A method and system for predicting inter-layer assembly gap based on BP neural network

Abstract

The present invention provides a method and system for predicting inter-layer assembly gaps based on a BP neural network, and relates to the technical field of quality control of layered product assembly processes. The method uses a random error matrix to produce a random error surface, and an ideal surface is superimposed on the random error surface to form a simulated product surface, and a simulated product model is established using the simulated product surface. A product model library is established, and the model library also contains a simulated product three-dimensional model and a measured product three-dimensional model. Feature points are extracted from the three-dimensional point cloud of the product surface, and used together with assembly process parameters as the input of the prediction model, and the assembly gap is used as the output of the prediction model. The assembly gap data comes from simulation and experiment. The assembly gap is measured by making the contact surface of the product after assembly equivalent to the exposed surface.

Description

Technical Field

[0001] The present invention relates to the technical field of quality control of layered product assembly process, and in particular to a method and system for predicting inter-layer assembly gap based on BP neural network. Background Art

[0002] Complex electromechanical products represented by radar array antennas have the characteristics of strong electromechanical coupling. Such products are usually based on layered parts, and various electrical connectors are used between layers for electrical signal transmission. The gap between assembly layers, as the decisive factor for the quality of electrical signal transmission of such products, is mainly affected by the processing accuracy of layered parts. At present, such products are gradually developing in the direction of high integration and complex surface. In order to meet the requirements of the design, the layered parts are partially produced by composite material additive manufacturing process. This production process can produce products with complex surfaces, but its surface manufacturing accuracy is poor. The actual product often has a large error with the ideal surface, which leads to a large and uneven gap between layers after assembly. In the current assembly method, the control of the assembly gap between layers of such products adopts post-detection means, resulting in a low yield rate. At the same time, it is impossible to subjectively grasp the assembly quality of the product, which is easy to cause great losses to the enterprise.

[0003] The existing assembly quality prediction method predicts the product processing surface through theoretical calculation and simulation, and then predicts the assembly clearance based on the predicted product surface.

[0004] A Chinese invention patent with publication number CN116165967A discloses a method for predicting and repairing the surface assembly quality. The method uses machine tool geometric error and thermal error to predict the machining surface error, and combines the calculated assembly clearance error to predict the surface assembly quality of the part. When the surface assembly quality does not meet the preset assembly quality requirements, the assembly error is compensated by repairing the corresponding assembly surface.

[0005] A Chinese invention patent with publication number CN112765729A discloses a method for controlling the gap in the assembly of a passenger car bumper mask. The method uses simulation software to simulate the parts production process to optimize the parts production process parameters, control product deformation, simulate the assembly process, and predict and evaluate whether the gap after assembly meets the design tolerance. If not, it is optimized through structural and process parameters, and finally a product that meets the tolerance requirements is obtained.

[0006] The surface parameters used in the above methods are all derived from theoretical calculations and predictions of the processing process. The surface parameters obtained in this way often have certain errors with the actual product surface, which will greatly affect the accuracy of the assembly clearance prediction results. In addition, the above methods mainly change the product assembly results by adjusting the product processing links, and cannot control the product assembly clearance through the assembly link. Summary of the invention

[0007] Purpose of the invention: A method and system for predicting inter-layer assembly gap based on BP neural network is proposed to solve the above problems existing in the prior art.

[0008] To achieve the above object, the first aspect of the present invention provides a method for predicting inter-layer assembly gap based on BP neural network, comprising the following steps:

[0009] S1. Based on the ideal product surface model, simulate the product manufacturing error in the actual processing process, establish a product model library, simulate the assembly process of the models in the product model library under different process parameter conditions, and obtain assembly clearance data;

[0010] S2, setting model parameters, and taking the simulated shape and assembly process parameters as input, and the assembly gap obtained by simulation as output, to construct and train a layer gap prediction model based on BP neural network;

[0011] S3, parameterizing the physical product surface based on 3D point cloud scanning, inputting the measured product surface and assembly process parameters into the interlayer gap prediction model, outputting the prediction results, and selecting the optimal assembly gap to determine the assembly process parameters;

[0012] S4. Compare the measured results of the assembly clearance with the predicted results, optimize the model parameters, and add the measured product model made according to the measured surface parameters and the measured assembly clearance of the assembled product to the product model library.

