CN116757068B - Prediction method for CFRP self-piercing riveting forming process based on deep learning - Google Patents

Prediction method for CFRP self-piercing riveting forming process based on deep learning Download PDF

Info

Publication number
CN116757068B
CN116757068B CN202310660481.8A CN202310660481A CN116757068B CN 116757068 B CN116757068 B CN 116757068B CN 202310660481 A CN202310660481 A CN 202310660481A CN 116757068 B CN116757068 B CN 116757068B
Authority
CN
China
Prior art keywords
layer
deep learning
self
convolution
cfrp
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202310660481.8A
Other languages
Chinese (zh)
Other versions
CN116757068A (en
Inventor
刘洋
吴庆军
代祎琳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Qingdao University of Technology
Original Assignee
Qingdao University of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Qingdao University of Technology filed Critical Qingdao University of Technology
Priority to CN202310660481.8A priority Critical patent/CN116757068B/en
Publication of CN116757068A publication Critical patent/CN116757068A/en
Application granted granted Critical
Publication of CN116757068B publication Critical patent/CN116757068B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0475—Generative networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/094—Adversarial learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/10—Numerical modelling
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Medical Informatics (AREA)
  • Mathematical Physics (AREA)
  • Molecular Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Multimedia (AREA)
  • Databases & Information Systems (AREA)
  • Geometry (AREA)
  • Computer Hardware Design (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a prediction method for a CFRP self-piercing riveting forming process based on deep learning, which establishes a new deep learning framework, and can rapidly predict model results under different displacement percentages in an SPR connection forming process by inputting the stroke displacement percentages of a punch into a model, so as to obtain corresponding section shapes, namely, the model results are predicted from the input of different displacement percentages, namely, the SPR connection is adopted, the layering and upper surface damage of a carbon fiber composite material, the deformation of an aluminum alloy and the dynamic change process of the position of a rivet in the riveting process are completed, and the analysis efficiency is improved.

