data, the 3D spline interpolation technique is used to fill these missing points. Reasonable intermediate data points are generated by interpolating the surrounding known data points. This process produces the third structure data, making the point cloud data more complete and consistent. Finally, based on the obtained third structure data, that is, the point cloud data after outlier filtering, noise removal, and interpolation, the next step is performed: using this data as the input of the convolutional neural network, combined with the application environment data, to build a preliminary geometric model of the plastic spiral bevel gear. The convolutional neural network is good at extracting features from complex 3D data, so in this process, it can accurately generate a preliminary geometric model of the gear based on the processed point cloud data, laying the foundation for subsequent optimization and simulation steps.

[0039] In one example, after obtaining the application environment data of the plastic spiral bevel gear in the gear box of the water flosser, the gear model determination method also includes: normalizing the amplitude of the application environment data to reduce the variation range of the application environment data to within a preset range to obtain first environment data; using time series generation technology to perform expected performance expansion processing on the first environment data to obtain second environment data; based on the second environment data, rigidly transforming the third structure data to align the third structure data with the second environment data to obtain fourth structure data; fusing the fourth structure data and the second environment data to obtain a comprehensive data set, wherein the comprehensive data set is used to execute the step of "using a convolutional neural network based on deep learning to construct a preliminary geometric model of the plastic spiral bevel gear".

[0040] That is, the application environment data obtained refers to the performance data of the plastic spiral bevel gear in the actual working environment. These data may include information such as load, temperature, vibration, etc. of the gear under different operating conditions. However, the amplitude and range of these data may vary and may not meet the standards required for subsequent analysis. Therefore, the first step is to normalize the amplitude of the application environment data. This process reduces the numerical range of the application environment data to a preset standard range, thereby obtaining the first environment data. This normalization process ensures the consistency of the data, making subsequent processing more stable and reliable.

[0041] Next, the first environmental data is augmented using time series generation techniques. This technique usually involves predicting and generating future trends and possible changes based on existing data. In this step, the second environmental data is obtained by analyzing historical data and generating forecasts of future performance. This augmentation process can help us foresee changes in environmental conditions over different time periods and provide more comprehensive reference data for the modeling process.

[0042] After obtaining the second environment data, the next step is to perform a rigid transformation on the third structure data. The third structure data is obtained from the point cloud data processing, including the data after outlier filtering and interpolation processing. The goal of this step is to align the third structure data with the second environment data in terms of data. This means that the position, size or direction of the third structure data needs to be adjusted so that it can be consistent with the range and distribution of the second environment data, thereby obtaining the fourth structure data. This rigid transformation process ensures the spatial consistency of the data, allowing the subsequent fusion steps to proceed smoothly.

[0043] Finally, the fourth structural data and the second environmental data will be fused. The process of data fusion involves integrating the two sets of data into a comprehensive data set, which will contain comprehensive information of structural data and environmental data. This fusion can provide more comprehensive and accurate model input information for subsequent modeling and analysis. In the present method, this comprehensive data set will be used to perform the step of "constructing a preliminary geometric model of the plastic spiral bevel gear based on a convolutional neural network based on deep learning". By combining environmental data and structural data, the convolutional neural network can generate a more accurate geometric model, thereby improving the accuracy and practicality of the model.

[0044] This process ensures that the performance of the gear model in the actual working environment can be more accurately reflected, improving the reliability and accuracy of the model.

[0045] In one example, based on the expected structure data and the application environment data, a convolutional neural network based on deep learning is used to construct a preliminary geometric model and shrinkage rate preset data of the plastic spiral bevel gear, including: extracting and processing the local geometric features and global morphological features of the plastic spiral bevel gear in the comprehensive data set through the convolutional layer of the convolutional neural network based on deep learning to obtain a geometric feature set; performing dimensionality reduction processing on the geometric feature set through the pooling layer of the convolutional neural network, and after performing the dimensionality reduction processing, introducing the application environment data in the comprehensive data set to obtain a performance feature set; fusing the performance feature set with the geometric feature set through the fully connected layer of the convolutional neural network, and optimizing the parameters of the convolutional neural network through the back propagation algorithm; until a preliminary geometric model and shrinkage rate preset data reflecting the geometric features of the plastic spiral bevel gear are output, wherein the shrinkage rate preset data represents the predicted shrinkage rate value of each position in the preliminary geometric model.