[0013] In a further embodiment of the first aspect, step S1 further comprises:

[0014] S11. Determine the number of control point array rows m and columns n according to the complexity of the product surface;

[0015] S12, generate the surface error matrix A = {a ij} m×n , according to the part surface error range in actual production, add the same scaling factor k to the random array;

[0016] In the formula, a ij Indicates XXX; a ij ∈[0,1];

[0017] S13, establish an ideal shape surface space coordinate system, and express the ideal shape surface control point matrix space coordinates as:

[0018] X={x ij} m×n ,Y={y ij} m×n ,Z={z ij} n×n ;

[0019] The coordinates of the control points of the simulated error surface are expressed as:

[0020] X={x ij} m×n ,Y={y ij} m×n ,A={a ij} m×n ;

[0021] Use the above control points to generate the simulated error surface P w ;

[0022] S14, simulate the error shape surface and superimpose the ideal shape surface to obtain the simulated product shape surface P m , repeating step S11 to step S12 to obtain a number of simulated product surfaces, drawing the simulated product surfaces into simulated product models in the three-dimensional modeling software, storing and managing the simulated product models with unified numbers, and establishing a product model library;

[0023] S15. Design several assembly processes. The assembly process parameters include product screw connection method M, interlayer connection sequence C, connection force F m , loading percentage E, forming assembly process parameter vector group B = {M, C, F m ,E};

[0024] S16. Use finite element analysis software to simulate the assembly process of all products in the product library model under different process parameter conditions to obtain the corresponding assembly layer gap Q f , add the assembly clearance information to the product model library accordingly.

[0025] In a further embodiment of the first aspect, the process of constructing and training an interlayer gap prediction model based on a BP neural network further includes:

[0026] S21, extracting features from the point cloud data of the preset product surface model; extracting the product surface parameters, assembly process parameters, assembly layer gap Q f Perform normalization processing and use them as input and output data of the inter-layer gap prediction model respectively; establish data samples, use 80% of the data samples as training sets and 20% of the data samples as test sets;

[0027] S22, build an interlayer gap prediction model, the interlayer gap prediction model includes an input layer, a hidden layer and an output layer, the input layer is composed of preset product surface parameters and preset assembly process parameters, including N x neurons, and the output layer contains N y neurons, i.e., the gap between the assembly layers; the number of hidden layers is initially set to one layer, including N h neurons;

[0028] S23, initialize the neural network parameters, input the product surface parameters and assembly process parameters into the neural network prediction model, and output the predicted value of the gap between the assembly layers after forward propagation; set the error threshold, if the error between the predicted assembly layer gap value and the actual assembly layer gap value does not meet the requirements, then back propagate the error in the neural network.

[0029] In a further embodiment of the first aspect, the connection force F m The calculation formula is as follows:

[0030]

[0031] Where P is the screw pitch, μ g is the friction coefficient of the thread pair, d2 is the screw diameter, D km is the surface size of the screw head, μ k is the friction coefficient of the screw head surface.

[0032] In a further embodiment of the first aspect, the number of neurons N included in the hidden layer is h The calculation formula is as follows:

[0033]

[0034] Where N x is the number of neurons in the input layer, N y is the number of neurons in the output layer, N s is the number of samples and c is a constant.

[0035] In a further embodiment of the first aspect, in the process of training the inter-layer gap prediction model based on the BP neural network, a ReLU activation function is used, and the He initialization method is used for the ReLU activation function to initialize the weight and bias parameters, and the initial learning rate is 1e-3;

[0036] The Adam optimization method is used to adjust the learning rate for model optimization. The formula for updating the weight and bias parameter θ is as follows:

[0037]

[0038] In the formula, m t and v t are the estimates of the first and second moments of the corrected gradient, η is the learning rate, and ∈ is a parameter to prevent division by zero.

[0039] In a further embodiment of the first aspect, step S3 further comprises:

[0040] S31. Use a 3D point cloud camera to take pictures of the product in different areas to obtain surface point cloud data, cut and splice the point cloud data to obtain complete product surface point cloud data; import the processed point cloud data into the point cloud surface fitting software to produce the measured surface P c ;

[0041] S34, the product measured surface P c And the preset assembly process parameter B is input into the prediction model to obtain the assembly layer gap prediction result Q y ;

[0042] S35. Design the following assembly quality judgment criteria: Maximum gap Max{q i ,1≤i≤k} does not exceed the design allowable value q max , and the assembly clearance space q total Less than the preset value;

[0043] Assembly clearance space calculation formula q total =∫∫(P c1 -P c2 )dxdy;

[0044] Where P c1 is the measured surface of the upper part, P c2 Measure the surface of the lower part;

[0045] S36. Compare the prediction results, select the optimal result according to the assembly quality judgment criteria and determine the assembly process parameters.