Description

Prediction method for CFRP self-piercing riveting forming process based on deep learning
Technical Field
The invention relates to the technical field of self-piercing riveting, in particular to a prediction method for CFRP self-piercing riveting forming process based on deep learning.
Background
As lightweight materials are increasingly used in vehicle body structures, multi-material hybrid structures present challenges to the joining technology. The light material is used for the car body to replace the traditional steel, so that the light car is one of effective ways for realizing the light weight of the car, in the car manufacture, new light materials such as aluminum alloy, titanium alloy, composite material and the like are widely applied, and carbon fiber reinforced composite material (CFRP, carbon Fiber Reinforced Polymer) is applied to new energy automobiles and high-end automobiles because of high specific strength and specific rigidity. In the design of a multi-material hybrid vehicle body structure, CFRP needs to be connected with a metal frame to form a complete vehicle body structure, and challenges are presented to the connection technology of the CFRP and the metal. SPR (self-piercing riveting) is a novel mechanical joining technique that allows for the efficient joining of CFRP and metal without the need for pre-forming holes. SPR has advantages of high connection efficiency, low cost, good reliability, and the like, and the joint has high strength and good sealing property.
Self-piercing riveting belongs to mechanical internal locking connection, and the deformation process of a base plate and a rivet cannot be known from the outside of a joint. In order to understand the forming mechanism of the joint, researchers currently mainly set the punch stroke to a specific value which is sequentially increased through a connection breaking experiment, then cut the joint from the middle part, and observe the dynamic forming process of the joint in the whole connection process by adopting a section visual detection method. However, this method requires a lot of time and economic costs, and the joint may have a large error in the cutting process. Compared with a metal material, the CFRP consists of a plurality of layers, the deformation state of each layer is inconsistent in the riveting process, and secondary damage is caused to the material by joint cutting, so that the analysis difficulty of the CFRP self-piercing riveting forming process is increased. In addition, in the process of self-piercing riveting of the composite material, damage can occur to the riveting area, such as fiber fracture, matrix cracking, delamination and the like, and the forming damage defect influences the sealing performance and the mechanical property of the joint. Therefore, there is a need to develop a method that can rapidly and accurately predict CFRP-metal self-piercing riveting forming process and damage.
Disclosure of Invention
The invention aims to design and develop a prediction method for a CFRP self-piercing riveting forming process based on deep learning, which accurately predicts the dynamic deformation process and upper surface damage of a base plate in the self-piercing riveting process of CFRP and alloy plates, shortens the prediction time and improves the accuracy.
The technical scheme provided by the invention is as follows:
a prediction method for CFRP self-piercing riveting forming process based on deep learning comprises the following steps:
step one, collecting the stroke displacement percentage of a punch;
step two, establishing a prediction model;
wherein the predictive model includes a generator and a arbiter;
the 1 st, 3 rd, 5 th, 7 th and 9 th layers of the generator are all convolution residual blocks, the 2 nd, 4 th, 6 th and 8 th layers are all transposed residual blocks, and the 10 th layer is an activation function layer;
the 1 st to 4 th layers of the discriminator consist of a convolution layer, a batch standardization layer and a rectification layer, and the 5 th layer is the convolution layer;
training the prediction model by using carbon fiber composite materials and alloy self-piercing riveting images under a plurality of different punch stroke displacement percentages to obtain a deep learning-based condition generation countermeasure network;
and fourthly, inputting random noise and punch stroke displacement percentage into the deep learning-based condition generation countermeasure network to obtain a predicted section segmentation and upper surface damage image.
Preferably, the convolution residual block includes a convolution layer, a batch normalization layer, a rectification layer, a convolution layer, and a batch normalization layer connected in sequence.
Preferably, the transposed residual block comprises a transposed convolution adjustment channel arranged in parallel, and a transposed convolution layer, a batch standardization layer and a rectification layer which are connected in sequence;
the convolution kernel size of the transposed convolution adjustment channel is 1×1, and the convolution kernel size of the transposed convolution layer is 3×3.
Preferably, the objective function of the generator is:
wherein V (G) is the least squares loss of the generator G, E c,z For the expectation when the condition information is c and the random noise is z, D (G (c, z), c) gives the arbiter D the probability that the generated data G (c, z) and the condition information is c is the true data, λ is the regularization coefficient, and F (G) is the regularization term.
Preferably, the objective function of the discriminator is:
wherein V (D) is the least squares loss of the arbiter D, E x,c For the expectation when the real data is x and the condition information is c, D (x, c) gives the probability that the real data x and the condition information c are the real data to the arbiter D.
Preferably, the third step specifically includes:
step 1, constructing an initial training set by using a plurality of carbon fiber composite materials and alloy self-piercing riveting images with different punch stroke displacement percentages;
step 2, carrying out data preprocessing on the initial training set;
step 3, performing image enhancement processing on the training set sample subjected to data preprocessing to generate a corresponding section segmentation image;
and 4, training the prediction model by taking the training set sample obtained after the corresponding section segmentation image and the data preprocessing as a data set to obtain a condition generation countermeasure network based on deep learning.
Preferably, the data preprocessing includes:
and labeling the initial sample image in the initial training set by a labeling tool, and labeling the outlines of the rivet, the lower plate, the carbon fiber composite material layering and the damaged area on the upper surface of the carbon fiber, wherein other areas are used as the background.
Preferably, the image enhancement process is:
and carrying out color adjustment, brightness adjustment, translation, blurring, clipping, sharpening, overturning, scaling or rotation of random angles on the training set samples after the data preprocessing.
Preferably, the training parameters of the prediction model include: training optimizers, learning rate, L2 regularization coefficients, batch size, and loss function.
Preferably, the alloy is an aluminum alloy.
The beneficial effects of the invention are as follows:
(1) The prediction method for the CFRP self-piercing riveting forming process based on deep learning solves the problems that in the prior art, experiments are complex, financial resources and materials are consumed, the period is long, accurate parameters are required in the process of carrying out numerical simulation analysis on complex models, a large amount of time and energy are consumed, and the like, greatly improves the measurement time of SPR measurement section shape, improves the analysis efficiency, overcomes the defects of repeated observation, manual measurement and section image labeling in the prior experiments, provides support for the follow-up improvement of efficiency and automatic measurement section geometric parameters, reduces the complexity and error of the experiments, and the network architecture is based on experimental data, does not have any assumption, so that the prediction result is closer to an actual value, and has the characteristics of simple analysis process, low cost and the like.
(2) According to the prediction method for the CFRP self-piercing riveting forming process based on deep learning, which is designed and developed by the invention, the characteristics of the connection process between the CFRP and the alloy are learned through a deep learning network, so that the cross-sectional shapes of the CFRP and the metal plate in the SPR connection forming process, namely the layering structure of the carbon fiber, the upper surface damage and the metal change process, can be rapidly predicted, the analysis efficiency is improved, and compared with the traditional finite element method, the deep learning method can avoid complex physical models and calculation processes, improve the prediction accuracy and efficiency, and provide better support for SPR connection design and optimization.
Drawings
FIG. 1 is a schematic diagram of a prediction model according to the present invention.
Fig. 2 is a schematic structural diagram of a convolution residual block according to the present invention.
Fig. 3 is a schematic structural diagram of a transposed residual block according to the present invention.
Fig. 4 is a schematic structural diagram of the discriminator according to the invention.
Fig. 5 is a schematic diagram of a training process of the prediction model according to the present invention.
Fig. 6 is a schematic structural diagram of a lower die according to the embodiment of the invention.
Fig. 7 is a schematic structural view of a rivet according to an embodiment of the present invention.
Fig. 8 is a schematic view of a finite element extracted image according to an embodiment of the present invention.
Fig. 9 is a schematic diagram of an image after data preprocessing according to an embodiment of the present invention.
FIG. 10 is a schematic representation of a predicted cross-sectional segmentation and upper surface damage image for a punch stroke displacement percentage of 19% according to the present invention.
Fig. 11 is a schematic diagram of a connecting section cloud image and an upper plate damage image obtained by finite element simulation when the stroke displacement percentage of the punch is 19%.
FIG. 12 is a schematic representation of predicted cross-sectional segmentation and upper surface damage image at a punch stroke displacement percentage of 68% according to the present invention.
Fig. 13 is a schematic diagram of a connecting section cloud image and an upper plate damage image obtained by finite element simulation when the stroke displacement percentage of the punch is 68%.
FIG. 14 is a schematic representation of predicted cross-sectional segmentation and upper surface damage images at a punch stroke displacement percentage of 100% according to the present invention.
Fig. 15 is a schematic diagram of a connecting section cloud image and an upper plate damage image obtained by finite element simulation when the stroke displacement percentage of the punch is 100%.
Detailed Description
The present invention is described in further detail below to enable those skilled in the art to practice the invention by reference to the specification.
The invention provides a prediction method for CFRP self-piercing riveting forming process based on deep learning, which comprises the following steps:
step one, collecting the stroke displacement percentage of a punch;
step two, establishing a prediction model;
wherein the prediction model is a new generation model of an countermeasure network based on the condition of CNN with residual blocks, namely a generation model, which is composed of two neural networks, a generator and a discriminator, as shown in fig. 1, the generator generates images from random noise, and the discriminator tries to distinguish between real images and generated images, the two networks are mutually game until the generator can generate sufficiently realistic images, in order to improve the performance of the generation model, the residual blocks are used to construct the generator, and the residual blocks are a structure capable of skipping one or more layers of networks, which can avoid the problem of gradient disappearance, and promote the input images, and concretely, the generator comprises:
the 1 st, 3 rd, 5 th, 7 th and 9 th layers are all convolution residual blocks, the 2 nd, 4 th, 6 th and 8 th layers are all transposed residual blocks, the 10 th layer is an activation function layer and is used for improving an input image, and the objective function of the generator is as follows:
wherein V (G) is the least squares loss of the generator G, E c,z For the expectation that the punch stroke displacement percentage is c and the random noise is z, G (c, z) is a function of the generator G generating data according to the punch stroke displacement percentage c and the random noise z, D (c, z), c giving the discriminator D the probability that the generated data G (c, z) and the punch stroke displacement percentage is c are real data, λ is a regularization coefficient, and F (G) is a regularization term for constraining the output range or distribution of the generator G;
the regularization term satisfies:
F(G)＝F L1 (G)＝E x,c,z [||x-G(c,z)|| 1 ]；
wherein F is L1 (G) To use the L1 norm as a function of regularization term, E x,c,z For the expectation that the true data is x, the percent punch travel displacement is c and the random noise is z, i x-G (c, z) i 1 For L1 range between real data x and generated data G (c, z)I.e. the sum of absolute values between them.