[0046] That is, the convolutional layer of the deep learning-based convolutional neural network is used to extract local geometric features and global morphological features of the plastic spiral bevel gear in the comprehensive dataset. The convolutional layer is the core part of the convolutional neural network, which is responsible for extracting features from the input data. It scans the data through convolution operations to capture local structures and patterns in the data. For plastic spiral bevel gears, this means that the convolutional layer will process various details of the gear, including local features such as the gear's tooth shape and tooth spacing, as well as global features such as the overall shape and size of the gear. The extraction results of these features constitute a set of geometric features that reflect the geometric structure of the spiral bevel gear.

[0047] Next, the pooling layer of the convolutional neural network performs dimensionality reduction on the geometric feature set. The role of the pooling layer is to reduce the feature dimension and retain important information while reducing the computational burden and preventing overfitting. After dimensionality reduction, the application environment data in the comprehensive dataset is introduced. These environmental data contain the operating conditions of the spiral bevel gear in actual applications, such as load, temperature, etc. Introducing this data can help the convolutional neural network understand the performance characteristics in a specific environment, thereby generating a performance feature set. These performance feature sets combine geometric features with environmental conditions to provide a more comprehensive understanding of the actual performance of the gear.

[0048] Subsequently, the performance feature set is fused with the geometric feature set through the fully connected layer of the convolutional neural network. The fully connected layer is responsible for integrating the features to form the final model input. The purpose of the fusion process is to combine local and global geometric features with the application environment data to generate a more accurate preliminary geometric model and shrinkage preset data. The shrinkage preset data represents the predicted shrinkage at each location in the model for further model optimization.

[0049] Finally, the parameters of the convolutional neural network are optimized by the back-propagation algorithm. The back-propagation algorithm calculates the error between the model prediction and the actual result and transmits the error back to the network, thereby adjusting the weights and biases of the network. This process continues until the model can accurately reflect the geometric characteristics and shrinkage prediction values ​​of the plastic spiral bevel gear. The optimized model can output a preliminary geometric model that reflects the actual geometric characteristics of the spiral bevel gear and the corresponding preset shrinkage data, providing a basis for subsequent model adjustment and optimization.

[0050] This process ensures that the generated geometric model not only meets the design requirements, but also accurately reflects the performance in actual applications, thereby improving the practicality and accuracy of the model.

[0051] In one example, the generative adversarial network is used to perform multiple iterative optimizations on the preliminary geometric model based on the preset shrinkage rate data, including: starting the local shrinkage rate adjustment mechanism of the generative adversarial network, so that the generator of the generative adversarial network locally optimizes the preliminary geometric model according to the predicted shrinkage rate values ​​at each position of the preliminary geometric model; and inputting the locally optimized preliminary geometric model into the judgement of the generative adversarial network to determine the local optimization data of the preliminary geometric model after local optimization and the overall geometric data of the preliminary geometric model before optimization; in combination with a preset global loss function, the smoothness and consistency of the locally optimized preliminary geometric model are evaluated for the local optimization data and the overall geometric data to obtain initial evaluation data; based on the initial evaluation data, a mechanism adjustment parameter for adjusting the local shrinkage rate adjustment mechanism is generated, and the generator is optimized by the judgement based on the mechanism adjustment parameter; using the optimized generator, the preliminary geometric model is locally optimized iteratively based on the predicted shrinkage rate values ​​at each position, until a final geometric model whose initial evaluation data meets a preset threshold is obtained.

[0052] That is, the local shrinkage rate adjustment mechanism of the generative adversarial network is activated. This means that the generator (G) of the generative adversarial network will perform local optimization processing based on the predicted shrinkage rate values ​​of each position in the preliminary geometric model. Specifically, the generator will adjust different parts of the model based on these shrinkage rate data to ensure that these parts can exhibit the expected shrinkage characteristics in practical applications. The goal of local optimization is to make the geometry of each area closer to the needs of practical applications by adjusting different areas of the model.

[0053] Next, the locally optimized preliminary geometric model is input into the judge (D) of the generative adversarial network. The task of the judge is to evaluate the relationship between the local data of the optimized model and the overall geometric data before optimization. The judge compares the local optimization data with the overall geometric data of the preliminary geometric model before optimization to determine whether the model meets the global geometric requirements after local optimization. The judge evaluates the smoothness (i.e., the consistency of the model surface) and consistency (i.e., the coordination of the various parts of the model) of the optimized model through this comparison.

