Component size classification method and production equipment

Abstract

This application provides a component size classification method and related equipment. The method includes: acquiring multiple size data of multiple components; determining multiple classification schemes for the multiple components based on the multiple size data; determining a matching scheme for the multiple components based on the multiple classification schemes; calculating a fitness score for each matching scheme; and determining a matching scheme for a preset assembly based on the fitness score for each matching scheme. This application realizes automatic optimization of the component size matching scheme for a preset assembly, effectively improving the assembly yield of the preset assembly.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application is a Chinese national segment application for international patent application filed on May 21, 2024, with application number PCT / CN2024 / 094560 and title "Method and Production Equipment for Component Size Segmentation". It claims priority to patent application filed on February 2, 2024, with application number 202410150681.3 and title "Method and Related Equipment for Size Segmentation of Shaft Assembly". The contents of the aforementioned patent application are incorporated herein by reference. Technical Field

[0003] This application relates to the field of smart terminal manufacturing technology, and in particular to a component size grading method and manufacturing equipment. Background Technology

[0004] In the process of assembling multiple components into a mechanical assembly, the combination of components of different sizes may affect the assembly yield of the mechanical assembly. Therefore, it is necessary to classify the component dimensions of the mechanical assembly according to tolerances. However, the methods for determining the component dimension classification scheme of mechanical assemblies through tolerance classification in related technologies are relatively complex and time-consuming. Summary of the Invention

[0005] In view of the above, it is necessary to provide a component size classification method and production equipment to solve the problem that the method of determining the component size classification scheme of mechanical components by tolerance classification in the above-mentioned related technologies is relatively complicated and time-consuming.

[0006] In a first aspect, embodiments of this application provide a component size grading method applied to production equipment. The method includes: acquiring multiple size data of multiple components; determining multiple grading schemes for the multiple components based on the multiple size data; determining a matching scheme for the multiple components based on the multiple grading schemes; calculating a fitness score for each matching scheme; and determining a matching scheme for a preset component based on the fitness score for each matching scheme.

[0007] The above technical solution automatically optimizes the grading and matching schemes of multiple size data of multiple components, improving the efficiency and accuracy of grading and matching the size data of preset components, and effectively improving the assembly yield of preset components.

[0008] In one possible implementation, determining multiple grading schemes for the multiple components based on the multiple size data includes: setting grading levels for the multiple size data of the multiple components and grading thresholds for each level; and determining the grading levels for the multiple size data of the multiple components and the grading thresholds for each level as multiple grading schemes for the multiple components.

[0009] The above technical solution uses a genetic algorithm to automatically optimize the component size data grading scheme, which improves the efficiency and intelligence of component size grading and matching.

[0010] In one possible implementation, determining the matching scheme of the multiple components based on the multiple grading schemes of the multiple components includes: determining the quantile corresponding to each grading threshold of the multiple size data of the multiple components based on the inverse cumulative distribution function; grouping the multiple size data of the multiple components according to the quantile corresponding to each grading threshold; and matching the grouped size data of different components to obtain the matching scheme of the multiple size data of the multiple components.

[0011] The above technical solution uses a genetic algorithm to automatically optimize the matching scheme of component size data, which improves the efficiency and intelligence of component size classification and matching.

[0012] In one possible implementation, determining the quantile corresponding to each threshold of the multiple size data of the multiple elements according to the inverse cumulative distribution function includes: calculating the size data corresponding to each threshold using the inverse cumulative distribution function of the normal distribution, and using the calculated size data corresponding to each threshold as the quantile.

[0013] The above technical solution can quickly calculate the quantile corresponding to the grading threshold, thereby improving the matching efficiency of the size data of multiple components.

[0014] In one possible implementation, calculating the fitness score for each combination scheme includes: using a fitness function to calculate the fitness score for each combination scheme based on the grouping size data of the different elements.

[0015] Using the above technical solutions, the fitness score for each tiered scheme can be accurately calculated.

[0016] In one possible implementation, determining the combination scheme of the preset component based on the fitness score of each combination scheme includes: if the fitness score of any combination scheme is greater than or equal to a preset fitness score threshold, determining the any combination scheme as the combination scheme of the preset component; or determining the combination scheme with the highest fitness score among multiple combination schemes as the combination scheme of the preset component.

[0017] The above technical solution can accurately determine the combination scheme of preset components.

[0018] In one possible implementation, the step of using a fitness function to calculate the fitness score of each combination scheme based on the grouping size data of the different components includes: inputting the grouping size data of the different components corresponding to each combination scheme into the prediction model to obtain multiple prediction results for each combination scheme; and using the fitness function to obtain the fitness score of each combination scheme based on the multiple prediction results for each combination scheme.

[0019] Using the above technical solution, the fitness score of each combination scheme can be accurately calculated.

[0020] In one possible implementation, the preset component is a rotating shaft component, which has a flattening angle in a flattened state, and the fitness function is a formula for calculating the yield of all predicted flattening angle values ​​under the matching scheme.

[0021] By using the above technical solution, the formula for calculating the yield of the flattening angle prediction value can be used as the fitness function, which can improve the calculation efficiency of fitness score.

[0022] In one possible implementation, the step of using a fitness function to calculate the fitness score of each combination scheme based on the group size data of the different components includes: inputting the group size data of the different components corresponding to each combination scheme into the flattening angle prediction model to obtain multiple predicted flattening angles for each combination scheme; calculating the yield of the multiple predicted flattening angles for each combination scheme to obtain the fitness score of each combination scheme.

[0023] Using the above technical solution, the yield rate of the predicted flattening angle is used as the fitness score of the matching scheme, which facilitates the evaluation of different matching schemes.

[0024] In one possible implementation, the method further includes: extracting multiple features that affect the flattening angle of the pivot assembly, wherein the multiple features include dimensional data of the elements of the pivot assembly; and establishing the flattening angle prediction model based on the multiple features and the flattening angle data.

[0025] The above technical solution establishes a flattening angle prediction model based on multiple features that affect the flattening angle of the rotating shaft assembly, thereby improving the efficiency of determining the corresponding flattening angle data based on the dimensional data of multiple components.

[0026] In one possible implementation, the extraction of multiple features affecting the flattening angle of the pivot assembly includes: obtaining the force information of the element in the flattened state of the pivot assembly; obtaining sampling data corresponding to the size data of the element based on the force information of the element in the flattened state of the pivot assembly; and performing regression analysis on the sampling data corresponding to the size data of the element to obtain the multiple features affecting the flattening angle of the pivot assembly.

[0027] The above technical solution can quickly and accurately determine multiple features that affect the flattening angle in the rotating shaft assembly, effectively improving the efficiency of component size classification and matching.

[0028] In one possible implementation, the regression analysis of the sampled data corresponding to the size data of the component to obtain the multiple features affecting the flattening angle of the pivot assembly includes: standardizing the sampled data; dividing the sampled data into a training set and using the training set to train a fitted cable regression model; obtaining the coefficients of all sampled data according to the trained cable regression model; using the absolute value of the coefficients as the contribution of the component's size data to the flattening angle; sorting the component's size data according to the descending order of the contribution; and selecting the top-ranked size data according to a preset percentage to obtain the multiple features.

[0029] The above technical solution uses a Lasso regression model to screen out features that affect the flattening angle from a large amount of dimensional data. Subsequently, a flattening angle prediction model is established based on the features, and the grading and matching schemes are automatically optimized, which effectively improves the efficiency of establishing the flattening angle prediction model and the efficiency of optimizing the grading and matching schemes.

[0030] In one possible implementation, establishing the flattening angle prediction model based on the plurality of features and the flattening angle data includes: training a preset deep learning model based on the plurality of features and the flattening angle data to obtain the flattening angle prediction model.

[0031] By using the above technical solution, a precise flatness angle prediction model can be obtained by training a preset deep learning model with the component size data and the flatness angle data of the rotating shaft assembly, thereby improving the accuracy of the flatness angle prediction results.

[0032] In one possible implementation, training a preset deep learning model based on the multiple features and the flattening angle data to obtain the flattening angle prediction model includes: establishing a training sample set based on multiple sets of features and the flattening angle data corresponding to each set of features; inputting a set of features and the corresponding flattening angle data from the training sample set as training data into the preset deep learning model, and calculating the loss function value of the preset deep learning model; if the loss function value of the preset deep learning model is greater than a preset value, adjusting the parameters of the preset deep learning model, and continuing to train the preset deep learning model until the loss function value of the preset deep learning model is less than or equal to the preset value, determining that the preset deep learning model has been trained, and determining the trained preset deep learning model as the flattening angle prediction model.

[0033] By employing the above technical solution, a flattening angle prediction model is established by training a preset deep learning model using multiple features that affect the flattening angle of the rotating shaft assembly. This can improve the accuracy of the flattening angle prediction value output by the flattening angle prediction model.

[0034] Secondly, embodiments of this application provide a component size grading method applied to production equipment. The method includes: acquiring multiple size data of multiple first components and multiple second components; determining multiple grading schemes for the multiple first components and multiple second components based on the multiple size data; determining a matching scheme for the multiple first components and multiple second components based on the multiple grading schemes; calculating a fitness score for each matching scheme; and determining a matching scheme for multiple preset components based on the fitness score for each matching scheme, wherein each preset component includes a first component and a second component.

[0035] In one possible implementation, determining multiple grading schemes for the multiple first elements and the multiple second elements based on the multiple size data includes: setting grading levels and grading thresholds for the multiple size data of the multiple first elements and the multiple second elements; and determining the grading levels and grading thresholds for the multiple size data of the multiple first elements and the multiple second elements as multiple grading schemes for the multiple first elements and the multiple second elements.

[0036] In one possible implementation, determining the pairing scheme of the plurality of first elements and the plurality of second elements based on the plurality of grading schemes includes: determining the quantile corresponding to each grading threshold of the plurality of size data of the plurality of first elements and the plurality of second elements based on the inverse cumulative distribution function; grouping the plurality of size data of the plurality of first elements and the plurality of second elements according to the quantile corresponding to each grading threshold; and pairing the grouped size data of the plurality of first elements and the plurality of second elements to obtain the pairing scheme of the plurality of size data of the plurality of first elements and the plurality of second elements.

[0037] In one possible implementation, calculating the fitness score for each combination scheme includes: using a fitness function to calculate the fitness score for each combination scheme based on the grouping size data of the plurality of first elements and the plurality of second elements.

[0038] In one possible implementation, the preset component is a rotating shaft component, which has a flattening angle in a flattened state, and the fitness function is a formula for calculating the yield of all predicted flattening angle values ​​under the matching scheme.

[0039] In one possible implementation, the step of using a fitness function to calculate the fitness score of each combination scheme based on the grouped size data of the plurality of first elements and the plurality of second elements includes: inputting the grouped size data of the plurality of first elements and the plurality of second elements corresponding to each combination scheme into the flattening angle prediction model to obtain multiple predicted flattening angles for each combination scheme; calculating the yield of the multiple predicted flattening angles for each combination scheme to obtain the fitness score of each combination scheme.

[0040] Thirdly, embodiments of this application provide a production device, the production device comprising: a feeding component for receiving multiple components; a processing component for executing the above-described component size grading method to determine a preset component matching scheme; a sorting component for matching the multiple components according to the preset component matching scheme; and a discharging component for outputting the matched multiple components.

[0041] In one possible implementation, the production equipment further includes a measuring component for measuring multiple dimensional data of the plurality of elements and sending the measured multiple dimensional data of the plurality of elements to the processing component.

[0042] In one possible implementation, the production equipment further includes an assembly component for assembling the assembled plurality of elements into the preset assembly.

[0043] Fourthly, embodiments of this application provide a production device, the production device comprising: a feeding component for receiving a plurality of first components and a plurality of second components; a processing component for performing the above-described component size grading method to determine a matching scheme for a plurality of preset components; a sorting component for matching the plurality of first components and the plurality of second components according to the matching scheme for the plurality of preset components; and a discharging component for outputting the matched plurality of first components and the plurality of second components.

[0044] Fifthly, this application provides a computer storage medium storing program instructions that, when executed on a production device, cause the processor of the production device to perform the aforementioned component size classification method.

[0045] Furthermore, the technical effects brought about by the second to fifth aspects can be found in the descriptions of the methods in the above-mentioned method section, and will not be repeated here. Attached Figure Description

[0046] Figure 1 This is a perspective view of a terminal device provided in an embodiment of this application when it is in a folded state.

[0047] Figure 2This is a perspective view of a terminal device provided in an embodiment of this application when it is in a flattened state.

[0048] Figure 3 This is a perspective view of another foldable terminal device provided in an embodiment of this application.

[0049] Figure 4 This is a flowchart of a component size classification method provided in an embodiment of this application.

[0050] Figure 5 This is a flowchart of a component size classification method provided in another embodiment of this application.

[0051] Figure 6 This is a flowchart of a component size classification method provided in another embodiment of this application.

[0052] Figure 7 This is a flowchart of feature extraction provided in one embodiment of this application.

[0053] Figure 8 This is a flowchart of model training provided in one embodiment of this application.

[0054] Figure 9 This is a perspective view of the main swing arm provided in one embodiment of this application.

[0055] Figures 10 to 12 This is a perspective view of a base provided in an embodiment of this application.

[0056] Figure 13 This is a perspective view of the auxiliary swing arm provided in one embodiment of this application.

[0057] Figures 14 to 16 This is a perspective view of a wedge block provided in an embodiment of this application.

[0058] Figure 17 This is a schematic diagram of the structure of a preset deep learning model provided in an embodiment of this application.

[0059] Figure 18 This is a flowchart of the grading level and grading threshold optimization provided in one embodiment of this application.

[0060] Figure 19 This is a schematic diagram of the grading levels and grading thresholds provided in an embodiment of this application.

[0061] Figure 20 This is a schematic diagram of the architecture of a production equipment provided in one embodiment of this application.

[0062] Figure 21 This is a schematic diagram illustrating an application scenario of the component size classification method provided in one embodiment of this application. Detailed Implementation

[0063] The terms "first" and "second" used in the embodiments of this application are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0064] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this application's specification is for the purpose of describing particular embodiments only and is not intended to limit the application. It should be understood that, unless otherwise stated, " / " in this application means "or". For example, A / B can mean A or B. "And / or" in this application is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. "At least one" refers to one or more. "More than one" refers to two or more. For example, at least one of a, b, or c can represent: a, b, c, a and b, a and c, b and c, and a, b, and c. Where there is no conflict, the following embodiments and features described herein can be combined with each other.

[0065] For ease of understanding, the following descriptions of some concepts related to the embodiments of this application are provided for reference:

[0066] Finite element simulation is a numerical simulation method that treats a continuum as a discrete set of a finite number of units connected together in a certain way. It is used to solve thermal, mechanical, and electromagnetic problems of the continuum. Finite element simulation is based on the finite element method, which decomposes complex structures or systems into a finite number of simple elements. Numerical calculations are then performed on these elements to obtain the behavior of the overall system.

[0067] Monte Carlo sampling, also known as statistical simulation sampling, is a method of approximating numerical calculations by randomly sampling from a probabilistic model. Specifically, simulation refers to replacing or simulating a certain characteristic or partial state of a real or abstract system with another system (called a simulation model). To solve a problem, it is transformed into a problem of solving a probabilistic model. Then, a large number of random numbers that conform to the model are generated, and the generated random numbers are analyzed to solve the problem. This method is called random simulation sampling, also known as Monte Carlo sampling.

[0068] Tolerance: The allowable variation in the actual value of a parameter. Parameters can be geometric parameters in machining, or parameters from disciplines such as physics, chemistry, and electricity. In mechanical manufacturing, the purpose of setting tolerances is to determine the geometric parameters of a product, ensuring that their variation is within a certain range to meet interchangeability or fit requirements.