As shown in fig. 2, the convolution residual block includes a convolution layer, a batch normalization layer, a rectification layer, a convolution layer and a batch normalization layer which are sequentially connected, and is used for extracting features, wherein the convolution kernel size of the convolution layer is 3×3;
the rectifying layer is a ReLU activation function and meets the following conditions:
wherein f 1 (x) For the output value of the ReLU activation function, x 1 The input value for the ReLU activation function.
As shown in fig. 3, the transposed residual block includes a transposed convolution adjustment channel, a transposed convolution layer, a batch standardization layer, and a rectification layer that are sequentially connected, where the convolution kernel size of the transposed convolution adjustment channel is 1×1, and is used to adjust the number of image channels of the transposed residual block, the convolution kernel size of the transposed convolution layer is 3×3, if the number of channels of the image output by the convolution residual block is the same as the number of channels of the transposed convolution layer, the output of the convolution residual block directly passes through the transposed convolution layer, the batch standardization layer, and the rectification layer that are sequentially connected, and if the number of channels of the image output by the convolution residual block is different from the number of channels of the transposed convolution layer, the output of the convolution residual block adjusts the number of channels through the transposed convolution adjustment channel, and the rectification layer is a ReLU activation function.
The activation function layer may use Softmax, sigmoid, tanh, relu, leak Relu, PRelu, RRelu, elu, selu, swish or Maxout but because the softmax function may map the inputs to a probability distribution between (0, 1) such that the sum of the outputs is 1, the probability of different classes (mainly for multi-classification) can be better distinguished, so the activation function layer of the present application selects the softmax function.
As shown in FIG. 4, layers 1-4 of the discriminator are all composed of a convolution layer, a batch standardization layer and a rectification layer, layer 5 is a convolution layer, the effect obtained by adding a conventional Loss function L1Loss (or CEE) into an original cGAN Loss is clearer than the edge of the original Loss function, and the objective function of the discriminator is as follows:
wherein V (D) is the least squares loss of the arbiter D, E x,c For the expectation that the real data is x and the punch stroke displacement percentage is c, D (x, c) gives the probability that the real data x and the punch stroke displacement percentage c are the real data for the discriminator D;
the rectification layer is a LeakReLU function and meets the following conditions:
wherein f 2 (x 2 ) For the output value of the LeakReLU function, x 2 The input value of the LeakReLU function is that alpha is a smaller positive number;
in this embodiment, α=0.3, and 0.3 represents the slope of the negative half-section, i.e., when the input is positive, the output is equal to the input, and when the input is negative, the output is equal to the input multiplied by 0.3.
In the present embodiment, the cGAN model is built by one of the deep learning frameworks Tensorflow, keras, pytorch, caffe, theano, paddlePaddle, MXNet, CNTK, chainer, deeplearning j;
the loss function of the prediction model is LSGAN loss and traditional L1 loss.
Step three, as shown in fig. 5, training the prediction model with a training data set to obtain a deep learning-based condition generation countermeasure network, which specifically includes the following steps:
1. constructing an initial training set:
the initial training set should contain enough samples, so that the initial training set is used for SPR connection technology, and a joint profile image of the connecting piece under the selected punch displacement is obtained through experiments and simulation to serve as an initial sample image, so that a prediction model can learn from the initial sample image, and the model of the initial training set is trained;
2. data preprocessing:
labeling an initial sample image in an initial training set through a labeling tool, labeling a rivet, a lower plate, a layered outline of a carbon fiber composite material and a damaged area on the upper surface of the carbon fiber, taking other areas as a background, removing unnecessary information contained in an OM image obtained through experiments, such as surface details, texture of the material, image contrast, noise and the like, and only retaining basic information such as the geometric shape and the position of the material to obtain a section segmentation image corresponding to the initial sample image after labeling as an initial training sample;
in this embodiment, the labeling tool may select one or more combinations of EISeg, CVAT, VIA, labelme, pixlAnnotationTool.
3. Performing image enhancement processing on the initial training sample, generating a corresponding profile segmentation image, and taking the initial training sample and the enhanced image as a data set:
randomly translating, overturning, rotating, adding noise, blurring, sharpening, cutting, scaling, color adjusting and shading adjusting the sectional segmented image;
in this embodiment, the tool may be one or more of OpenCV, imgauge, skimage, PIL, augmentor, albumentations.
The data enhancement is not limited to the data pretreatment, and the data enhancement can be performed first and then the data pretreatment can be performed;
4. training the cGAN model:
wherein the training parameters include: training an optimizer, a learning rate, an L2 regularization coefficient, a batch size and a loss function;
in the training process, firstly, parameters of a generator and a discriminator are initialized, learning rate, an optimizer, a loss function and other super parameters are set, then a batch of real data x and corresponding punch stroke displacement percentages c are randomly extracted from a data set formed by an initial training sample and an enhanced image, random noise z and the punch stroke displacement percentages c are input into the generator to obtain a batch of generated data G (c, z), the generated data G (c, z) and the punch stroke displacement percentages c are input into the discriminator to obtain a discrimination result D (G, z), the real data x and the corresponding punch stroke displacement percentages c are input into the discriminator to obtain a discrimination result D (x, c), the loss function of the discriminator is calculated, the parameters of the discriminator are updated according to the loss function, D (x, c) is close to 1, D (G (c, z), c) is close to 0, then the training is carried out once, the loss function of the generator is calculated, and the parameters of the generator are updated through feedback of the discriminator, D (G (c, z) is close to 1, and convergence is achieved.