[0046] In a further embodiment of the first aspect, step S4 further comprises:

[0047] S41. Draw the measured surface solid model based on the measured product surface in the 3D modeling software, and use the equivalent idea to measure the assembly gap Q of the actual product. c , the measured surface parameters are drawn into the measured product entity model in the 3D modeling software, the measured product model and the corresponding measured assembly gap are added to the product model library, the training data set is enriched, and the inter-layer gap prediction model is continued to be trained to achieve model iteration;

[0048] Mounting product assembly clearance Q c The calculation formula is as follows:

[0049] Q c =P cd1 -P cd2 -(l1+l2);

[0050] Where P cd1 is the equivalent surface of the upper part, P cd2 is the equivalent surface of the lower part, l1 and l2 are the thickness of the upper and lower parts respectively;

[0051] S42. Compare the prediction results of the gap between assembly layers Q y Assembly clearance Q with installed product c , modify the prediction algorithm parameters and optimize the assembly inter-layer gap prediction model.

[0052] According to a second aspect of the present invention, a layer-to-layer assembly gap prediction system is provided. The prediction system includes an assembly gap data acquisition unit, a prediction model construction unit, a result output unit, and an optimization unit.

[0053] The assembly clearance data acquisition unit is used to simulate the product manufacturing error of the actual processing process according to the ideal product surface model, establish a product model library, simulate the assembly process of the models in the product model library under different process parameter conditions, and obtain the assembly clearance data;

[0054] The prediction model building unit is used to set model parameters, and takes the simulated shape and assembly process parameters as input, and the assembly gap obtained by simulation as output, and builds and trains the interlayer gap prediction model based on BP neural network;

[0055] The result output unit parameterizes the physical product surface based on three-dimensional point cloud scanning, inputs the measured product surface and assembly process parameters into the interlayer gap prediction model, outputs the prediction result, and selects the optimal assembly gap to determine the assembly process parameters;

[0056] The optimization unit is used to compare the measured results and predicted results of the assembly clearance, optimize the model parameters, and add the measured surface parameters and the measured assembly clearance of the assembled product to the product model library.

[0057] Compared with the prior art, the present invention has the following significant advantages: by inputting different assembly process parameters and measured product surface parameters, the gap between the assembly layers of layered products can be effectively predicted. The main purpose is to establish a product model library, add simulated product models and measured product models to the product model library, enrich the prediction model training data set, and improve the prediction model training effect. In addition, by using back propagation, the connection weights between neurons in each layer can be adjusted layer by layer, so that the error is continuously reduced until it is less than the set error threshold. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a schematic diagram of the process of the present invention.

[0059] Figure 2 It is a model structure diagram of the BP neural network of the present invention.

[0060] Figure 3 It is a flowchart for training BP neural network. DETAILED DESCRIPTION

[0061] In the following description, a large number of specific details are provided to provide a more thorough understanding of the present invention. However, it is apparent to those skilled in the art that the present invention can be implemented without one or more of these details. In other examples, in order to avoid confusion with the present invention, some technical features well known in the art are not described.

[0062] In order to solve the problem of predicting the distribution of assembly gaps between layers of complex-surface products mentioned in the background of the invention, the present invention provides a method for predicting assembly gaps between layers based on a BP neural network. A random error surface is produced using a random error matrix. An ideal surface is superimposed with the random error surface to form a simulated product surface. A simulated product model is established using the simulated product surface. A product model library is established, which contains both a simulated product three-dimensional model and a measured product three-dimensional model. Feature points are extracted from the three-dimensional point cloud of the product surface and used as the prediction model input together with assembly process parameters, and the assembly gap is used as the prediction model output. The assembly gap data comes from simulation and experiment. The assembly gap is measured by treating the contact surface of the product after assembly as equivalent to an exposed surface.

[0063] Combination Figure 1 The specific technical scheme of the inter-layer assembly gap prediction method is disclosed as follows:

[0064] Step S1: Make a model learning sample: Based on the ideal product surface model, simulate the product manufacturing error of the actual processing process, establish a product model library, simulate the assembly process of the models in the product model library under different process parameter conditions, and obtain assembly clearance data.