[0054] Combined with the preset global loss function, the local optimization data and the overall geometric data are evaluated to obtain the initial evaluation data. The global loss function is used to measure the performance of the optimized model in terms of smoothness and consistency. If the value of the loss function is low, it means that the model performs well in these two aspects; otherwise, further optimization may be required. Based on the initial evaluation data, the mechanism adjustment parameters are generated. These parameters are used to adjust the local shrinkage adjustment mechanism in the generative adversarial network to improve the optimization strategy of the generator. The generation process of the mechanism adjustment parameters is to ensure that the generator can better handle the adjustment of the local shrinkage rate in the next round of optimization and improve the overall quality of the model.

[0055] Subsequently, the generator is optimized by the judge based on the mechanism adjustment parameters. This process includes updating the weights and biases of the generator so that it can more effectively apply the adjustment parameters, thereby improving the local optimization effect of the model. The optimized generator will further perform local optimization on the preliminary geometric model. Finally, the preliminary geometric model is iterated with the optimized generator for multiple local optimizations until the initial evaluation data meets the preset threshold. This preset threshold usually represents that the model meets certain standards in terms of smoothness and consistency. Through continuous iterative optimization, the final geometric model will be able to better meet the design requirements and the needs of practical applications.

[0056] In summary, this process uses the generator and judge of the generative adversarial network to continuously improve the accuracy and applicability of the preliminary geometric model through a combination of local optimization and global evaluation, and finally obtains a fully optimized geometric model.

[0057] In one example, based on the initial evaluation data, a mechanism adjustment parameter for adjusting the local shrinkage rate adjustment mechanism is generated, and the generator is optimized based on the mechanism adjustment parameter by the judge, including: based on the initial evaluation data, determining a target area where the local loss is higher than a threshold, and using a gradient descent optimization algorithm on the initial evaluation data to calculate the loss gradient of the target area; determining a mechanism adjustment parameter based on the loss gradient of the target area, wherein the mechanism adjustment parameter includes at least any one of the following: a predicted value of the shrinkage rate of the target area, a weight value of a network layer of the generator; using a back-propagation algorithm by the judge to transmit the mechanism adjustment parameter to each neural network layer of the generator, and optimizing the parameters in each neural network layer.

[0058] That is, based on the initial evaluation data, target areas where the local loss is higher than the threshold are determined. The initial evaluation data is a measurement of the smoothness and consistency of the locally optimized geometric model. By analyzing this data, poorly performing areas of the model can be found, where the local loss values ​​are higher than the preset threshold. These target areas are usually the areas where the generator needs to focus on optimization. Next, a gradient descent optimization algorithm is applied to the initial evaluation data to calculate the loss gradient of the target area. The gradient descent algorithm is a common optimization technique used to minimize the loss function. In this process, the loss gradient can be obtained by calculating the gradient of the loss function relative to each model parameter. These gradients represent the steepest descent direction of the loss function in the target area, thereby providing guidance on how to adjust the model parameters to reduce the loss.

[0059] Then, based on the calculated loss gradient, the mechanism adjustment parameters are determined. The mechanism adjustment parameters are key parameters used to optimize the local shrinkage adjustment mechanism. These parameters may include the predicted shrinkage value of the target area and the weight value of the generator network layer. Specifically, the predicted shrinkage value can adjust the shape of certain areas in the model, while the adjustment of the weight value affects the performance of the generator in the entire network. Finally, the mechanism adjustment parameters are transmitted to each neural network layer of the generator through the back-propagation algorithm by the judge, and the parameters in these layers are optimized. The back-propagation algorithm is a common technique for training neural networks. It calculates the gradient of the error and back-propagates it to each neural network layer, thereby adjusting the network weights to reduce the error. In this process, the judge adjusts the layers of the generator according to the mechanism adjustment parameters to optimize the performance of the generator.

[0060] In summary, this process uses initial evaluation data to identify target areas that need to be optimized, calculates the loss gradient through gradient descent, and then the generation mechanism adjusts the parameters and implements these adjustments in the generator using the back-propagation algorithm. This can improve the effect of the generator in local optimization and ultimately obtain a more accurate and effective geometric model.

[0061] In order to facilitate the understanding of the technical solution of this application, Figure 2-Figure 5 Explain the technical solution.

[0062] Specifically, the main focus of the present invention is the gear accuracy problem caused by the difference between the actual shrinkage rate and the theoretical shrinkage rate of the plastic spiral bevel gear after injection molding. The present invention proposes a method of applying a multi-shrinkage design to adjust the tooth surface of the spiral bevel gear in the tooth width direction, so that the shrinkage rate of each tooth is different in the tooth width direction. The purpose of this design is to improve the accuracy of the injection molded gear and reduce noise.

[0063] First, if Figure 2 and Figure 3 As shown, starting from the traditional tooth flosser structure of a combination of spur teeth and face teeth, the spur teeth are driven by a motor, and the face teeth are used for transmission. In order to reduce the noise of the gearbox, the present invention introduces a spiral bevel gear structure. Although the gears of the traditional structure can achieve power transmission, there are certain deficiencies in noise and precision. Therefore, the use of a spiral bevel gear structure can significantly improve these problems.

[0064] In order to verify the effect of the multi-shrinkage design proposed in the present invention, a standard model with a shrinkage rate of 1.02 was first constructed. Figure 4 As shown. Subsequently, a multi-shrinkage design was adopted to draw a 3D model with different shrinkage rates. Through this design method, the present invention can adjust the shrinkage rates at different positions, thereby optimizing the final accuracy of the gear. The comparison results show that under the same conditions, the precision of the gears molded by the spiral bevel gear designed with multiple shrinkage rates is significantly higher than that of the gears designed with equal shrinkage rates, and the noise value is also improved.

[0065] The specific implementation steps are as follows: First, according to the general shrinkage rate of POM materials, the 3D product drawing of the mold is designed and produced, such as Figure 5 As shown. At this stage, a detailed comparison of the difference between the actual size and the theoretical size is required. By comparing the difference in shrinkage between different positions, especially in the tooth width direction and tooth height direction, it is found that the difference in shrinkage in these directions is significant. In order to solve this problem, two aspects of comparative analysis were carried out.

[0066] In the tooth width direction, we found that the shrinkage difference was about 0.0005, while in the tooth height direction, the shrinkage difference was about 0.0009375. To ensure the accuracy of the gear, we incorporated these differences into the modeling process and adjusted the geometry of the tooth surface to compensate for the shrinkage difference caused by the size change. In the end, we obtained a trimmed 3D spiral bevel gear model with variable shrinkage modeling in both the tooth width and tooth height directions. This model can effectively optimize the accuracy of the gear and make the gear size closer to the design value, thereby achieving higher overall performance and lower noise levels.

[0067] In general, by precisely adjusting the shrinkage rate of the gear design, we can significantly improve the manufacturing accuracy of the gear and reduce the performance differences in actual use. This technical solution has broad application prospects in the design and production of water flosser gears.

[0068] like Figure 6 As shown, an embodiment of the present invention further provides a parameter determination device for a plastic spiral bevel gear with a variable shrinkage ratio, comprising:

[0069] An acquisition unit 1 is used to acquire expected structural data of the plastic spiral bevel gear and application environment data of the plastic spiral bevel gear in a tooth flosser gear box, wherein the application environment data at least includes: expected performance data of the plastic spiral bevel gear under multiple working conditions;

[0070] A construction unit 2 is used to construct a preliminary geometric model and preset shrinkage rate data of the plastic spiral bevel gear based on the expected structure data and the application environment data by using a convolutional neural network based on deep learning;

[0071] An iterative unit 3 is used to input the preliminary geometric model of the plastic spiral bevel gear and the preset shrinkage rate data into a generative adversarial network based on deep learning, and use the generative adversarial network to perform multiple iterative optimizations on the preliminary geometric model based on the preset shrinkage rate data;

[0072] The output unit 4 is used to obtain a target geometric model that is optimized according to the shrinkage rates at different positions in the tooth width direction, and to output parameter data of the target geometric model.

[0073] In this embodiment, for the specific implementation of each unit in the above device embodiment, please refer to the above method embodiment, which will not be repeated here.

[0074] Reference Figure 7 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 7 As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0075] Those skilled in the art will understand that Figure 7 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0076] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0077] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0078] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0079] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for determining parameters of a variable shrinkage plastic spiral bevel gear, characterized in that: The following steps are involved: Acquire expected structural data of the plastic spiral bevel gear and application environment data of the plastic spiral bevel gear in a tooth flosser gear box, wherein the application environment data at least includes: expected performance data of the plastic spiral bevel gear under multiple working conditions; Based on the expected structure data and the application environment data, a preliminary geometric model and preset shrinkage rate data of the plastic spiral bevel gear are constructed using a convolutional neural network based on deep learning; Inputting the preliminary geometric model and the preset shrinkage rate data of the plastic spiral bevel gear into a generative adversarial network based on deep learning, and using the generative adversarial network to perform multiple iterations of optimization on the preliminary geometric model based on the preset shrinkage rate data, wherein the preset shrinkage rate data represents the predicted shrinkage rate value of each position in the preliminary geometric model; Obtaining a target geometric model optimized according to the shrinkage rates at different positions in the tooth width direction, and outputting parameter data of the target geometric model; Among them, the generative adversarial network is used to perform multiple iterative optimizations on the preliminary geometric model based on the preset shrinkage rate data, including: starting the local shrinkage rate adjustment mechanism of the generative adversarial network, so that the generator of the generative adversarial network performs local optimization on the preliminary geometric model according to the predicted shrinkage rate values ​​of each position of the preliminary geometric model; and, the preliminary geometric model after local optimization is input into the judge of the generative adversarial network to determine the local optimization data of the preliminary geometric model after local optimization and the overall geometric data of the preliminary geometric model before optimization; in combination with a preset global loss function, the smoothness and consistency of the preliminary geometric model after local optimization are evaluated for the local optimization data and the overall geometric data to obtain initial evaluation data; based on the initial evaluation data, a mechanism adjustment parameter for adjusting the local shrinkage rate adjustment mechanism is generated, and the generator is optimized by the judge based on the mechanism adjustment parameter; using the optimized generator, the preliminary geometric model is iteratively optimized locally based on the predicted shrinkage rate values ​​of each position until a final geometric model whose initial evaluation data meets a preset threshold is obtained; and, Wherein, based on the initial evaluation data, a mechanism adjustment parameter for adjusting the local shrinkage rate adjustment mechanism is generated, and the generator is optimized based on the mechanism adjustment parameter by the judge, including: based on the initial evaluation data, determining a target area where the local loss is higher than a threshold, and using a gradient descent optimization algorithm on the initial evaluation data to calculate the loss gradient of the target area; determining a mechanism adjustment parameter based on the loss gradient of the target area, wherein the mechanism adjustment parameter includes at least any one of the following: a predicted value of the shrinkage rate of the target area, a weight value of the network layer of the generator; using a back-propagation algorithm by the judge to transmit the mechanism adjustment parameter to each neural network layer of the generator, and optimizing the parameters in each neural network layer.

2. The parameter determination method according to claim 1, characterized in that: After obtaining the target geometric model optimized according to the shrinkage rates at different positions in the tooth width direction, the parameter determination method further includes: Performing simulation verification on the target geometric model to predict deformation data of the target geometric model under different loads, wherein the deformation data at least includes: dimensional change data in the tooth width direction and dimensional change data in the tooth height direction; Based on the deformation data, determining whether the target geometric model passes the expected design requirements; If it passes, continue to execute the step of "outputting the target geometric model"; If it fails, the deformation data is fed back to the deep learning-based generative adversarial network, and the parameters of the generator in the generative adversarial network are adjusted according to the deformation data, and the step of "using the generative adversarial network to perform multiple iterative optimizations on the preliminary geometric model" is continued.

3. The parameter determination method according to claim 1, characterized in that: After obtaining the expected structural data of the plastic spiral bevel gear, the parameter determination method further includes: Converting the expected structure data into a point cloud data format to obtain first structure data, wherein each data point in the first structure data includes an XYZ coordinate value; Filtering outliers in the first structure data to obtain second structure data; Using the DBSCAN algorithm, clustering the data points in the second structure data according to the density values ​​of the data points in the second structure data, to obtain a plurality of data clusters and a plurality of data points that do not belong to any data cluster; Deleting the data points that do not belong to any data cluster as noise, and using a three-dimensional spline interpolation technique to fill in the missing points in the second structure data after the deletion process, to obtain third structure data; and The third structural data is used as the expected structural data to execute the step of "based on the expected structural data and the application environment data, using a convolutional neural network based on deep learning to construct a preliminary geometric model and shrinkage rate preset data of the plastic spiral bevel gear".

4. The parameter determination method according to claim 3, characterized in that: After obtaining the application environment data of the plastic spiral bevel gear in the tooth flosser gear box, the parameter determination method further includes: Normalizing the amplitude of the application environment data so that the variation range of the application environment data is reduced to a preset range, thereby obtaining first environment data; Using a time series generation technology to perform expected performance expansion processing on the first environmental data to obtain second environmental data; Based on the second environment data, a rigid transformation process is performed on the third structure data to align the third structure data with the second environment data to obtain fourth structure data; The fourth structure data and the second environment data are fused to obtain a comprehensive data set, wherein the comprehensive data set is used to execute the step of "constructing a preliminary geometric model of the plastic spiral bevel gear using a convolutional neural network based on deep learning".

5. The parameter determination method according to claim 4, characterized in that: Based on the expected structure data and the application environment data, a preliminary geometric model and shrinkage rate preset data of the plastic spiral bevel gear are constructed using a convolutional neural network based on deep learning, including: Extracting and processing the local geometric features and global morphological features of the plastic spiral bevel gear in the comprehensive data set through the convolution layer of the deep learning-based convolutional neural network to obtain a geometric feature set; Performing dimensionality reduction processing on the geometric feature set through the pooling layer of the convolutional neural network, and after performing the dimensionality reduction processing, introducing the application environment data in the comprehensive data set to obtain a performance feature set; The performance feature set and the geometric feature set are fused through a fully connected layer of the convolutional neural network, and the parameters of the convolutional neural network are optimized through a back propagation algorithm; Until a preliminary geometric model reflecting the geometric features of the plastic spiral bevel gear and preset shrinkage rate data are output, wherein the preset shrinkage rate data represents the predicted shrinkage rate value of each position in the preliminary geometric model.

6. A parameter determination device for a variable shrinkage plastic spiral bevel gear, characterized in that: include: An acquisition unit, used for acquiring expected structural data of the plastic spiral bevel gear and application environment data of the plastic spiral bevel gear in a tooth flosser gear box, wherein the application environment data at least includes: expected performance data of the plastic spiral bevel gear under multiple working conditions; A construction unit, configured to construct a preliminary geometric model and preset shrinkage rate data of the plastic spiral bevel gear based on the expected structure data and the application environment data by using a convolutional neural network based on deep learning; An iterative unit, used for inputting the preliminary geometric model and the preset shrinkage rate data of the plastic spiral bevel gear into a generative adversarial network based on deep learning, and using the generative adversarial network to perform multiple iterative optimizations on the preliminary geometric model based on the preset shrinkage rate data, wherein the preset shrinkage rate data represents the predicted shrinkage rate value of each position in the preliminary geometric model; An output unit, used to obtain a target geometric model optimized according to the shrinkage rates at different positions in the tooth width direction, and to output parameter data of the target geometric model; Among them, the generative adversarial network is used to perform multiple iterative optimizations on the preliminary geometric model based on the preset shrinkage rate data, including: starting the local shrinkage rate adjustment mechanism of the generative adversarial network, so that the generator of the generative adversarial network performs local optimization on the preliminary geometric model according to the predicted shrinkage rate values ​​of each position of the preliminary geometric model; and, the preliminary geometric model after local optimization is input into the judge of the generative adversarial network to determine the local optimization data of the preliminary geometric model after local optimization and the overall geometric data of the preliminary geometric model before optimization; in combination with a preset global loss function, the smoothness and consistency of the preliminary geometric model after local optimization are evaluated for the local optimization data and the overall geometric data to obtain initial evaluation data; based on the initial evaluation data, a mechanism adjustment parameter for adjusting the local shrinkage rate adjustment mechanism is generated, and the generator is optimized by the judge based on the mechanism adjustment parameter; using the optimized generator, the preliminary geometric model is iteratively optimized locally based on the predicted shrinkage rate values ​​of each position until a final geometric model whose initial evaluation data meets a preset threshold is obtained; and, Wherein, based on the initial evaluation data, a mechanism adjustment parameter for adjusting the local shrinkage rate adjustment mechanism is generated, and the generator is optimized based on the mechanism adjustment parameter by the judge, including: based on the initial evaluation data, determining a target area where the local loss is higher than a threshold, and using a gradient descent optimization algorithm on the initial evaluation data to calculate the loss gradient of the target area; determining a mechanism adjustment parameter based on the loss gradient of the target area, wherein the mechanism adjustment parameter includes at least any one of the following: a predicted value of the shrinkage rate of the target area, a weight value of the network layer of the generator; using a back-propagation algorithm by the judge to transmit the mechanism adjustment parameter to each neural network layer of the generator, and optimizing the parameters in each neural network layer.

7. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images