The training process is to input the punch stroke displacement percentage and the random noise vector into the generator, the discriminator receives three inputs, namely the punch stroke displacement percentage, the generated picture and the real picture, the punch stroke displacement percentage and the real picture are adjusted to be the same in length and height through the python code, the punch stroke displacement percentage and the real picture are spliced together in the channel dimension, the training process can guide the generator to generate a profile image conforming to the condition, the generator generates a segmented image of a profile joint and transmits the segmented image to the discriminator, the discriminator also receives an external real image and the punch stroke displacement percentage, the discriminator also splices the punch stroke displacement percentage and the image in the channel dimension to help the discriminator distinguish the real profile image and the generated profile image, the generator and the discriminator are continuously trained so that the generator can generate more real data, and the steps are repeated until the generator can generate a sufficiently real image with sufficiently high accuracy.
In this embodiment, the loss function curves of the generator and the arbiter are observed to see if they tend to be stable or periodically change to determine if convergence is achieved, or the quality of the sample generated by the generator is observed to see if it is clear, and if it is desired to determine if convergence is achieved.
In this embodiment, the optimizer of the overall training process may select one or more combinations of Adam, adamax, nadam, BGD, SGD, MBGD, momentum, adagrad, adadelta or RMSprop.
And fourthly, inputting random noise and punch stroke displacement percentage into the deep learning-based condition generation countermeasure network to obtain a predicted section segmentation and upper surface damage image.
Examples
Step one, determining that the total displacement of the punch stroke is 4.621, and collecting the percentage of the displacement of the punch stroke to be 19%, 68% and 100%;
wherein, as shown in fig. 6, the structure of the lower die is schematically shown, the diameter of the groove is 12mm, the maximum depth is 1.8mm, as shown in fig. 7, the structure of the rivet with the maximum depth is 5mm, and the carbon fiber reinforced composite material and the aluminum alloy are subjected to SPR.
Step two, establishing a prediction model;
training the prediction model by using a data set formed by the initial training sample and the enhanced image to obtain a condition generation countermeasure network based on deep learning;
the initial training set is 40 SPR riveted images with different punch displacement percentages, as shown in fig. 8, is an image extracted by finite elements, as shown in fig. 9, is a cross-sectional segmented image obtained by performing data preprocessing on the finite element image in fig. 8, and then performing image enhancement processing on the cross-sectional segmented image, and specifically includes: performing brightness adjustment and color adjustment on all the section segmentation images, performing random angle rotation with a maximum angle of 6 degrees on all the label images, performing weak image changes such as clipping with a scale of 0.9 on one half of all the section segmentation images, and generating 8000 data sets;
the Adam optimizer adopted in the whole training process, bayesian optimization is a sequential design strategy for global optimization, and the global optimal solution can be found faster without assuming any functional form, so that the learning rate is set to be 2.562 multiplied by 10 through Bayesian optimization -5 The L2 regularization coefficient lambda is set to 10 -5 The batch size is 16, 2 dropout layers are used in the discriminator to reduce the interdependence among the features and prevent overfitting, so that the generalization capability of the model is improved;
step four, respectively inputting random noise and 3 punch stroke displacement percentages into the deep learning-based condition generation countermeasure network to obtain predicted section segmentation and upper surface damage images;
fig. 10 shows a cross-sectional segmentation and upper surface damage image predicted when the punch stroke displacement percentage is 19%, and fig. 11 shows a connecting cross-sectional cloud image and upper plate damage image obtained by finite element simulation performed at the same punch stroke displacement percentage; as shown in fig. 12 and 13, in this embodiment, the cross-section segmentation and upper surface damage image predicted when the stroke displacement percentage of the punch is 68% and the connection cross-section cloud image and upper plate damage image obtained by finite element simulation are shown; as shown in fig. 14 and 15, in this embodiment, when the stroke displacement percentage of the punch is 100%, the predicted sectional segmentation and upper surface damage image and the connection sectional cloud image and upper plate damage image obtained by finite element simulation are measured, the interlocking length, the bottom thickness and the nail head height of the two images are measured, and the predicted result in this embodiment can reach 95.36% of the segmented sectional image in the simulation model, thereby meeting the requirement of measuring the geometrical parameters of the subsequent SPR riveting.
According to the prediction method for the CFRP self-piercing riveting forming process based on deep learning, based on experimental data, no assumption exists, so that a prediction result is closer to an experimental result, the complexity of operation is reduced, meanwhile, the complex and nonlinear relation in the rivet change process is captured by utilizing the feature extraction and expression capability of deep learning, the accuracy is improved, and the generalization capability of a model is improved; compared with the traditional simulation, once training is finished, the prediction time is extremely short (within a few seconds), in addition, the problems of grid division and time step are solved, the damage to the section shape and the upper surface of the rivet under different displacement percentages is reduced, the measurement time of geometric parameters of the SPR riveting section is effectively shortened, the subjective factor interference of a measurer is avoided, and the measurement precision is improved.
Although embodiments of the present invention have been disclosed above, it is not limited to the details and embodiments shown, it is well suited to various fields of use for which the invention is suited, and further modifications may be readily made by one skilled in the art, and the invention is therefore not to be limited to the particular details and examples shown and described herein, without departing from the general concepts defined by the claims and the equivalents thereof.