[0065] Step S2: Model setting and training: setting model parameters, taking the simulated shape and assembly process parameters as input, and the assembly gap obtained by simulation as output, and training the interlayer gap prediction model based on BP neural network.

[0066] Step S3: Assembly gap prediction: parameterize the physical product surface based on 3D point cloud scanning. Input the measured product surface and assembly process parameters into the prediction model to obtain the prediction result, and select the optimal assembly gap to determine the assembly process parameters.

[0067] Step S4: Model optimization: Compare the measured results of the assembly clearance with the predicted results, and optimize the model parameters. The measured product model made according to the measured surface parameters and the measured assembly clearance of the assembled product are added to the product model library to enrich the prediction model learning samples and improve the model prediction accuracy.

[0068] As a preferred embodiment, in order to study the changes in product assembly gaps under various process parameters under different manufacturing error conditions, a more comprehensive and rich sample is provided for the training of the model in step S2. According to the variation range of each process parameter in the actual assembly, multiple sets of process parameter combinations are preset. Several error surfaces made based on random arrays are superimposed on the ideal surface to form a simulated product surface. The finite element analysis software is used to perform assembly simulation of the simulated product surface under different process parameters to obtain the distribution of the assembly gap surface. The step S1 specifically includes:

[0069] Step S11: To better parametrically express the product surface P, the number of control point array rows m and columns n is determined according to the complexity of the product surface.

[0070] Step S12: Generate a shape error matrix A = {a ij} m×n ,a ij ∈[0,1], according to the part surface error range in actual production, add the same scaling factor k to the random array.

[0071] When making error surfaces, the error matrix is ​​subjectively controlled based on the product manufacturing process and engineering experience, so that the error matrix can simulate the possible manufacturing errors of the product to the greatest extent.

[0072] Step S13: Establish an ideal surface space coordinate system, and express the ideal surface control point space coordinates as X = {x ij} m×n ,Y={y ij} m×n ,Z={z ij} m×n , then the coordinates of the control points of the simulated error surface can be expressed as X = {x ij} m×n ,Y={y ij} m×n ,A={a ij} m×n , use the above control points to generate the simulated error surface P w .

[0073] Step S14: Simulate the error surface and superimpose the ideal surface to obtain the simulated product surface P m Repeat steps S11 to S12 to obtain several simulated product surfaces, draw the simulated product surfaces into simulated product models using the surface thickening function in the 3D modeling software, and uniformly number, store and manage the simulated product models and the measured product models obtained in step S41 to establish a product model library.

[0074] Step S15: Design several assembly processes, and the specific process parameters include the product screw connection method M, which depends on the type of screw or bolt, nominal diameter, number of threads, and screw material properties. Interlayer connection sequence C, connection force F m , loading percentage E={e1,e2,…e k}, where k is the number of loading times. Form an assembly process parameter vector group B = {M, C, F m ,E}.

[0075] Connection force calculation formula:

[0076]

[0077] Where P is the screw pitch, μ g is the friction coefficient of the thread pair, d2 is the screw diameter, D km is the surface size of the screw head, μ k is the friction coefficient of the screw head surface.

[0078] Step S16: Use finite element analysis software to simulate the assembly process of all products in the product library model under different process parameter conditions to obtain the corresponding assembly layer gap Q f , add the assembly clearance information to the product model library accordingly.

[0079] As a preferred embodiment, in order to predict the assembly gap after the product is actually assembled, a product assembly gap prediction neural network is constructed to obtain the mapping relationship between the assembly process parameters and assembly errors and the product assembly gap surface distribution. The step S2 specifically includes:

[0080] Step S21: Establish a data set, and extract features from the preset product surface model point cloud data using a point-based feature extraction method to train the neural network. The product surface parameters and assembly process parameters obtained in step S1, and the gap between assembly layers obtained by assembly simulation are normalized and used as input and output data of the model to establish data samples, 80% of the data samples are used as training sets, and 20% of the data samples are used as test sets. In step S23, the ratio of the test set to the training set can be adjusted, and the network model parameters can be adjusted at the same time to perform multiple trainings to make the network have stronger generalization capabilities.