[0069] Position tolerance: Used to specify the allowable range of positional tolerance or positional deviation of a part.

[0070] The flattening angle is a parameter used to describe displacement relationships under stress, and is frequently used in mechanical engineering and structural analysis. The flattening angle describes the twisting or deviation of an object under stress, and allows us to understand the deformation state of a structure.

[0071] In the process of assembling multiple components into a mechanical assembly, the combination of components of different sizes may affect the assembly yield of the mechanical assembly. Therefore, it is necessary to classify the component dimensions of the mechanical assembly according to tolerances. However, the methods for determining the component dimension classification scheme of mechanical assemblies through tolerance classification in related technologies are relatively complex and time-consuming.

[0072] With the development of smart terminal technology, users have an increasingly strong demand for large-screen terminals. However, large-screen terminals are not convenient to carry due to their large screen size. Foldable terminal devices use a folding screen design, which increases the screen size without significantly increasing the overall size of the device, and has become the mainstream large-screen terminal solution.

[0073] Foldable terminal devices typically use hinge assemblies to achieve the folding function. However, during the assembly process of foldable terminal devices, hinge assemblies are prone to defects such as under-folding or over-folding. As a result, the production yield of hinge assemblies is low, which in turn leads to a low production yield of foldable terminal devices.

[0074] See Figure 1 The image shown is a perspective view of the terminal device provided in this application embodiment when it is in a folded state. The terminal device 200 provided in this application embodiment is a foldable terminal device, including a first display screen 10, a second display screen 20, and a hinge assembly 30. The first display screen 10 is the outer screen of the terminal device 200, and the second display screen 20 is the inner screen of the terminal device 200. The terminal device 200 is in a folded or unfolded state based on the rotation of the hinge assembly 30. In the folded state, the terminal device 100 uses the first display screen 10.

[0075] See Figure 2 The image shown is a perspective view of the terminal device 200 in a flattened state according to an embodiment of this application. The user can drive the rotating shaft assembly 30 to rotate, thereby controlling the terminal device 200 from... Figure 1The folded state shown has changed to Figure 2 The flattened state shown indicates that the terminal device 100 uses a second display screen 20, which includes a first display area 21, a second display area 22, and a third display area 23.

[0076] See Figure 3 The image shown is a perspective view of another foldable terminal device provided in an embodiment of this application. The terminal device 200 includes a display screen 40 and a hinge assembly 50. The display screen 40 includes a first display area 41, a second display area 42, and a third display area 43.

[0077] Taking the hinge assembly of a foldable terminal device as an example, the hinge assembly is a crucial component enabling the folding and flattening functions of the foldable terminal device. However, during the assembly process of foldable terminal devices, due to limitations in assembly technology, the assembled hinge assembly is prone to defects such as under-folding and over-folding, resulting in a low mass production yield. Specifically, under-folding and over-folding of the hinge assembly will cause corresponding under-folding or over-folding of the display screen of the foldable terminal device. Under-folding refers to the flattening angle of the display screen in the flattened state being less than 180 degrees, that is, the angle between the first display area 21 or the third display area 23 and the second display area 22 is less than 180 degrees. Over-folding refers to the flattening angle of the display screen in the flattened state being greater than 180 degrees, that is, the angle between the first display area 21 or the third display area 23 and the second display area 22 is greater than 180 degrees. Both under-folding and over-folding will result in an uneven display screen in the flattened state of the foldable terminal device, affecting the display effect of each display area and reducing the user experience.

[0078] Due to the low precision assembly threshold and the large processing tolerances of materials, blind assembly methods cannot be used for the current assembly of foldable terminal devices, necessitating the use of tolerance grading technology. However, the complex structure of the hinge assembly and the slab angle are easily affected by various factors such as material deformation and fluctuations in processing tolerances, making it technically challenging to establish a high-accuracy automatic grading scheme. In the traditional tolerance simulation-based grading and matching scheme, the grading threshold requires Design of Experiments (DOE) step calculations, necessitating continuous adjustments to find the optimal yield threshold. This calculation method is not intelligent enough, requiring a long calculation cycle (e.g., two days) each time, resulting in significant computation time. Furthermore, the DOE step calculation gradient is large; for example, using 10% as a threshold step can easily miss the optimal threshold scheme.

[0079] To address the above issues, this application provides a component size classification method. The component size classification scheme is intelligently and automatically optimized, reducing the difficulty of implementing component size classification, shortening the time required to determine the component size classification scheme of mechanical components, and effectively improving the assembly yield of mechanical components, such as improving the yield of the flattening angle of the rotating shaft assembly.

[0080] See Figure 4 The diagram shown is a flowchart of a component size classification method provided in an embodiment of this application. The method is applied in production equipment, and the component size classification method includes:

[0081] S101, acquire multiple dimension data of multiple components.

[0082] In one embodiment of this application, multiple dimensional data of multiple components can be measured using a measuring device. In another embodiment of this application, the memory of the production equipment can also pre-store multiple dimensional data of multiple components, thus allowing the multiple dimensional data of multiple components to be retrieved from the memory of the production equipment. In another embodiment of this application, the production equipment can also receive input multiple dimensional data of multiple components. In one embodiment of this application, the multiple components can be components for assembling into a preset assembly. For example, the preset assembly is a hinge assembly of a foldable terminal device, and the multiple components include a main swing arm, a base, a secondary swing arm, and a wedge block of the hinge assembly.

[0083] S102 determines multiple grading schemes for multiple components based on multiple dimensional data.

[0084] In one embodiment of this application, multiple size data of multiple components are set into multiple grading levels and grading thresholds for each level, and the multiple size data of multiple components and grading thresholds for each level are determined as multiple grading schemes for multiple components.

[0085] S103, determine the combination scheme of multiple components based on multiple grading schemes of multiple components.

[0086] In one embodiment of this application, the quantiles corresponding to each threshold of multiple size data of multiple components are determined according to the inverse cumulative distribution function. The multiple size data of multiple components are grouped according to the quantiles corresponding to each threshold. The grouped size data of different components are matched to obtain a matching scheme of multiple size data of multiple components.

[0087] S104, calculate the fitness score for each combination scheme.

[0088] In one embodiment of this application, a fitness function is used to calculate the fitness score of each combination scheme based on the grouping size data of different components. The grouping size data of different components corresponding to each combination scheme are input into the prediction model to obtain multiple prediction results for each combination scheme. The fitness function is then used to obtain the fitness score of each combination scheme based on the multiple prediction results for each combination scheme.

[0089] In one embodiment of this application, the preset component is a rotating shaft component, which has a flattening angle in a flattened state. The fitness function is a formula for calculating the yield of all predicted flattening angle values ​​under a given configuration scheme, and the prediction model is a flattening angle prediction model. The grouped size data of different components corresponding to each configuration scheme are input into the flattening angle prediction model to obtain multiple predicted flattening angles for each configuration scheme. The yield of the multiple predicted flattening angles for each configuration scheme is calculated to obtain the fitness score for each configuration scheme.

[0090] S105, determine the combination scheme of the preset components based on the fitness score of each combination scheme.

[0091] In one embodiment of this application, if the fitness score of any combination is greater than or equal to a preset fitness score threshold, that combination is determined as a combination of preset components. In another embodiment of this application, the combination with the highest fitness score among multiple combinations can also be determined as a combination of preset components.

[0092] In one embodiment of this application, the component size classification method further includes: extracting multiple features that affect the flattening angle of the pivot assembly, wherein the multiple features include multiple size data of multiple components of the pivot assembly; and establishing a flattening angle prediction model based on the multiple features and the flattening angle data.

[0093] In one embodiment of this application, multiple features affecting the flattening angle of the pivot assembly are extracted, including: obtaining the force information of multiple components of the pivot assembly in the flattened state; obtaining sampling data corresponding to the size data of the components based on the force information of the components in the flattened state; and performing regression analysis on the sampling data corresponding to the size data of the components to obtain multiple features affecting the flattening angle of the pivot assembly.

[0094] In one embodiment of this application, regression analysis is performed on the sampled data corresponding to the component's size data to obtain multiple features affecting the flattening angle of the rotating shaft assembly. This includes: standardizing the sampled data; dividing the sampled data into a training set and using the training set to train a fitted cable regression model; obtaining the coefficients of all sampled data based on the trained cable regression model; using the absolute value of the coefficients as the contribution of the component's size data to the flattening angle; sorting the component's size data according to the descending order of contribution; and selecting the top-ranked size data according to a preset percentage to obtain multiple features.

[0095] In one embodiment of this application, a planarization angle prediction model is established based on multiple features and planarization angle data, including: training a preset deep learning model based on multiple features and planarization angle data to obtain the planarization angle prediction model.

[0096] In one embodiment of this application, a pre-defined deep learning model is trained based on multiple features and flattening angle data to obtain a flattening angle prediction model. This includes: establishing a training sample set based on multiple sets of features and the flattening angle data corresponding to each set of features; inputting a set of features and the corresponding flattening angle data from the training sample set as training data into the pre-defined deep learning model, and calculating the loss function value of the pre-defined deep learning model; if the loss function value of the pre-defined deep learning model is greater than a preset value, adjusting the parameters of the pre-defined deep learning model, and continuing to train the pre-defined deep learning model until the loss function value of the pre-defined deep learning model is less than or equal to the preset value, determining that the pre-defined deep learning model training is complete, and identifying the trained pre-defined deep learning model as the flattening angle prediction model.

[0097] See Figure 5 The diagram shown is a flowchart of a component size classification method provided in another embodiment of this application. The method is applied in production equipment, and the component size classification method includes:

[0098] S201, acquire multiple dimension data of multiple first elements and multiple second elements.

[0099] S202, determine multiple grading schemes for multiple first elements and multiple second elements based on multiple dimensional data.

[0100] S203, determine the combination scheme of multiple first elements and multiple second elements based on multiple grading schemes of multiple first elements and multiple second elements.

[0101] S204, calculate the fitness score for each combination scheme.

[0102] S205, based on the fitness score of each combination scheme, determine the combination scheme of multiple preset components, wherein each preset component includes a first element and a second element.

[0103] See Figure 6 The diagram shown is a flowchart of a component size classification method provided in another embodiment of this application. The method is applied in production equipment, and the component size classification method includes:

[0104] S301, extract multiple features that affect the flattening angle of the pivot assembly, including the dimensional data of multiple components.

[0105] In one embodiment of this application, multiple features affecting the flattening angle of the hinge assembly are extracted based on the design drawing of the hinge assembly. The design drawing of the hinge assembly is a three-dimensional model image, which is generated by drawing the hinge assembly to scale according to the standard dimensions beforehand. The design drawing of the hinge assembly is input into finite element simulation software. The finite element simulation software is used to perform force analysis on multiple components of the hinge assembly in the flattened state to determine the force information of multiple components, including the force contact points and force deformation. The force contact points and force deformation of multiple components are input into tolerance simulation software. The tolerance simulation software performs Monte Carlo sampling on the dimensional data of multiple components to determine the sampling data corresponding to the dimensional data of multiple components. Regression analysis is performed on the sampling data to determine multiple features in the sampling data, wherein the features include the dimensional data of multiple components, for example, the dimensional data includes the positional tolerance of the components.

[0106] S302, establishes a flattening angle prediction model based on multiple features and flattening angle data.

[0107] In one embodiment of this application, a preset deep learning model is trained based on multiple features and flattening angle data to obtain a flattening angle prediction model. The preset deep learning model is a fully connected neural network model, and the multiple features and flattening angle data serve as training data for the fully connected neural network model. The multiple features are used as input data to the fully connected neural network model, and the flattening angle data is used as output data to train the fully connected neural network model. The trained fully connected neural network model is then used as the flattening angle prediction model. In other embodiments of this application, the preset deep learning model can also be other types of machine learning models, such as convolutional neural network models.

[0108] S303, based on the flattening angle prediction model and genetic algorithm, determines multiple grading schemes and combination schemes for multiple components. The grading scheme includes the grading level and grading threshold of the dimensional data for each component. The combination scheme includes the grading combination of dimensional data between different components.

[0109] In one embodiment of this application, the grading levels of the size data for each component are customized, for example, they can be set to 2 to n levels, where n can be an integer greater than 2. The grading thresholds of the size data for each component are traversed, and the quantiles of each customized grading threshold are determined according to the inverse cumulative distribution function. Monte Carlo sampling calculations are performed for different grading thresholds using a genetic algorithm. The size data of different grading levels of multiple components are combined, and the size data of different grading levels of the combined components are input into the flattening angle prediction model to obtain the corresponding predicted flattening angle. The overall defect rate after combining each grading level is determined based on the predicted flattening angle and the preset flattening angle range. The optimal grading scheme and combination scheme are determined based on the minimum value of the overall defect rate.

[0110] The above embodiments of this application determine multiple features affecting the flattening angle in the pivot assembly through finite element stress analysis, Monte Carlo sampling, and regression analysis. Based on these features, a preset deep learning model is trained to establish a flattening angle prediction model. The flattening angle prediction model and a genetic algorithm are used to automatically optimize the grading and matching schemes of the component's size data, thereby improving the efficiency and accuracy of grading the size data of the pivot assembly and effectively increasing the assembly yield of the pivot assembly.

[0111] See Figure 7 The diagram shown is a flowchart of feature extraction provided in an embodiment of this application.

[0112] S3011. Input the design drawing of the rotating shaft assembly into the finite element simulation software to obtain the force information of multiple components of the rotating shaft assembly in the flattened state.

[0113] In one embodiment of this application, the design drawing of the pivot assembly is a finite element model generated in advance based on a standard-sized pivot assembly and drawn to scale. For example, the finite element simulation software can be ANSYS or Abaqus software, which can perform force analysis on multiple components of the pivot assembly in a flattened state to determine the force information of multiple components in the pivot assembly. The flattened state refers to the pivot assembly being flattened at an angle of 180 degrees. The force information of multiple components includes, but is not limited to, the force contact points m of the components. i and the deformation δ i Among them, the force contact point m i The rotation axis assembly is represented using three-dimensional coordinates in a three-dimensional Cartesian coordinate system, which can be established based on the position of the rotating shaft assembly. In one embodiment of this application, the rotating shaft assembly includes, but is not limited to, a main swing arm, a secondary swing arm, a base, a wedge block, a hinge, a main shaft, a support portion, and a slider.

[0114] In one embodiment of this application, a geometric model of the rotating shaft assembly (e.g., a design drawing of the rotating shaft assembly) is created or imported using finite element simulation software. The geometric model of the rotating shaft assembly is discretized into a finite element mesh. Material properties, such as elastic modulus, Poisson's ratio, density, etc., are defined for each element of the rotating shaft assembly. These properties determine the material's response to external loads. Boundary conditions and loadings are defined on the finite element model of the rotating shaft assembly. Boundary conditions and loadings include constraint conditions (e.g., fixed supports or axial constraints) and loading conditions (e.g., force, pressure, temperature, etc.). In this embodiment, the boundary conditions and loadings are based on the rotating shaft assembly being in a flattened state. The finite element model of the rotating shaft assembly is solved using numerical methods (e.g., the finite element method), solving linear or nonlinear equations on the mesh to obtain the nodal displacements and stress distributions in the finite element model. The results are visualized and analyzed using finite element simulation software. By viewing the displacement, stress, deformation, and other results, the stress and deformation of the rotating shaft assembly under a given load (i.e., in a flattened state) are evaluated, and the contact conditions between multiple elements of the rotating shaft assembly and the stress deformation of multiple elements are obtained. The contact between multiple components is represented by the force-bearing contact points.

[0115] S3012 inputs the force information of multiple components of the rotating shaft assembly in the flattened state into the tolerance simulation software to obtain the sampling data corresponding to the dimensional data of multiple components.

[0116] In one embodiment of this application, the tolerance simulation software is Monte Carlo simulation software. Using this software, based on the force information of multiple components in the flattened state of the shaft assembly, the dimensional data of the multiple components are sampled a preset number of times using Monte Carlo simulation to obtain the sampled data corresponding to the dimensional data of the multiple components. For example, the preset number of times may be one million, one million two hundred thousand, or other numbers; this embodiment of the application does not limit this. The dimensional data of the multiple components includes, but is not limited to, the positional tolerances of each surface of the components.

[0117] In one embodiment of this application, the stress information of multiple components is input as a virtual feature into the three-dimensional tolerance simulation software of the mechanism. The tolerance simulation software outputs the dimensional data of each component that matches the stress information of the corresponding component. The dimensional data of each component that matches the stress information of the corresponding component is subjected to a preset number of Monte Carlo samplings to obtain the sampling data corresponding to the dimensional data of each component.

[0118] S3013 performs regression analysis on the sampled data corresponding to the dimensional data of multiple components to obtain multiple features that affect the flattening angle of the rotating shaft assembly.

[0119] In one embodiment of this application, a preset regression analysis model is used to perform regression analysis on the sampled data corresponding to the dimensional data of multiple components. This yields the contribution of each dimensional data point for each component, identifies key influencing factors, and sorts the multiple dimensional data points of the multiple components in descending order of contribution. In other words, the multiple dimensional data points of the multiple components are sorted in descending order of contribution, and the dimensional data points of the components with the highest preset percentage are used as features affecting the flattening angle of the rotating shaft assembly. For example, the preset regression analysis model is a LASSO (cable) regression analysis model, and the preset percentage is 90%, 95%, or other percentages.

[0120] In one embodiment of this application, the sampled data is standardized, for example, by standardizing the sampled data to a mean of 0 and a standard deviation of 1, eliminating differences in measurement units and proportions between different feature variables. The sampled data is divided into a training set and a test set. The training set includes multiple sets of training data, where the input data for each set is the sampled data, i.e., the size data of the component, and the output data is the flattening angle of the hinge assembly. The test set includes multiple sets of test data, where the input data for each set is the sampled data, i.e., the size data of the component, and the output data is the flattening angle of the hinge assembly. The LASSO regression model is fitted and trained using the training set. During the fitting process, the sparsity of the features is controlled by adjusting the L1 regularization parameter; for example, the optimal L1 regularization parameter can be found using methods such as cross-validation or grid search. Based on the trained LASSO regression model, the coefficients of all features are obtained, and the absolute values ​​of the coefficients are sorted in descending order to filter out the features. The model performance is evaluated using the test set, for example, using metrics such as root mean square error (RMSE) and R-squared.

[0121] In one embodiment of this application, the absolute value of the coefficient is used as the contribution of each feature (i.e., the size data of each element) to the flattening angle of the rotating shaft assembly. Multiple features are sorted in descending order of contribution, and the feature with the highest percentage is selected to obtain the final feature.

[0122] The embodiments described above in this application can accurately screen out multiple features affecting the flattening angle in the rotating shaft assembly through finite element stress analysis, Monte Carlo sampling, and regression analysis.

[0123] See Figure 8 The diagram shown is a flowchart of model training provided in an embodiment of this application.

[0124] S3021, Establish a training sample set based on multiple sets of features and the flattening angle data corresponding to each set of features.

[0125] In one embodiment of this application, the feature corresponding to each flattening angle data includes the dimensional data of multiple components, where the dimensional data of the components are the dimensional data of the key structural feature surfaces of the components. For example, the multiple components include a main swing arm A, a base B, a secondary swing arm C, and a wedge block D. See also... Figure 9 The image shown is a perspective view of a main swing arm provided in an embodiment of this application. The dimensional data of the key structural feature surfaces of the main swing arm 11 include the positional tolerance A1 of the mating hole 111 between the main swing arm and the wedge block, the positional tolerance A2 of the stop surface 112 between the main swing arm and the base, and the positional tolerance A3 of the mating arc surface 113 between the main swing arm and the base. (See reference...) Figure 10-12 The image shown is a perspective view of a base provided in an embodiment of this application. The dimensional data of the key structural feature surfaces of the base 12 include the positional tolerance B1 of the mating pin 121 between the base and the secondary swing arm, the positional tolerance B2 of the stop surface 122 between the base and the secondary swing arm, the positional tolerance B3 of the mating arc surface 123 between the base and the main swing arm, and the positional tolerance B4 of the stop surface 124 between the base and the main swing arm. See also... Figure 13 The image shown is a perspective view of a secondary swing arm provided in an embodiment of this application. The dimensional data of the key structural feature surfaces of the secondary swing arm 13 include the positional tolerance C1 of the mating surface 131 between the secondary swing arm and the wedge block, the positional tolerance C2 of the stop surface 132 between the secondary swing arm and the base, and the positional tolerance C3 of the mating hole 133 between the secondary swing arm and the base. (See also...) Figure 14-16 The image shown is a perspective view of a wedge block provided in an embodiment of this application. The dimensional data of the key structural feature surfaces of the wedge block 14 include the positional tolerance D1 of the mating hole 141 between the wedge block and the main swing arm, and the positional tolerance D2 of the mating surface 142 between the wedge block and the auxiliary swing arm.

[0126] S3022: Take a set of features and corresponding flattening angle data as training data, input them into a preset deep learning model, and calculate the loss function value of the preset deep learning model.

[0127] See Figure 17 The diagram shown is a schematic representation of the structure of a preset deep learning model provided in one embodiment of this application. In one embodiment of this application, the preset deep learning model is a fully connected neural network model (also known as a multilayer perceptron). The fully connected neural network model is initialized by importing it into the deep learning framework and setting initial parameters. Then, the set of features is used as the input data of the fully connected neural network model, and the flattened angle data is used as the output data of the fully connected neural network model. The fully connected neural network model is trained, and the loss function of the preset deep learning model is calculated.

[0128] In one embodiment of this application, the fully connected neural network model includes an input layer, two hidden layers, and an output layer. The input layer comprises 34 neurons, corresponding to the first parameter of the linear transformation function nn.Linear(34, 128). The first hidden layer is a fully connected layer, defined by the linear transformation function self.fc1 = nn.Linear(34, 128), and includes 128 neurons. The first hidden layer also includes a Dropout layer (self.dropout1 = nn.Dropout(p = 0.5)) to reduce overfitting. The second hidden layer is also a fully connected layer, defined by the linear transformation function self.fc2 = nn.Linear(128, 64), and includes 64 neurons. The second hidden layer also includes a Dropout layer (self.dropout2 = nn.Dropout(p = 0.5)) to reduce overfitting. The output layer is also a fully connected layer, defined by the linear transformation function self.fc3 = nn.Linear(64, 1), and includes one neuron, used to generate the final output of the fully connected neural network model.

[0129] In one embodiment of this application, the activation function of the fully connected neural network model is self.relu = nn.ReLU(), and the computation code of the fully connected neural network model is:

[0130] def__init__(self):

[0131] super(Net,self).__init__();

[0132] self.fc1=nn.Linear(34,128);

[0133] self.dropout1=nn.Dropout(p=0.5);

[0134] self.fc2=nn.Linear(128,64);

[0135] self.dropout2=nn.Dropout(p=0.5);

[0136] self.fc3 = nn.Linear(64, 1);

[0137] self.relu = nn.ReLU().

[0138] In one embodiment of this application, the loss function of the preset deep learning model is data loss, which is the mean square error (MSE) between the predicted flattening angle and the actual flattening angle data.

[0139] S3023, Determine whether the loss function value of the preset deep learning model is less than or equal to the preset value. If the loss function value of the preset deep learning model is greater than the preset value, execute S3024; if the loss function value of the preset deep learning model is less than or equal to the preset value, execute S3025.

[0140] S3024, adjust the parameters of the preset deep learning model, and then return to execute S3022. The parameters of the fully connected neural network model include the connection weights between different layers, the bias value of each neuron, etc.

[0141] S3025, confirm that the preset deep learning model has been trained and determine the trained preset deep learning model as the flattening angle prediction model.

[0142] In one embodiment of this application, if the loss function of the preset deep learning model is less than or equal to a preset value, it is determined that the preset deep learning model has converged, and then it is determined that the training of the preset deep learning model has been completed.

[0143] The above embodiments of this application use the features of the pivot assembly and the flattening angle data to train the neural network model to obtain the flattening angle prediction model. By inputting the component size data of the pivot assembly into the flattening angle prediction model, the corresponding predicted flattening angle data can be obtained quickly, thus improving the calculation efficiency of the flattening angle data.

[0144] See Figure 18 The diagram shown is a flowchart of the grading level and grading threshold optimization provided in an embodiment of this application.

[0145] S3031, a fitness function that determines the matching scheme of the size data of multiple components.

[0146] In one embodiment of this application, the fitness function for determining the matching scheme of the size data of multiple components is a formula for calculating the yield of all flattening angle prediction values ​​under the matching scheme. The yield of all flattening angle prediction values ​​corresponding to each matching scheme is used as the fitness score. Specifically, the size data of multiple components corresponding to each matching scheme are input into the flattening angle prediction model to obtain multiple flattening angle prediction values ​​for each matching scheme. It is determined whether each flattening angle prediction value is within a preset flattening angle range. If the flattening angle prediction value is within the preset flattening angle range, it is determined to be qualified; if the flattening angle prediction value is not within the preset flattening angle range, it is determined to be unqualified. The ratio between all qualified flattening angle prediction values ​​and all flattening angle prediction values ​​is calculated to obtain the yield of all flattening angle prediction values. The matching scheme includes the matching of size data levels between different components. For example, the preset flattening angle range is 180.2 degrees to 181.8 degrees, that is, greater than or equal to 180.2 degrees and less than or equal to 181.8 degrees.

[0147] In another embodiment of this application, the fitness function of the matching scheme of the size data of multiple components can also be the reciprocal of the sum of the mean square error of the size data of each component in each segment of the matching scheme, the reciprocal of the sum of the standard deviation of the size data of each component in each segment of the matching scheme, and the sum of the yield of all flattening angle prediction values.

[0148] In one embodiment of this application, the number of size categories for the component's dimensional data is the number of categories after size classification, and the classification threshold is the percentage value for classifying the dimensional data. See also... Figure 19 The diagram shown illustrates the grading levels and grading thresholds provided in an embodiment of this application. For example, for multiple dimensional data of a component conforming to a normal distribution, the number of grading levels is 5, the number of grading thresholds is 4, and grading threshold one is 15%, grading threshold two is 30%, grading threshold three is 70%, and grading threshold four is 85%. Therefore, the grading levels include: level one is 0%–15%, level two is 15%–30%, level three is 30%–70%, level four is 70%–85%, and level five is 85%–100%.

[0149] S3032 sets the size data of multiple components into different grading levels and the grading threshold for each level, and determines the quantile corresponding to each grading threshold based on the inverse cumulative distribution function.

[0150] In one embodiment of this application, a genetic algorithm is used to optimize the grading levels and grading thresholds of the dimensional data of multiple components. First, an initial population representing different grading schemes is randomly generated. Multiple components and multiple dimensional data for each component are initialized. The grading levels of the multiple dimensional data for each component are customized to 2 to n levels, where n is a positive integer greater than 2, for example, n is set to 4. For example, the dimensional data of a component may be the positional tolerance of its critical structural feature surface, where the critical structural feature surface refers to a specific structural feature that plays an important role in the component's performance, reliability, or function.

[0151] In one embodiment of this application, a target probability is determined, and the size data corresponding to the target probability is calculated using the inverse cumulative distribution function of a preset probability distribution function. The target probability is a threshold for classifying multiple size data of a component, and the size data corresponding to each target probability is used as the quantile corresponding to the threshold. The initialized multiple size data of the component conform to a normal distribution, and the inverse cumulative distribution function is the inverse cumulative distribution function corresponding to the normal distribution. The inverse cumulative distribution function maps the target probability to the corresponding cumulative distribution function value, and the cumulative distribution function value is used as the quantile corresponding to each target probability. Accordingly, the threshold is input into the inverse cumulative distribution function, and the quantile corresponding to the threshold is output through the inverse cumulative distribution function, thereby obtaining the size data corresponding to each threshold among the multiple size data.

[0152] In one embodiment of this application, after customizing the grading levels of the size data, the number of grading thresholds for the size data is one less than the number of grading levels. Each grading threshold of the size data for each component is iterated over, and each grading threshold is treated as a target probability and input into the inverse cumulative distribution function to obtain the quantiles corresponding to the grading thresholds. The iteration method can be to iterate over each grading threshold at preset percentage intervals, such as a preset percentage of 5%, or to iterate over each grading threshold within a preset percentage range, such as a preset percentage range of 15% to 25%.

[0153] For example, if the size data grading levels are defined as 4, then the number of size data grading thresholds is 3. The 3 grading thresholds are traversed and set to 20%, 50%, and 70% respectively. 20%, 50%, and 70% are respectively input as target probabilities into the inverse cumulative distribution function corresponding to the normal distribution to obtain the quantiles corresponding to the 20% grading threshold, the 50% grading threshold, and the 70% grading threshold.

[0154] S3033 groups the dimensional data of multiple components according to the quantile corresponding to the threshold of each grade.

[0155] In one embodiment of this application, the dimensional data of multiple components are grouped according to a custom grading level and the quantile corresponding to the grading threshold of each grading level. For example, a component has 4 grading levels, and the grading thresholds are 20%, 50%, and 70%. The dimensional data of the component is grouped according to the quantile corresponding to the 20% grading threshold, the 50% grading threshold, and the 70% grading threshold, resulting in four groups of dimensional data: dimensional data between 0% and 20%, 20% and 50%, 50% and 70%, and 70% and 100%, respectively.

[0156] S3034 uses a fitness function to calculate the fitness score for each combination scheme based on the grouping size data of multiple components.

[0157] In one embodiment of this application, multiple gear positions of different components are arranged and combined to determine the matching scheme of different components. For example, the first to fourth gear positions of the main swing arm, the first to fourth gear positions of the base, the first to fourth gear positions of the auxiliary swing arm, and the first to fourth gear positions of the wedge block are matched with each other to generate a matching scheme. The grouped size data of different components corresponding to the matching scheme are input into the flattening angle prediction model to obtain multiple predicted flattening angles corresponding to the matching scheme. The yield of the multiple predicted flattening angles of the matching scheme is calculated to obtain the fitness score of the matching scheme. During the gearing process, the gear positions of different components can be freely matched to increase the number of matching schemes, and are not limited to the above example.

[0158] For example, the grouping of the size data of the main swing arm A includes AX1 (first gear), AX2 (second gear), AX3 (third gear), and AX4 (fourth gear); the grouping of the size data of the base B includes BX1 (first gear), BX2 (second gear), BX3 (third gear), and BX4 (fourth gear); the grouping of the size data of the auxiliary swing arm C includes CX1 (first gear), CX2 (second gear), CX3 (third gear), and CX4 (fourth gear); and the grouping of the size data of the wedge block D includes DX1 (first gear), DX2 (second gear), DX3 (third gear), and DX4 (fourth gear). The following combinations are used as a configuration scheme: AX1, BX1, CX1, DX1, AX2, BX1, CX1, DX1, ..., AX4, BX4, CX4, DX4. This configuration scheme includes all combinations of gear positions between the main swing arm A, base B, auxiliary swing arm C, and wedge block D. Monte Carlo sampling is used to select sampled dimensional data from the dimensional data of multiple components corresponding to the configuration scheme. The sampled dimensional data is then input into the flattening angle prediction model to obtain the corresponding predicted flattening angle. After multiple samplings, all predicted flattening angles under this configuration scheme are obtained. The yield of all predicted flattening angles is calculated and used as the fitness score of this configuration scheme.

[0159] S3035, determine whether the fitness score is greater than or equal to a preset fitness score threshold. If the fitness score is greater than or equal to the preset fitness score threshold, proceed to S3036; if the fitness score is less than the preset fitness score threshold, return to S3032, reset the grading levels of the size data of multiple components and the grading threshold of each level, and re-determine the quantiles corresponding to the grading threshold of each grading level according to the inverse cumulative distribution function. Here, the fitness score is the yield of multiple predicted flattening angles corresponding to the grading scheme, and the preset fitness score threshold is a preset yield threshold. For example, the preset fitness score threshold is 94%, 95%, 96%, or other values.

[0160] S3036, determine the current combination scheme as the optimal combination scheme.

[0161] In one embodiment of this application, the optimal pairing scheme includes a tiered pairing of size data for different components corresponding to fitness scores that are less than or equal to a fitness score threshold. The tiering scheme corresponding to the optimal pairing scheme is the optimal tiering scheme.

[0162] The selection process based on a genetic algorithm selects individuals with better performance based on their fitness scores for reproduction. Accordingly, the pairing schemes with higher flattening angle yield are considered the optimal pairing schemes. The crossover process based on a genetic algorithm generates offspring by exchanging genes between parent individuals. Different pairing schemes are obtained by adjusting the size data of different components, and the flattening angle yield corresponding to each pairing scheme is calculated. The mutation process based on a genetic algorithm randomly changes certain genes in individuals to increase genetic diversity. Different grading schemes and pairing schemes are obtained by adjusting the size data grading of each component and its corresponding grading threshold, and the flattening angle yield corresponding to each pairing scheme is calculated. The iterative process based on a genetic algorithm repeats the selection, crossover, and mutation processes. Each generation of the population is evaluated according to the fitness function until a preset number of iterations is reached or the fitness score stabilizes. Accordingly, the grading levels, grading thresholds, and grading combinations of the dimensional data of multiple components are continuously adjusted to adjust the grading and combination schemes of the dimensional data of multiple components. The adaptability score, i.e., the flattening angle yield, is calculated for each combination scheme until the flattening angle yield is greater than or equal to a preset yield. The combination scheme with a flattening angle yield greater than or equal to the preset yield is then identified as the optimal combination scheme. Alternatively, when the number of adjustments to the combination scheme reaches a preset number, the combination scheme with the highest flattening angle yield among the multiple combination schemes is determined as the optimal combination scheme.

[0163] The above embodiments of this application use genetic algorithms to automatically optimize the size data grading scheme and matching scheme, which improves the efficiency and intelligence of tolerance grading of component size data of the rotating shaft assembly.

[0164] In one embodiment of this application, the size classification method further includes: automatically assembling the pivot assembly according to a matching scheme (e.g., the optimal matching scheme) of the size data of a plurality of components.

[0165] This application also provides a computer storage medium storing computer instructions. When the computer instructions are executed on a production device, the production device performs the aforementioned related method steps to implement the component size classification method in the above embodiments.

[0166] See Figure 20As shown in the illustration, this application also provides a production equipment 300, including: an infeed component 301, a measuring component 302, a processing component 303, a sorting component 304, an outfeed component 305, and an assembly component 306. The processing component 303 is used to execute the above-described component size classification method to determine the optimal matching scheme of the size data of multiple components, that is, the matching scheme of a preset assembly. The infeed component 301 includes, but is not limited to, a carrying device and a conveyor belt, for receiving multiple components to be assembled into a preset assembly and conveying the multiple components to the measuring component 302. The measuring component 302 is communicatively connected to the processing component 303, for measuring the size data of the multiple components and sending the measured size data of the multiple components to the processing component 303. The sorting component 304 may be a robotic arm, for matching multiple components according to the matching scheme of the preset assembly and conveying the matched multiple components to the outfeed component 305. The outfeed component 305 is used to output the matched multiple components. The outfeed component 305 may be a conveyor belt. The assembly component 306 is used to assemble the matched multiple components into a preset assembly. In other embodiments of this application, the production equipment 300 may also exclude the measuring component 302, and the dimensional data of multiple components may be measured in advance.

[0167] See Figure 21 The diagram illustrates an application scenario of the component size classification method provided in an embodiment of this application. Multiple components of different sizes are input into a production equipment 300. The production equipment 300 can apply the component size classification method provided in this embodiment to determine a matching scheme for the multiple size data of the components, match the components according to the determined matching scheme, and output the matched components. For example, components A1 to An are first components of different sizes, components B1 to Bn are second components of different sizes, components C1 to Cn are third components of different sizes, and components D1 to Dn are fourth components of different sizes. Inputting multiple components of different sizes into the production equipment 300, the production equipment 300 outputs the matched components, which include a first component A1 of a specified size, a second component B2 of a specified size, a third component C3 of a specified size, and a third component D4 of a specified size.

[0168] This application also provides a computer program product that, when run on a computer, causes the computer to perform the aforementioned steps to implement the component size classification method described in the above embodiments.

[0169] In addition, embodiments of this application also provide an apparatus, which may specifically be a chip, component or module. The apparatus may include a connected processor and a memory; wherein the memory is used to store computer execution instructions, and when the apparatus is running, the processor may execute the computer execution instructions stored in the memory to cause the chip to execute the component size classification method in the above method embodiments.

[0170] In this embodiment, the production equipment, computer storage medium, computer program product or chip are all used to execute the corresponding methods provided above. Therefore, the beneficial effects that can be achieved can be referred to the beneficial effects of the corresponding methods provided above, and will not be repeated here.

[0171] Through the above description of the embodiments, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0172] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0173] The units described as separate components may or may not be physically separate. A component shown as a unit can be one or more physical units; that is, it can be located in one place or distributed in multiple different locations. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0174] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0175] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, essentially or in other words, the parts that contribute to the prior art, or all or part of the technical solutions, can be embodied in the form of a software product. The software product is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0176] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit it. Although this application has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of this application without departing from the spirit and scope of the technical solutions of this application.

Claims

1. A component size grading method, applied to production equipment, characterized in that, The method includes: Obtain the force information of multiple components of the rotating shaft assembly in the flattened state; Based on the force information of the multiple components of the rotating shaft assembly in the flattened state, sampled data corresponding to the size data of the multiple components are obtained; Regression analysis was performed on the sampled data corresponding to the size data of the multiple components to obtain multiple features that affect the flattening angle of the rotating shaft assembly; A flattening angle prediction model is established based on the aforementioned features and flattening angle data; Obtain multiple dimensional data of multiple components in the rotating shaft assembly; Based on the multiple size data, multiple grading schemes for the multiple components are determined; The combination scheme of the multiple components is determined based on the multiple grading schemes; Based on the flattening angle prediction model, a fitness function is used to calculate the fitness score of each combination scheme. The fitness function is the formula for calculating the yield of all flattening angle prediction values ​​under the combination scheme. Based on the fitness score of each combination scheme, the combination scheme of the preset components is determined.

2. The component size grading method as described in claim 1, characterized in that, The step of determining multiple grading schemes for the multiple components based on the multiple size data includes: Set the grading levels for multiple dimensions of the multiple components and the grading threshold for each level; The grading levels of the multiple size data of the multiple components and the grading threshold of each level are determined as multiple grading schemes for the multiple components.

3. The component size grading method as described in claim 2, characterized in that, The step of determining the combination scheme of the multiple components according to the multiple grading schemes includes: The quantiles corresponding to each threshold of the multiple size data are determined based on the inverse cumulative distribution function. The multiple size data are grouped according to the quantile corresponding to each grading threshold; By combining the group size data of different components, a combination scheme for the multiple components is obtained.

4. The component size grading method as described in claim 3, characterized in that, The step of determining the quantiles corresponding to each threshold of the multiple size data of the multiple elements based on the inverse cumulative distribution function includes: The inverse cumulative distribution function of the normal distribution is used to calculate the size data corresponding to each threshold, and the calculated size data corresponding to each threshold is used as the quantile.

5. The component size grading method as described in claim 1, characterized in that, The fitness score for each combination scheme is calculated using a fitness function based on the flattening angle prediction model, including: Based on the flattening angle prediction model, the fitness function is used to calculate the fitness score of each combination scheme according to the group size data of different components.

6. The component size grading method as described in claim 5, characterized in that, The step of determining the combination scheme of preset components based on the fitness score of each combination scheme includes: If the fitness score of any combination is greater than or equal to a preset fitness score threshold, then that combination is determined as a combination of the preset components; or The combination scheme with the highest fitness score among multiple combination schemes is determined as the combination scheme of the preset components.

7. The component size grading method as described in claim 5, characterized in that, The fitness function, based on the flattening angle prediction model and the grouping size data of different components, is used to calculate the fitness score of each combination scheme, including: The group size data of different components corresponding to each combination scheme are input into the flattening angle prediction model to obtain multiple predicted flattening angles for each combination scheme. Calculate the yield of multiple predicted flattening angles for each of the above combinations to obtain the fitness score for each combination.

8. The component size grading method as described in claim 1, characterized in that, The regression analysis performed on the sampled data corresponding to the dimensional data of the multiple components yields several features affecting the flattening angle of the rotating shaft assembly, including: The sampled data is standardized. The sampled data is divided into a training set, and the training set is used to train a fitted Lasso regression model. Based on the trained Lasso regression model, obtain the coefficients of all sampled data; The absolute value of the coefficient is used as the contribution of the component's size data to the flattening angle. The component's size data is sorted in descending order of the contribution. Multiple size data that are ranked first are selected according to a preset percentage to obtain the multiple features.

9. The component size grading method as described in claim 1, characterized in that, The step of establishing a flattening angle prediction model based on the multiple features and flattening angle data includes: The preset deep learning model is trained based on the multiple features and the flattening angle data to obtain the flattening angle prediction model.

10. The component size grading method as described in claim 9, characterized in that, The multiple components include a main swing arm, a base, a secondary swing arm, and a wedge block. The dimensional data of the main swing arm includes the positional tolerance of the mating hole between the main swing arm and the wedge block, the positional tolerance of the stop surface between the main swing arm and the base, and the positional tolerance of the mating arc surface between the main swing arm and the base. The dimensional data of the base includes the positional tolerance of the mating hole pin between the base and the secondary swing arm, the positional tolerance of the stop surface between the base and the secondary swing arm, the positional tolerance of the mating arc surface between the base and the main swing arm, and the positional tolerance of the stop surface between the base and the main swing arm. The dimensional data of the auxiliary swing arm includes the positional tolerance of the mating surface between the auxiliary swing arm and the wedge block, the positional tolerance of the stop surface between the auxiliary swing arm and the base, and the positional tolerance of the mating hole between the auxiliary swing arm and the base. The dimensional data of the key structural feature surfaces of the wedge block include the positional tolerance of the mating hole between the wedge block and the main swing arm, and the positional tolerance of the mating surface between the wedge block and the auxiliary swing arm.

11. The component size grading method as described in claim 10, characterized in that, The step of training a preset deep learning model based on the multiple features and the flattening angle data to obtain the flattening angle prediction model includes: A training sample set is established based on the size data of the main swing arm, the base, the auxiliary swing arm, and the wedge block, and the flattening angle data; The size data and corresponding flattening angle data of a set of the main swing arm, the base, the secondary swing arm and the wedge block in the training sample set are used as training data and input into the preset deep learning model to calculate the loss function value of the preset deep learning model. If the loss function value of the preset deep learning model is greater than the preset value, the parameters of the preset deep learning model are adjusted, and the preset deep learning model is trained again until the loss function value of the preset deep learning model is less than or equal to the preset value. The preset deep learning model is then determined to be trained and the trained preset deep learning model is determined as the flattening angle prediction model.

12. A component size grading method, applied to production equipment, characterized in that, The method includes: Obtain the force information of multiple components of the rotating shaft assembly in the flattened state; Based on the force information of the multiple components of the rotating shaft assembly in the flattened state, sampled data corresponding to the size data of the multiple components are obtained; Regression analysis was performed on the sampled data corresponding to the size data of the multiple components to obtain multiple features that affect the flattening angle of the rotating shaft assembly; A flattening angle prediction model is established based on the aforementioned features and flattening angle data; Obtain multiple dimensional data of multiple first elements and multiple second elements in the rotating shaft assembly; Based on the multiple size data, multiple grading schemes are determined for the multiple first elements and the multiple second elements; The combination scheme of the plurality of first elements and the plurality of second elements is determined according to the plurality of grading schemes of the plurality of first elements and the plurality of second elements; Based on the flattening angle prediction model, a fitness function is used to calculate the fitness score of each combination scheme. The fitness function is the formula for calculating the yield of all flattening angle prediction values ​​under the combination scheme. Based on the fitness score of each of the aforementioned combinations, a plurality of combinations of the rotating shaft components are determined, wherein each of the rotating shaft components includes a first element and a second element.

13. The component size grading method as described in claim 12, characterized in that, The step of determining multiple grading schemes for the multiple first elements and the multiple second elements based on the multiple size data includes: The multiple size data of the plurality of first elements and the plurality of second elements are set into multiple size ranges and the threshold of each size range; The grading levels and grading thresholds of the multiple size data of the multiple first elements and the multiple second elements are determined as multiple grading schemes for the multiple first elements and the multiple second elements.

14. The component size grading method as described in claim 13, characterized in that, The step of determining the combination scheme of the plurality of first elements and the plurality of second elements according to the plurality of grading schemes includes: The quantiles corresponding to each threshold of the multiple size data of the multiple first elements and the multiple second elements are determined according to the inverse cumulative distribution function; The multiple size data of the multiple first elements and the multiple second elements are grouped according to the quantile corresponding to each grading threshold; The grouped size data of the plurality of first elements and the plurality of second elements are matched to obtain a matching scheme of multiple size data of the plurality of first elements and the plurality of second elements.

15. The component size grading method as described in claim 14, characterized in that, The fitness score for each combination scheme is calculated using a fitness function based on the flattening angle prediction model, including: Based on the flattening angle prediction model, the fitness function is used to calculate the fitness score of each combination scheme according to the grouping size data of the plurality of first elements and the plurality of second elements.

16. The component size grading method as described in claim 15, characterized in that, The fitness function, based on the flattening angle prediction model and the grouping size data of the plurality of first elements and the plurality of second elements, is used to calculate the fitness score of each combination scheme, including: The grouped size data of the plurality of first elements and the plurality of second elements corresponding to each combination scheme are input into the flattening angle prediction model to obtain multiple predicted flattening angles for each combination scheme; Calculate the yield of multiple predicted flattening angles for each of the above combinations to obtain the fitness score for each combination.

17. A production equipment, characterized in that, The production equipment includes: The feeding unit is used to receive multiple components; A processing unit is used to perform the component size classification method as described in any one of claims 1 to 11 to determine the matching scheme of the shaft assembly; The sorting component is used to match the plurality of components according to the matching scheme of the rotating shaft assembly; The discharge component is used to output the assembled plurality of components.

18. The production equipment as described in claim 17, characterized in that, The production equipment also includes: A measuring component is used to measure multiple dimensional data of the plurality of components and send the measured multiple dimensional data of the plurality of components to the processing component.

19. The production equipment as described in claim 17, characterized in that, The production equipment also includes: Assembly components are used to assemble the assembled plurality of elements into the shaft assembly.

20. A production equipment, characterized in that, The production equipment includes: A feeding component for receiving multiple first elements and multiple second elements; A processing unit for performing the component size grading method as described in any one of claims 12 to 16 to determine a matching scheme for multiple shaft assemblies; The sorting component is used to match the plurality of first elements and the plurality of second elements according to the matching scheme of the plurality of rotating shaft assemblies; The discharge component is used to output the assembled plurality of first elements and the plurality of second elements.

Images