Claims (6)

1. The prediction method for the CFRP self-piercing riveting forming process based on deep learning is characterized by comprising the following steps of:
step one, collecting the stroke displacement percentage of a punch;
step two, establishing a prediction model;
wherein the predictive model includes a generator and a arbiter;
the 1 st, 3 rd, 5 th, 7 th and 9 th layers of the generator are all convolution residual blocks, the 2 nd, 4 th, 6 th and 8 th layers are all transposed residual blocks, and the 10 th layer is an activation function layer;
the 1 st to 4 th layers of the discriminator consist of a convolution layer, a batch standardization layer and a rectification layer, and the 5 th layer is the convolution layer;
the convolution residual block comprises a convolution layer, a batch standardization layer, a rectifying layer, a convolution layer and a batch standardization layer which are sequentially connected;
the transposition residual block comprises transposition convolution adjusting channels which are arranged in parallel, and a transposition convolution layer, a batch standardization layer and a rectification layer which are connected in sequence;
the convolution kernel size of the transposed convolution adjustment channel is 1 multiplied by 1, and the convolution kernel size of the transposed convolution layer is 3 multiplied by 3;
the objective function of the generator is:
wherein V (G) is the least squares loss of the generator G, E c,z D (G (c, z) gives the arbiter D the generation data G (c, z) and the condition information for the expectation that the condition information is c and the random noise is zFor c is the probability of the real data, λ is the regularization coefficient, and F (G) is the regularization term;
the objective function of the discriminator is:
wherein V (D) is the least squares loss of the arbiter D, E x,c For the expectation that the real data is x and the condition information is c, D (x, c) is a probability that the real data x and the condition information c are the real data given by the discriminator D;
training the prediction model by using carbon fiber composite materials and alloy self-piercing riveting images under a plurality of different punch stroke displacement percentages to obtain a deep learning-based condition generation countermeasure network;
and fourthly, inputting random noise and punch stroke displacement percentage into the deep learning-based condition generation countermeasure network to obtain a predicted section segmentation and upper surface damage image.
2. The method for predicting CFRP self-piercing riveting forming process based on deep learning of claim 1, wherein said step three specifically comprises:
step 1, constructing an initial training set by using a plurality of carbon fiber composite materials and alloy self-piercing riveting images with different punch stroke displacement percentages;
step 2, carrying out data preprocessing on the initial training set;
step 3, performing image enhancement processing on the training set sample subjected to data preprocessing to generate a corresponding section segmentation image;
and 4, training the prediction model by taking the training set sample obtained after the corresponding section segmentation image and the data preprocessing as a data set to obtain a condition generation countermeasure network based on deep learning.
3. The method for predicting CFRP self-piercing riveting forming process based on deep learning of claim 2 wherein the data preprocessing comprises:
and labeling the initial sample image in the initial training set by a labeling tool, and labeling the outlines of the rivet, the lower plate, the carbon fiber composite material layering and the damaged area on the upper surface of the carbon fiber, wherein other areas are used as the background.
4. A prediction method for CFRP self-piercing riveting forming process based on deep learning as in claim 3, wherein said image enhancement process is:
and carrying out color adjustment, brightness adjustment, translation, blurring, clipping, sharpening, overturning, scaling or rotation of random angles on the training set samples after the data preprocessing.
5. The method for predicting CFRP self-piercing riveting forming process based on deep learning of claim 4 wherein the training parameters of the prediction model comprise: training optimizers, learning rate, L2 regularization coefficients, batch size, and loss function.
6. The method for predicting CFRP self-piercing riveting process based on deep learning of claim 5 wherein said alloy is aluminum alloy.
CN202310660481.8A 2023-06-06 2023-06-06 Prediction method for CFRP self-piercing riveting forming process based on deep learning Active CN116757068B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310660481.8A CN116757068B (en) 2023-06-06 2023-06-06 Prediction method for CFRP self-piercing riveting forming process based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310660481.8A CN116757068B (en) 2023-06-06 2023-06-06 Prediction method for CFRP self-piercing riveting forming process based on deep learning