[0081] Step S22: Model building. The prediction model based on BP neural network includes an input layer, a hidden layer and an output layer. The input layer is composed of preset product surface parameters and preset assembly process parameters, including N x neurons, x = m × n + k + 3, the output layer contains N y Contains neurons, that is, the gap between assembly layers, y is the number of high-value points. The number of hidden layers is initially set to one layer, including the number of neurons Nh Use the empirical formula to make an initial estimate:

[0082]

[0083] Where N x is the number of neurons in the input layer, N y is the number of neurons in the output layer, N s is the number of samples, c is a constant, and is initially set to 5. The ReLU activation function is used, and the He initialization method is used to initialize the weight and bias parameters for the ReLU activation function. The initial learning rate is 1e-3. Considering that the point cloud data is relatively large and concentrated, the Adam optimization method is used to adjust the learning rate for model optimization. The formula for updating the weight and bias parameters θ is as follows:

[0084]

[0085] Among them, m t and v t They are respectively the estimates of the first-order moment (mean) and the second-order moment (center-free variance) of the corrected gradient, η is the learning rate, ∈ is a parameter to prevent division by 0, usually 1e -8 .

[0086] Step S23: Model training: Initialize the neural network parameters, input the product surface parameters and assembly process parameters into the neural network prediction model, and output the predicted value of the gap between the assembly layers after forward propagation; set the error threshold, if the error between the predicted assembly layer gap value and the actual assembly layer gap value does not meet the requirements, then back propagate the error in the neural network, in this process, adjust the connection weights and offsets between the neurons in each layer layer by layer, so that the error is continuously reduced, that is, the predicted value is continuously close to the actual value, and repeat this process until the training reaches the set error threshold.

[0087] As a preferred embodiment, in order to study and verify the accuracy of the prediction model and provide measured samples for step S4 to optimize the model, the three-dimensional point cloud data of the product parts is obtained based on the three-dimensional point cloud camera, and the feature points m×n parameterized expression of the part surface are extracted, and the prediction model trained in step S2 is input to obtain the prediction results, and the assembly gap is compared to determine the optimal assembly process, and the product is actually installed and the interlayer gap is measured. The step S3 specifically includes.

[0088] Step S31: Use a 3D point cloud camera to take pictures of the product in different areas to obtain surface point cloud data, and cut and splice the point cloud data to obtain complete product surface point cloud data. Import the processed point cloud data into the point cloud surface fitting software to produce the measured surface P c .

[0089] Step S34: The product measured surface P cAnd the preset assembly process parameter B is input into the prediction model to obtain the assembly layer gap prediction result Q y .

[0090] Step S35: Design assembly quality judgment criteria: (i) Maximum gap Max{q i ,1≤i≤k} does not exceed the design allowable value q max ; (ii) Assembly clearance space q total Relatively small.

[0091] Assembly clearance space calculation formula q total =∫∫(P c1 -P c2 )dxdy

[0092] Where P c1 is the measured surface of the upper part, P c2 Measure the shape and surface of the lower layer parts.

[0093] Step S36: Compare the prediction results, select the best result according to the assembly quality judgment criteria and determine the assembly process parameters.

[0094] As a preferred embodiment, in order to enrich the prediction model learning samples and improve the model prediction accuracy, the measured product model and the measured assembly interlayer gap are prepared as new learning samples of the prediction model. At the same time, the measured assembly interlayer gap and the predicted assembly interlayer gap are compared to adjust the prediction algorithm parameters. The step S4 specifically includes:

[0095] Step S41: Draw a solid model of the measured surface based on the measured product surface in the 3D modeling software, and measure the assembly gap Q of the actual product using the equivalent idea c , the measured surface parameters are drawn as the measured product solid model in the 3D modeling software, the measured product model and its corresponding measured assembly gap are added to the product model library, the training data set is enriched, and the inter-layer gap prediction model is continued to be trained to realize model iteration.

[0096] Among them, since the products have the same thickness, the assembly gap of the installed products can be measured by equating the product contact surface to the exposed surface.

[0097] The calculation formula for the assembly clearance of the installed product is: Q c =P cd1 -P cd2 -(l1+l2)

[0098] Among them, P cd1 is the equivalent surface of the upper part, P cd2 is the equivalent surface of the lower part, l1 and l2 are the thickness of the upper and lower parts respectively.

[0099] Step S42: Compare Qy and Q c , modify the prediction algorithm parameters and optimize the assembly inter-layer gap prediction model.