Publications (2)

Publication Number Publication Date
CN116757068A CN116757068A (en) 2023-09-15
CN116757068B true CN116757068B (en) 2024-01-09

Family

ID=87954465

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310660481.8A Active CN116757068B (en) 2023-06-06 2023-06-06 Prediction method for CFRP self-piercing riveting forming process based on deep learning

Country Status (1)

Country Link
CN (1) CN116757068B (en)

Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108162420A (en) * 2017-12-26 2018-06-15 吉林大学 A kind of thermoplasticity carbon fibre composite plate connects method with alloy sheets self-piercing riveting device and willow
KR20190070050A (en) * 2017-12-12 2019-06-20 한국생산기술연구원 Connecting different materials, apparatus and method for joining different materials
CN110956591A (en) * 2019-11-06 2020-04-03 河海大学 Dam crack image data enhancement method based on depth convolution generation countermeasure network
CN112184731A (en) * 2020-09-28 2021-01-05 北京工业大学 Multi-view stereo depth estimation method based on antagonism training
KR20210065623A (en) * 2019-11-27 2021-06-04 한국생산기술연구원 Self piercing riveting device and self piercing riveting method
CN113971676A (en) * 2021-10-20 2022-01-25 长三角先进材料研究院 Automatic segmentation method for connecting piece section image based on deep learning
CN114119391A (en) * 2021-11-04 2022-03-01 深圳市智佐生物科技有限公司 Method for establishing anti-neural network denoising model and ultrasonic image denoising method
CN114706023A (en) * 2022-06-06 2022-07-05 中国科学技术大学 Magnetic detection method for biomolecular interaction
CN115019120A (en) * 2022-05-17 2022-09-06 长三角先进材料研究院 Connecting piece profile prediction method for generating countermeasure network based on conditions
WO2022267641A1 (en) * 2021-06-25 2022-12-29 南京邮电大学 Image defogging method and system based on cyclic generative adversarial network
CN116012739A (en) * 2023-02-01 2023-04-25 湖南大学 Unmanned aerial vehicle remote sensing video blind motion blur removing method based on countermeasure learning and contrast learning