[0100] As a preferred embodiment, a layer-to-layer assembly gap prediction system is proposed, which can be used to run the layer-to-layer assembly gap prediction method based on BP neural network disclosed in the above embodiment. The prediction system includes an assembly gap data acquisition unit, a prediction model construction unit, a result output unit, and an optimization unit.

[0101] The assembly clearance data acquisition unit is used to simulate the product manufacturing error of the actual processing process according to the ideal product surface model, establish a product model library, simulate the assembly process of the models in the product model library under different process parameter conditions, and obtain the assembly clearance data.

[0102] The prediction model building unit is used to set model parameters, and takes the simulated shape and assembly process parameters as input, and the simulated assembly gap as output, to build and train an interlayer gap prediction model based on the BP neural network.

[0103] The result output unit parameterizes the physical product surface based on three-dimensional point cloud scanning, inputs the measured product surface and assembly process parameters into the interlayer gap prediction model, outputs the prediction result, and selects the optimal assembly gap to determine the assembly process parameters.

[0104] The optimization unit is used to compare the measured results and predicted results of the assembly clearance, optimize the model parameters, and add the measured surface parameters and the measured assembly clearance of the assembled product to the product model library.

[0105] The above-mentioned embodiment discloses a method for predicting the interlayer assembly gap based on the BP neural network. In the actual application process, its operation logic can be written into a set of executable instructions, which are written into a storage medium and run on an electronic device. More specific examples of the computer-readable storage medium mentioned in this embodiment may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. The computer-readable storage medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The readable signal medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program used by or in combination with an instruction execution system, an apparatus, or a device. When the electronic device is running, the complete inter-layer assembly gap prediction method based on BP neural network disclosed in any of the above embodiments can be automatically executed, and the specific method will not be described in detail here.

[0106] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for predicting inter-layer assembly gap based on BP neural network, characterized in that: The steps include: S1. Based on the ideal product surface model, simulate the product manufacturing error of the actual processing process, establish a product model library, simulate the assembly process of the models in the simulated product model library under different process parameter conditions, and obtain assembly clearance data; S11. Determine the number of control point array rows m and columns n according to the complexity of the product surface; S12. Generate surface error matrix , according to the part surface error range in actual production, add the same scaling factor k to the random array; In the formula, ; S13, establish an ideal shape surface space coordinate system, and express the ideal shape surface control point matrix space coordinates as: ； The coordinates of the control points of the simulated error surface are expressed as: ； Generate the simulated error surface using the above control points ; S14. Superimpose the simulated error surface on the ideal surface to obtain the simulated product surface , repeating step S11 to step S12 to obtain a number of simulated product surfaces, drawing the simulated product surfaces into simulated product models in the 3D modeling software, storing and managing the simulated product models and the measured product models with unified numbers, and establishing a product model library; S15. Design several assembly processes. The assembly process parameters include product screw connection method M, interlayer connection sequence C, connection force F m , loading percentage E, forming the assembly process parameter vector group ; S16. Use finite element analysis software to simulate the assembly process of all products in the product library model under different process parameter conditions to obtain the corresponding gaps between assembly layers. , add the assembly clearance information to the product model library; S2, setting model parameters, and taking the simulated shape and assembly process parameters as input, and the assembly gap obtained by simulation as output, to construct and train a layer gap prediction model based on BP neural network; S3. Parameterize the physical product surface based on three-dimensional point cloud scanning, input the measured product surface and assembly process parameters into the interlayer gap prediction model, output the prediction result, select the optimal assembly gap and determine the assembly process parameters.

2. The method for predicting inter-layer assembly gap based on BP neural network according to claim 1, characterized in that: Also includes: S4. Compare the measured results of the assembly clearance with the predicted results, optimize the model parameters, and add the measured product model made according to the measured surface parameters and the measured assembly clearance of the assembled product to the product model library.

3. The method for predicting inter-layer assembly gap based on BP neural network according to claim 1, characterized in that: The process of constructing and training the interlayer gap prediction model based on the BP neural network further includes: S21, extracting features from the point cloud data of the preset product surface model; extracting product surface parameters, assembly process parameters, and gaps between assembly layers Perform normalization processing and use them as input and output data of the inter-layer gap prediction model respectively; establish data samples, use 80% of the data samples as training sets and 20% of the data samples as test sets; S22, build an interlayer gap prediction model, the interlayer gap prediction model includes an input layer, a hidden layer and an output layer, the input layer is composed of preset product surface parameters and preset assembly process parameters, including neurons, and the output layer contains neurons, i.e., the gap between the assembly layers; the number of hidden layers is initially set to one layer, including neurons; S23, initialize the neural network parameters, input the product surface parameters and assembly process parameters into the neural network prediction model, and output the predicted value of the gap between the assembly layers after forward propagation; set the error threshold, if the error between the predicted assembly layer gap value and the actual assembly layer gap value does not meet the requirements, then back propagate the error in the neural network.

4. The method for predicting inter-layer assembly gap based on BP neural network according to claim 1, characterized in that: The connection force F m The calculation formula is as follows: ； Where P is the screw pitch, is the friction coefficient of thread pair, is the screw diameter, is the surface size of the screw head, is the friction coefficient of the screw head surface, and T is the screw tightening torque.

5. The method for predicting inter-layer assembly gap based on BP neural network according to claim 3 is characterized in that: The number of neurons contained in the hidden layer The calculation formula is as follows: ； In the formula, is the number of neurons in the input layer, is the number of neurons in the output layer, is the number of samples and c is a constant.

6. The method for predicting inter-layer assembly gap based on BP neural network according to claim 3 is characterized in that: In the process of training the inter-layer gap prediction model based on BP neural network, the ReLU activation function is used. The He initialization method is used to initialize the weight and bias parameters for the ReLU activation function, and the initial learning rate is 1e-3; The Adam optimization method is used to adjust the learning rate for model optimization and update the weight and bias parameters. The formula is as follows: ； In the formula, and are respectively the estimates of the first and second moments of the corrected gradient, is the learning rate, It is a parameter to prevent division by 0.

7. The method for predicting inter-layer assembly gap based on BP neural network according to claim 1, characterized in that: Step S3 further comprises: S31. Use a 3D point cloud camera to take pictures of the product in different areas to obtain surface point cloud data, and then cut and splice the point cloud data to obtain complete product surface point cloud data; import the processed point cloud data into the point cloud surface fitting software to produce the measured surface ; S34, measure the product surface And the preset assembly process parameter B is input into the prediction model to obtain the prediction result of the gap between assembly layers ; S35. Design the following assembly quality judgment criteria: Maximum clearance at high-value points Not exceeding the design allowable value , and the assembly clearance space Less than the preset value; Assembly clearance space calculation formula ; In the formula, Measure the shape of the upper part. Measure the surface of the lower part; S36. Compare the prediction results, select the optimal result according to the assembly quality judgment criteria and determine the assembly process parameters.

8. The method for predicting inter-layer assembly gap based on BP neural network according to claim 7, characterized in that: Step S4 further comprises: S41. Draw a solid model of the measured surface based on the measured product surface in the 3D modeling software, and use the equivalent idea to measure the assembly gap of the actual product , in the 3D modeling software, the measured surface parameters are plotted as the measured product model, the measured product model and the corresponding measured assembly gap are added to the product model library, the training data set is enriched, and the inter-layer gap prediction model is continued to be trained to achieve model iteration; Assembly clearance of installed products The calculation formula is as follows: ； In the formula, is the equivalent surface of the upper part, is the equivalent surface of the lower part. and are the thickness of the upper and lower parts respectively; S42. Compare the prediction results of the gap between assembly layers Assembly clearance with installed products , modify the prediction algorithm parameters and optimize the assembly inter-layer gap prediction model.

9. An inter-layer assembly gap prediction system, using the inter-layer assembly gap prediction system to run the inter-layer assembly gap prediction method based on BP neural network as described in any one of claims 1 to 8, characterized in that: The inter-layer assembly gap prediction system includes: The assembly clearance data acquisition unit is used to simulate the product manufacturing error of the actual processing process according to the ideal product surface model, establish a preset product model library, simulate the assembly process of the models in the product model library under different process parameter conditions, and obtain the assembly clearance data; The prediction model building unit is used to set model parameters, and take the simulated surface and assembly process parameters as input, and the assembly gap obtained by simulation as output, so as to build and train a layer gap prediction model based on BP neural network; A result output unit, which parameterizes the physical product surface based on three-dimensional point cloud scanning, inputs the measured product surface and assembly process parameters into the interlayer gap prediction model, outputs the prediction result, and selects the optimal assembly gap to determine the assembly process parameters; The optimization unit is used to compare the measured results and predicted results of the assembly clearance, optimize the model parameters, and add the measured surface parameters and the measured assembly clearance of the assembled product to the product model library.

Images