Patent Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20190070050A (en) * 2017-12-12 2019-06-20 한국생산기술연구원 Connecting different materials, apparatus and method for joining different materials
CN108162420A (en) * 2017-12-26 2018-06-15 吉林大学 A kind of thermoplasticity carbon fibre composite plate connects method with alloy sheets self-piercing riveting device and willow
CN110956591A (en) * 2019-11-06 2020-04-03 河海大学 Dam crack image data enhancement method based on depth convolution generation countermeasure network
KR20210065623A (en) * 2019-11-27 2021-06-04 한국생산기술연구원 Self piercing riveting device and self piercing riveting method
CN112184731A (en) * 2020-09-28 2021-01-05 北京工业大学 Multi-view stereo depth estimation method based on antagonism training
WO2022267641A1 (en) * 2021-06-25 2022-12-29 南京邮电大学 Image defogging method and system based on cyclic generative adversarial network
CN113971676A (en) * 2021-10-20 2022-01-25 长三角先进材料研究院 Automatic segmentation method for connecting piece section image based on deep learning
CN114119391A (en) * 2021-11-04 2022-03-01 深圳市智佐生物科技有限公司 Method for establishing anti-neural network denoising model and ultrasonic image denoising method
CN115019120A (en) * 2022-05-17 2022-09-06 长三角先进材料研究院 Connecting piece profile prediction method for generating countermeasure network based on conditions
CN114706023A (en) * 2022-06-06 2022-07-05 中国科学技术大学 Magnetic detection method for biomolecular interaction
CN116012739A (en) * 2023-02-01 2023-04-25 湖南大学 Unmanned aerial vehicle remote sensing video blind motion blur removing method based on countermeasure learning and contrast learning

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
Equivalent characterization of pre-strained material properties and mechanical behavior prediction of steel/aluminum self-piercing riveted joints;Xue Zhigang等;《Thin-Walled Structures》(第182期);1-18 *
基于神经网络的医学图像融合算法研究;吴帆;《中国优秀硕士学位论文全文数据库 (信息科技辑)》(第12期);I138-310 *
自冲铆接工艺参数对接头性能影响的研究;杜爱民等;《新技术新工艺》(第4期);1-7 *

Also Published As

Publication number Publication date
CN116757068A (en) 2023-09-15

Similar Documents

Publication Publication Date Title
WO2021139069A1 (en) General target detection method for adaptive attention guidance mechanism
Liu et al. Deep learning in sheet metal bending with a novel theory-guided deep neural network
CN112465790A (en) Surface defect detection method based on multi-scale convolution and trilinear global attention
CN114638784B (en) Method and device for detecting surface defects of copper pipe based on FE-YOLO
CN111444865B (en) Multi-scale target detection method based on gradual refinement
CN116883393B (en) Metal surface defect detection method based on anchor frame-free target detection algorithm
CN113642486A (en) Unmanned aerial vehicle distribution network inspection method with airborne front-end identification model
CN113033547A (en) Welding state classification method based on MobileNet V2
Li et al. AUTOMATED CLASSIFICATION OF COLD ROLLED STRIP WELD SEAM DEFECTS USING LIGHTWEIGHT DEEP LEARNING NETWORKS
CN111310791A (en) Dynamic progressive automatic target identification method based on small sample number set
CN116757068B (en) Prediction method for CFRP self-piercing riveting forming process based on deep learning
Liu et al. Forming process prediction of a self-piercing riveted joint in carbon fibre reinforced composites and aluminium alloy based on deep learning
CN117392450A (en) Steel material quality analysis method based on evolutionary multi-scale feature learning
CN116630343A (en) Rivetless riveting joint profile image segmentation method based on deep learning
CN115496941B (en) Structural health diagnosis method based on knowledge enhanced computer vision
CN116883741A (en) Small sample insect pest detection method based on pyramid attention mechanism
CN116958662A (en) Steel belt defect classification method based on convolutional neural network
CN116740051A (en) Steel surface defect detection method based on improved YOLO model
Wu et al. Detection of defects on aluminum profile surface based on improved YOLO
CN116485021A (en) Coal enterprise technical skill person sentry matching prediction method and system
CN116311514A (en) Pedestrian detection and attitude estimation method based on 2D-3D coordinate matching strategy
CN111222529A (en) GoogLeNet-SVM-based sewage aeration tank foam identification method
CN116758394B (en) Prediction method for forming section of rivet-free riveting joint based on deep learning
Tian et al. Feature fusion–based preprocessing for steel plate surface defect recognition
CN113486857B (en) YOLOv 4-based ascending safety detection method and system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant