Physical information and deep learning fused bolt connection model construction method and device

Abstract

The invention discloses a bolt connection model construction method and device fusing physical information and deep learning, and the method comprises the steps: obtaining a hysteresis response data set of a bolt connection interface according to an Ivon model through bolt connection excitation; marking the hysteretic response data set to obtain a target data set; constructing a target loss function according to the hysteresis response data set; constructing an initial training model based on a convolutional neural network and a long-short-term memory network; according to the target loss function and the target data set, the initial training model is trained, and a bolt hysteresis system prediction model.The bolt connection model identification method can improve the efficiency and accuracy of bolt connection model identification and can be widely applied to the technical field of computer assistance.

Description

Technical Field

[0001] The present invention relates to the field of computer-aided technology, and in particular to a method and device for constructing a bolt connection model integrating physical information and deep learning. Background Art

[0002] Bolted connection is one of the most commonly used connection methods in mechanical structures and is widely used in industries such as construction, automobiles, and aerospace. Bolted connection forms a strong connection between two or more parts by tightening bolts and nuts to provide the necessary structural stability. The current bolt connection model methods can be divided into constitutive models and phenomenal models. The constitutive model refers to starting from the microscopic physical behavior of the connection interface and gradually establishing the mechanical model of the connection interface using physical laws. However, there are some simple assumptions that are relied on and the problem of the roughness of the actual surface of the bolted connection cannot be fully characterized. In addition, the variance of the statistical parameters is seriously dependent on the resolution of the measuring instrument, and it is impossible to uniquely describe and analyze the rough surface; the phenomenal model refers to using the macroscopic mechanical response of the connection structure and using system identification theory and methods to obtain the mechanical model of the connection interface. However, the phenomenal model has model errors, is difficult to characterize the viscous and sliding friction states, and is also difficult to characterize the complex connection interface and the nonlinear hysteresis response of the bolted connection under load. As a result, the efficiency of identifying the bolted connection model is low, the computational cost is high, and the accuracy of identification is reduced. Summary of the invention

[0003] In view of this, the main purpose of the embodiments of the present invention is to provide a method and device for constructing a bolt connection model that integrates physical information and deep learning, in order to solve at least one of the problems of the prior art. The present invention can improve the efficiency and accuracy of identifying the bolt connection model.

[0004] To achieve the above-mentioned purpose, an embodiment of the present invention provides a method for constructing a bolt connection model by integrating physical information and deep learning, the method comprising:

[0005] Through the bolt connection excitation, the hysteresis response data set of the bolt connection interface is obtained according to the Ivan model;

[0006] Performing labeling processing on the hysteresis response data set to obtain a target data set;

[0007] Constructing a target loss function according to the hysteresis response data set;

[0008] Construct an initial training model based on convolutional neural network and long short-term memory network;

[0009] The initial training model is trained according to the target loss function and the target data set to obtain a bolt hysteresis system prediction model.

[0010] In some embodiments, a method for constructing a bolt connection model integrating physical information and deep learning further includes:

[0011] The bolt hysteresis system prediction model is verified according to the verification data set in the target data set.

[0012] In some embodiments, the step of obtaining a hysteresis response data set of a bolt connection interface according to an Ivan model through bolt connection excitation includes the following steps:

[0013] According to the Ivan model, a bolt connection structure is simulated and modeled to obtain a bolt connection simulation model;

[0014] The bolt connection excitation is applied to the bolt connection simulation model to generate a hysteresis response data set of the bolt connection interface.

[0015] In some embodiments, the labeling process of the hysteresis response data set to obtain a target data set includes the following steps:

[0016] Randomly selecting a number of the hysteresis response data sets as known data sets, and dividing the known data sets according to a preset ratio to obtain a training data set and a verification data set;

[0017] Using the remaining hysteresis response data set as a test data set;

[0018] The target data set includes the training data set, the verification data set and the test data set.

[0019] In some embodiments, constructing a target loss function according to the hysteresis response data set comprises the following steps:

[0020] Obtaining a normalized hysteresis force in the hysteresis response data set, and obtaining a displacement of the bolt connection structure in the hysteresis response data set;

[0021] constructing a first loss function according to the normalized hysteresis force;

[0022] constructing a second loss function according to the smoothness between the displacement and the normalized hysteresis force;

[0023] The target loss function is obtained according to the first loss function and the second loss function.

[0024] In some embodiments, the initial training model is trained according to the target loss function and the target data set to obtain a bolt hysteresis system prediction model, comprising the following steps:

[0025] Classifying the training data set in the target data set according to the speed condition to obtain a first data set and a second data set;

[0026] The first data set and the second data set are respectively input into the first channel and the second channel of the initial training model, and the initial training model is trained in combination with the constraints of the target loss function to obtain the bolt hysteresis system prediction model.

[0027] To achieve the above object, another aspect of an embodiment of the present invention provides a bolt connection model construction device integrating physical information and deep learning, the device comprising:

[0028] The first module is used to obtain the hysteresis response data set of the bolt connection interface according to the Ivan model through bolt connection excitation;

[0029] The second module is used to label the hysteresis response data set to obtain a target data set;

[0030] A third module is used to construct a target loss function according to the hysteresis response data set;

[0031] The fourth module is used to build an initial training model based on convolutional neural networks and long short-term memory networks;

[0032] The fifth module is used to train the initial training model according to the target loss function and the target data set to obtain a bolt hysteresis system prediction model.

[0033] To achieve the above-mentioned purpose, another aspect of an embodiment of the present invention provides an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the aforementioned method for constructing a bolt connection model that integrates physical information and deep learning.

[0034] To achieve the above-mentioned purpose, another aspect of an embodiment of the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the aforementioned method for constructing a bolt connection model that integrates physical information and deep learning.

[0035] To achieve the above object, another aspect of an embodiment of the present invention provides a computer program product or a computer program, the computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the aforementioned method for constructing a bolt connection model that integrates physical information and deep learning.

[0036] The embodiments of the present invention include at least the following beneficial effects: The present invention provides a method for constructing a bolt connection model that integrates physical information and deep learning. The scheme obtains a hysteresis response data set of a bolt connection interface according to the Ivan model through bolt connection excitation; the hysteresis response data set is marked to obtain a target data set; a target loss function is constructed according to the hysteresis response data set; an initial training model is constructed based on a convolutional neural network and a long short-term memory network; the initial training model is trained according to the target loss function and the target data set to obtain a bolt hysteresis system prediction model, which can improve the efficiency and accuracy of identifying bolt connection models. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0038] Figure 1 It is a flow chart of a method for constructing a bolt connection model integrating physical information and deep learning provided by an embodiment of the present invention;

[0039] Figure 2 is a schematic diagram of a bolt connection structure provided by an embodiment of the present invention;

[0040] Figure 3a-Figure 3b is a schematic diagram of a bolt connection simulation model provided by an embodiment of the present invention;

[0041] Figure 4 is a schematic diagram of an LSTM unit structure provided by an embodiment of the present invention;

[0042] Figure 5 It is a schematic diagram of the deep learning framework of CNN-LSTM provided by an embodiment of the present invention;

[0043] Figure 6 is a schematic diagram of a bolt connection element subjected to tensile force in a simulation experiment provided by an embodiment of the present invention;

[0044] Figure 7a-7d is a schematic diagram of the recognition result of the deep learning model provided by an embodiment of the present invention;

[0045] Figure 8 is a schematic diagram of a torque-bearing bolt connection element in a simulation experiment provided by an embodiment of the present invention;

[0046] Fig. 9is a schematic diagram of a response result under a torsional hysteresis loop provided by an embodiment of the present invention;

[0047] Fig.10 It is a schematic diagram of the hardware structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of the present invention, and they are only examples of devices and methods consistent with some aspects of the embodiments of the present invention as detailed in the attached claims.

[0049] It should be noted that, although the functional modules are divided in the system schematic diagram and the logical order is shown in the flow chart, in some cases, the steps shown or described may be performed in a different order than the module division in the system or the flow chart. The terms "first / S100" and "second / S200" in the specification and claims and the above-mentioned drawings may be used to describe various concepts in this article, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiment of the present invention, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the words "if" and "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determination".

[0050] The terms "at least one", "multiple", "each", "any", etc. used in the present invention, at least one includes one, two or more, multiple includes two or more, each refers to each of the corresponding multiple, and any refers to any one of the multiple.

[0051] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein are only for the purpose of describing the embodiments of the present invention and are not intended to limit the present invention.

[0052] Before describing the embodiments of the present invention in detail, some nouns and terms involved in the embodiments of the present invention are first described. The nouns and terms involved in the embodiments of the present invention are subject to the following explanations.

[0053] Bolted connection: Bolted connection is a common mechanical connection method that uses bolts to connect two or more parts together to form a fixed structure.

[0054] Deep learning: originated from the study of artificial neural networks. It combines low-level features to form more abstract high-level representations of attribute categories or features to discover distributed feature representations of data.

[0055] Physical information: When constructing a physical model (such as a mechanical model, a thermodynamic model, an electromagnetic model, etc.), physical information includes various physical quantities related to the model and their relationships.

[0056] Hysteresis system: When the input and output of a system present a nonlinear characteristic, this characteristic is called hysteresis.

[0057] Bolted connection is one of the most commonly used connection methods in mechanical structures and is widely used in industries such as construction, automobiles, and aerospace. Bolted connection forms a strong connection between two or more components by tightening bolts and nuts to provide the necessary structural stability. With the development of industrial technology, the design and application of bolted connections have gradually become intelligent and automated. Especially in scenarios with high-strength and high-precision connections, the reliability and predictive performance of bolted connections have become increasingly important. The reliability of bolted connections not only depends on the material, size, and accuracy of bolt holes of the bolts, but is also affected by many factors such as connection preload, load changes, temperature fluctuations, and environmental factors. Modeling and identification of bolted connection structures remain a hot topic and frontier in research.

[0058] At present, the bolt connection model method can be divided into constitutive model and phenomenal model. The constitutive model refers to starting from the microscopic physical behavior of the connection interface and gradually establishing the mechanical model of the connection interface using physical laws. However, there are some simple assumptions that are relied on and the problem of not being able to fully characterize the roughness of the actual surface of the bolt connection exists. In addition, the variance of the statistical parameters is seriously dependent on the resolution of the measuring instrument and cannot uniquely describe and analyze the rough surface. When large-scale projects are subjected to dynamic loads, complex nonlinear hysteresis behaviors will occur. When dealing with such problems, the existing models will generate heavy computational costs. In addition, the phenomenal model refers to using the macroscopic mechanical response of the connection structure and using system identification theory and methods to obtain the mechanical model of the connection interface. However, the phenomenal model has the problem of zero-speed detection and the difficulty of switching between the viscosity and sliding friction state equations. In addition, the phenomenal model also has the problem of difficulty in characterizing the complex connection interface and the nonlinear hysteresis response of the bolt connection under load. It can be seen that the existing bolt connection model method has low efficiency, high computational cost and low accuracy in identifying the bolt connection model.

[0059] In view of this, if Figure 1As shown, an embodiment of the present invention provides a method for constructing a bolt connection model by integrating physical information and deep learning. The method may include but is not limited to steps S100 to S500:

[0060] Step S100, obtaining a hysteresis response data set of the bolt connection interface according to the Ivan model through bolt connection excitation;

[0061] Step S200, marking the hysteresis response data set to obtain a target data set;

[0062] Step S300, constructing a target loss function according to the hysteresis response data set;

[0063] Step S400, constructing an initial training model based on a convolutional neural network and a long short-term memory network;

[0064] Step S500: training the initial training model according to the target loss function and the target data set to obtain a bolt hysteresis system prediction model.

[0065] In some embodiments, step S100 may include but is not limited to steps S110 to S120:

[0066] Step S110, performing simulation modeling on the bolt connection structure according to the Ivan model to obtain a bolt connection simulation model;

[0067] Step S120: applying the bolt connection excitation to the bolt connection simulation model to generate a hysteresis response data set of the bolt connection interface.

[0068] In steps S110 to S120 of some embodiments, a bolt connection structure can be established using simulation modeling software to obtain a bolt connection simulation model. In establishing the bolt connection simulation model, a simulation model is constructed according to the Iwan model, and the model can include geometric parameters, material properties, preload, etc. of the bolt connection structure. The parameters of the Iwan model, such as spring stiffness, friction coefficient, etc., can also be pre-set in the simulation model, and these parameters will affect the nonlinear response of the model. According to the actual application scenario, the corresponding external excitation, such as load, temperature change, vibration, etc., is input into the simulation model. When running the simulation model, the behavior of the bolt connection structure under the action of bolt connection excitations of different amplitudes will be simulated to generate a hysteresis response data set of the bolt connection interface.

[0069] refer to Figure 2 , Figure 2The actual physical structure of the bolted connection structure is shown, which consists of bolts, nuts, connected parts, etc., and is used to fix two or more parts together. This structure is very common in engineering and construction, and is used to withstand different types of loads such as tension and shear. Based on the physical characteristics and behavior of the bolted connection structure, a bolted connection simulation model can be constructed using computer software or programming language. The simulation model can simulate the performance of the bolted connection structure under different conditions, such as the response when it is affected by external conditions such as load, temperature change, vibration, etc. Through the simulation model, engineers can predict and analyze the performance of the bolted connection structure without actually manufacturing and testing the physical structure, evaluate the impact of different design parameters on the structural performance, optimize the design, reduce costs, and shorten the development cycle.

[0070] like Figure 3a-Figure 3b The bolt connection simulation model shown in Figure 1 is shown in Figure 2, where Figure 3a The Jenkins unit is one of the basic units that constitute the Iwan model. Each Jenkins unit consists of a linear spring and a Coulomb friction element (slider) to simulate the nonlinear behavior in bolted connections, especially the stick-slip characteristics. This stick-slip characteristic is that during a certain loading process, the bolt connection interface first experiences a sticky state (i.e., relative stillness) and then slips under the action of external force. The Jenkins unit can capture this nonlinear dynamic behavior of the transition from micro-slip to macro-slip.

[0071] In some embodiments, in order to simulate the nonlinear behavior of the bolted connection structure, a mathematical model such as the Iwan model may be used. Figure 3b The Iwan model shown is composed of multiple Jenkins units connected in parallel. Using the superposition characteristics of the Jenkins unit, the constitutive equation of the Iwan model can be obtained as:

[0072]

[0073] In the formula, F I represents the hysteresis force of the Ivan model; φ represents the yield displacement of the Jenkins unit; ρ(φ) represents the distribution function of the Jenkins unit with yield displacement; U represents the displacement of the bolted connection structure; Indicates the speed of the bolted structure; Function representing the hysteresis force of the Ivan model; represents the function of the velocity of the Coulomb slider; p represents the parameter specified by each Ivan model; z(t,φ) represents the displacement of the Coulomb slider; represents the velocity of the Coulomb slider. In the following, the symbol z(t,φ) is simplified to z, and the symbol Simplified description

[0074] At the same time, considering the relationship between energy dissipation and the amplitude of external load, as well as the discontinuity of stiffness at the beginning of large sliding, the distribution function of the four-parameter Iwan model can be expressed as:

[0075] ρ(φ)=Rφ -X [H(φ)-H(φ-φ max )]+Sδ(φ-φ max ) (2)

[0076] Where R represents the coefficient of the power law part; χ represents the positive power law value of the distribution; H(·) represents the unit step function; δ(·) represents the Dirac trigonometric function; φ represents the yield displacement of the Jenkins unit; φ max represents the maximum yield displacement of all Jenkins units; S represents the slope of the force-displacement curve before the large sliding, considering that the stiffness may be discontinuous at the beginning of the large sliding. At this time, the hysteresis force of the four-parameter Ivan model can be expressed as:

[0077]

[0078] Where, PI = [R, χ, S, φ max ] is the parameter of Ivan model; u represents the displacement of bolt connection structure; Indicates the speed of the bolted structure; represents the hysteresis force of the four-parameter Ivan model; z and z(t,φ) both represent the displacement of the Coulomb slider; z(t,φ max ) represents the maximum yield displacement of the Coulomb slider; The function representing the Coulomb slider velocity of the four-parameter Ivan model; represents the function of the maximum yield displacement of the Coulomb slider velocity of the four-parameter Ivan model; R represents the coefficient of the power law part; χ represents the positive power law value of the distribution; S represents the slope of the force-displacement curve before the large sliding; φ represents the yield displacement of the Jenkins unit; φ max Represents the maximum yield displacement of all Jenkins elements.

[0079] Exemplarily, the equation in formula (1) is solved, and a synthetic database containing 100 samples (e.g., independent bolt response sequences) is generated by bolt excitations of different amplitudes in numerical simulations under an external excitation of u = 0.8 sin (πt), acting on a single degree of freedom (SDOF) nonlinear system. Each simulation is performed at a sampling frequency of 50 Hz, lasts for 160 seconds, and each record contains 9001 data points.

[0080] In some embodiments, step S200 may include but is not limited to steps S210 to S220:

[0081] Step S210, randomly selecting a number of the hysteresis response data sets as known data sets, and dividing the known data sets according to a preset ratio to obtain a training data set and a verification data set;

[0082] Step S220, using the remaining hysteresis response data set as a test data set;

[0083] The target data set includes the training data set, the verification data set and the test data set.

[0084] In some embodiments, in steps S210 to S220, the hysteresis response data set Manual labeling is performed using data simulation software. Ten data sets are randomly selected from the generated hysteresis response data sets. These data sets contain Gaussian white noise (BLWN) input and the corresponding structural displacement and velocity response. The 10 randomly selected data sets are used as "known" data sets for training and verification. The "known" data sets are divided into training data sets and verification data sets according to a preset ratio. Optionally, the "known" data set is divided into training data sets and verification data sets according to a ratio of 0.8 / 0.2. The remaining hysteresis response data sets are used as "unknown" data sets (i.e., test data sets) to test the prediction effect of the trained meta-model. In addition, 50 additional configuration samples with different amplitudes (i.e., bolt excitations of different amplitudes) can be provided to assist the subsequent model training process under physical constraints.

[0085] In some embodiments, step S300 may include but is not limited to steps S310 to S320:

[0086] Step S310, obtaining a normalized hysteresis force in the hysteresis response data set, and obtaining a displacement of the bolt connection structure in the hysteresis response data set;

[0087] Step S320, constructing a first loss function according to the normalized hysteresis force;

[0088] Step S330, constructing a second loss function according to the smoothness between the displacement and the normalized hysteresis force;

[0089] Step S340: Obtain the target loss function according to the first loss function and the second loss function.

[0090] In steps S310 to S340 of some embodiments, the Jenkins unit is composed of a spring with a stiffness of k and a Coulomb slider with a critical yield displacement of φ. In the Jenkins unit, the normalized hysteresis force F can be expressed as the displacement u of the bolt connection structure, the velocity of the bolt connection structure, and the displacement u of the bolt connection structure. If it is a function composed of the displacement z of the spring element relative to the Coulomb slider, the following expression can be obtained:

[0091]

[0092] In the formula, P J =[k,φ] is the Jenkins unit parameter; k is the stiffness of the Jenkins unit spring; represents the hysteretic force of the Jenkins unit; u represents the displacement of the bolted connection structure; represents the velocity of the bolted connection structure; z represents the displacement of the Coulomb slider; represents the velocity of the Coulomb slider; H(·) represents the unit step function; φ represents the yield displacement of the Jenkins unit.

[0093] When z = ±φ, the deformation velocity of the spring in the Jenkins unit is zero. The function indicates that the elongation of the spring element connected to the sliding element is limited within the interval of -φ < z < φ. When the elongation of the spring element connected to the sliding element is within -φ < z < φ, the relative velocity of the sliding unit must be zero.

[0094] Since multiple Jenkins units can be combined in parallel to form the Iwan model, the hysteretic force of the Jenkins unit is incorporated into the subsequent construction of the Iwan model deep learning model of physical information.

[0095] Since Jenkins has typical viscoplastic-slip characteristics and is a significant non-smooth unit, a target loss function integrating physical information is designed according to the characteristics of the Jenkins unit. This target loss function can be incorporated into deep learning training. The target loss function also ensures the smoothness of the prediction results by constraining the relationship between the hysteretic force and the displacement. Exemplarily, by obtaining the normalized hysteretic force F in the hysteretic response dataset representation in the Jenkins unit (i.e., the hysteretic force F of the Jenkins unit J ), multiple Jenkins units are combined in parallel to form the Iwan model, and the physical information of the hysteretic force of the Jenkins unit is introduced, and the hysteretic force F of the Iwan model can be obtained I . According to the hysteretic force of the Iwan model and the hysteretic force predicted by deep learning construct the first loss function; construct the second loss function according to the displacement u of the bolted connection structure in the hysteretic response dataset and the hysteretic force of the Iwan model; according to the first loss function and the second loss function, the target loss function can be obtained. Optionally, the target loss function can be defined as:

[0096] L=L1+L2 (6)

[0097] In the formula,

[0098]

[0099] Among them, L represents the target loss function; represents the hysteresis force predicted by deep learning; L1 represents the loss function composed of the hysteresis force of the Ivan model of nonlinear response calculated by the equation in formula (1) and the hysteresis force predicted by deep learning, that is, the first loss function; L2 represents the loss function that combines the displacement u and the hysteresis force F I The physical information loss function of the smoothness between them is the second loss function; λ, μ, ξ represent the parameters optimized by the deep learning model; λ, μ, ξ can be optimized by the Adam optimizer.

[0100] Based on the macroscopic viscosity-slip characteristics of the bolted connection structure, the physical information of the Jenkins unit is introduced, and a physical information loss function that integrates the macroscopic viscosity-slip characteristics, displacement state and smoothness of hysteresis force is constructed to evaluate the gap between the predicted output of the deep learning model and the actual target (label).

[0101] In step S400 of some embodiments, an initial training model is constructed based on a convolutional neural network and a long short-term memory network (CNN-LSTM). Among them, the convolutional neural network (CNN) is a type of feedforward neural network (Feedforward Neural Networks) that includes convolution calculations and has a deep structure, and is one of the representative algorithms of deep learning. Their design is inspired by the visual system in biology and aims to simulate the way human vision is processed. Recurrent Neural Network (RNN) is a neural network model widely used in sequence data processing. Long Short-Term Memory Network (LSTM) is an important variant of RNN (such as Figure 4 As shown in Figure 2, by introducing the gating mechanism, the gradient disappearance and gradient explosion problems in the traditional RNN model are effectively solved.

[0102] In some embodiments, Figure 4 As shown in Figure 1, the LSTM model consists of an input layer, a hidden layer, and an output layer. There are h hidden units, a batch size of n, and the number of inputs is d. The hidden state at the previous time step is The input gate is The forget gate is The output gate is Candidate memory cells Memory Unit Then there is the following mathematical expression of LSTM:

[0103] I t =σ(X t W xi +H t-1 W hi +b i ) (9)

[0104] F t =σ(X t W xf +H t-1 W hf +b f ) (10)

[0105] O t =σ(X t W xo +H t-1 W ho +b o ) (11)

[0106]

[0107] H t =O t ⊙tanh(C t ) (14)

[0108] In the formula, and is the weight matrix; is the bias parameter matrix; σ(·) represents the sigmoid activation function; tanh(·) represents the hyperbolic tangent function; ⊙ represents the Hadamard integral (element-wise product).

[0109] In some embodiments, Figure 5 The deep learning network architecture of CNN-LSTM is shown in the figure. In the CNN-LSTM network architecture, displacement, velocity, and hysteresis force at time t-1 are used as inputs, and hysteresis force at time t is used as output. and The data is divided into loading path and unloading path data. The data features are extracted by CNN. The two different data are trained with dual-channel CNN-LSTM. A LSTM layer with 32 hidden units is used later. The more hidden units, the more accurate the prediction data. Then, to output a sequence with the same number of channels as the input channels, a fully connected layer with the same output size and the same number of input data channels is used. Finally, a regression layer is used to get the predicted response.

[0110] In some embodiments, step S500 may include but is not limited to steps S510 to S520:

[0111] Step S510, classifying the training data set in the target data set according to the speed condition to obtain a first data set and a second data set;

[0112] Step S520, inputting the first data set and the second data set into the first channel and the second channel of the initial training model respectively, training the initial training model in combination with the constraints of the target loss function, and obtaining the bolt hysteresis system prediction model.

[0113] In steps S510 to S520 of some embodiments, Figure 5 As shown, the training data sets under different amplitude states are input into the network, and the speed and As classification criteria, they are respectively introduced into the two channels of the initial training model network structure to classify the input training data set. The first data set obtained by classification is input into the first channel of the initial training model, which meets the speed condition The second data set obtained by classification is input into the second channel of the initial training model to train the network. In the initial training model network, a convolutional neural network (CNN) is used to extract features. A 1×1 convolution kernel is inserted between the 3×3 convolution kernels to compress the features. The convergence speed of the network is accelerated by using batch normalization, and the Leaky ReLU activation function is combined to extract complex features. Then it is input into the LSTM network. Each LSTM network contains two LSTM layers and one fully connected (FC) layer. Combined with the constraints of the target loss function, the initial training model network is constrained during training. First, the model is pre-trained using the Adam optimizer, and the learning rate is set to 0.001 for the first 5000 epochs, and the learning rate is reduced to 0.0001 for the next 5000 epochs. Subsequently, the L-BFGS optimizer is used to further optimize the pre-trained model until the default convergence criteria are met.

[0114] In some embodiments, the hysteresis response data set of the preprocessed bolt connection interface is input into the network for training to form a bolt hysteresis system prediction model. By using the target data set and the supplementary data set, the model is further optimized so that it can learn the time series characteristics of the bolt connection and has the ability to model the bolt connection. The displacement, velocity, and force data in the supplementary data set (optionally, the supplementary data set can be an additional 50 configuration samples provided to assist the model training process under physical constraints) provide the model with diversified features under different amplitude scenarios, thereby improving the prediction accuracy and adaptability of the model in different working conditions.

[0115] In some embodiments, a method for constructing a bolt connection model that integrates physical information and deep learning may also include, but is not limited to, step S600: verifying the bolt hysteresis system prediction model according to a verification data set in the target data set. Optionally, a verification data set is constructed by a bolt hysteresis response data set constructed by a tensile simulation experiment and a torsion simulation experiment, and the bolt hysteresis system prediction model is verified using the verification data set. The effectiveness and applicability of the proposed bolt connection model that integrates physical information and deep learning in nonlinear dynamic systems are evaluated. The system considered is a bolt hysteresis system, including a four-parameter Iwan model and a torsion hysteresis experiment. Numerical examples are used to verify the recovery effects of different models, and the sensitivity of the bolt connection model construction method that integrates physical information and deep learning of the present invention to multiple factors is analyzed, including data usage, sampling frequency, simulation time step, selection of basis function library, degree of damage, noise type, and model accuracy. Exemplarily, the standard deviation of Gaussian noise is set to Randomly assigned to the manually labeled hysteresis response data set. The specific noise formula is as follows:

[0116]

[0117] Where F is the normalized hysteresis force vector of the hysteresis response data set; F Noise is the force vector after adding noise; ζ is a scaling factor used to adjust the noise intensity; R is a random noise vector generated from a standard normal distribution, whose elements are independent and identically distributed random variables; is a diagonal matrix whose elements are the standard deviations of the measured forces Used to scale noise.

[0118] In some embodiments, numerical examples are used to explore how to use deep learning models to predict the response of bolted connection structures when they are subjected to quasi-static loads and under the influence of specific dynamic loads u=Asin(πt). In this process, the performance of deep learning models in motion mechanisms in nonlinear dynamic systems will be focused on. That is, numerical simulation is used to analyze and predict the response of bolted connection structures when subjected to periodic dynamic loads, with special attention paid to the effectiveness of deep learning models in dealing with nonlinear dynamic problems. Optionally, a finite element simulation model is established through three-dimensional modeling software, as shown in Figure 6 The bolt connection structure shown in the figure. Under the dynamic load u=Asin(πt), a deep learning model is constructed based on formulas (4)-(6). The convolutional layer of the convolutional neural network is used to remove the redundant parameters caused by noise, prevent the deep learning model from being too complicated, remove interference information such as noise, ensure the robustness of the model, and make the model retain the main physical properties. Figure 7a-7d The recognition result of the deep learning model. Among them, Figure 7a The response prediction of the four-parameter Iwan model of the simulation model when A=0.3 is shown. Figure 7b The hysteresis loop of the four-parameter Iwan model of the simulation model when A=0.3 is shown. Figure 7c The response prediction of the four-parameter Iwan model of the simulation model when A=0.7 is shown. Figure 7d The hysteresis loop of the four-parameter Iwan model of the simulation model when A=0.7 is shown. Figure 7a-7d The results and data show that the constructed deep learning model has a good recognition effect. Furthermore, through simulation research, the bolt connection experiment with torsional load excitation is used to evaluate the performance of the method of the present invention in the actual model. Optionally, due to the excellent generalization and approximation ability of the four-parameter Iwan model, the four-parameter Iwan model is used to model the hysteresis curve obtained under the bolt preload, and the following can be obtained: Figure 8 The simulation experiment is shown in Figure 1. The torque bolt connection structure is subjected to torque. By using the method of the present invention to predict the hysteresis force and hysteresis loop of the torsion test data, the following can be obtained: Fig. 9 The response results under the torsional hysteresis loop are shown. In order to visualize the recognition performance, Fig. 9 The hysteresis curve predicted by the bolt hysteresis system prediction model is shown and compared with the experimental hysteresis curve results. It can be seen that the method of the present invention provides excellent recognition ability for the bolt connection model in terms of good fit between the predicted and experimental hysteresis curves.

[0119] The bolt connection model integrating physical information and deep learning was simulated and the accuracy of the method was verified by numerical simulation. Compared with traditional constitutive models and phenomenological models, the neural network integrating physical information shows strong robustness in processing noisy data, especially when the observation data is limited, which can effectively reduce the overfitting of the model. In the numerical example, the physical information neural network has a good prediction effect on non-smoothness (such as Figure 7b This is because the method of the present invention integrates the physical information of the Jenkins unit, and has better performance when dealing with non-smoothness. In order to further verify the advantages of the neural network that integrates the physical information of the Jenkins unit for hysteresis response prediction, in the simulation experiment, since the response data generated by the bolt connection excited by the torsional load has a certain smoothness, reference Fig. 9 It can be seen that good prediction results can be achieved by using the neural network model of the method of the present invention to make predictions.

[0120] The embodiment of the present invention further provides a bolt connection model construction device integrating physical information and deep learning, which can implement the above-mentioned bolt connection model construction method integrating physical information and deep learning. The device includes:

[0121] The first module is used to obtain the hysteresis response data set of the bolt connection interface according to the Ivan model through bolt connection excitation;

[0122] The second module is used to label the hysteresis response data set to obtain a target data set;

[0123] A third module is used to construct a target loss function according to the hysteresis response data set;

[0124] The fourth module is used to build an initial training model based on convolutional neural networks and long short-term memory networks;

[0125] The fifth module is used to train the initial training model according to the target loss function and the target data set to obtain a bolt hysteresis system prediction model.

[0126] It can be understood that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0127] The embodiment of the present invention further provides an electronic device, which includes a processor and a memory, wherein the memory stores a computer program, and when the processor executes the computer program, the method for constructing a bolt connection model integrating physical information and deep learning is implemented. The electronic device can be any intelligent terminal including a tablet computer, a vehicle-mounted computer, etc.

[0128] It can be understood that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0129] refer to Fig.10 , Fig.10 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0130] The processor 701 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present invention.

[0131] The memory 702 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 702 can store an operating system and other applications. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 702, and the processor 701 calls and executes a bolt connection model construction method integrating physical information and deep learning in an embodiment of the present invention;

[0132] Input / output interface 703, used to implement information input and output;

[0133] Communication interface 704, used to realize communication interaction between the device and other devices, which can be realized through wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.);

[0134] A bus 705 that transmits information between the various components of the device (e.g., the processor 701, the memory 702, the input / output interface 703, and the communication interface 704);

[0135] The processor 701 , the memory 702 , the input / output interface 703 and the communication interface 704 are connected to each other in communication within the device via a bus 705 .

[0136] An embodiment of the present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned method for constructing a bolt connection model that integrates physical information and deep learning.

[0137] It can be understood that the contents of the above method embodiments are all applicable to the present storage medium embodiments, the functions specifically implemented by the present storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0138] The embodiment of the present invention also provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the aforementioned method for constructing a bolt connection model integrating physical information and deep learning.

[0139] In summary, the bolt connection model construction method and device integrating physical information and deep learning according to the embodiment of the present invention have the following advantages:

[0140] 1. The embodiment of the present invention generates a synthetic data set through simulation modeling software, which ensures the validity and expansion of the data set on the one hand, and avoids the difficulties and labor and time consumption caused by collecting bolt node data sets on the other hand.

[0141] 2. The embodiment of the present invention constructs a physical information loss function that integrates the smoothness of the displacement state and the hysteresis force, which can model the bolt connection model and show good convergence.

[0142] 3. The embodiments of the present invention can efficiently and accurately identify bolt connection models, including complex working conditions such as tension and bending, and can also be applied to the identification of bolt connection structures under loads such as time-varying normal stiffness and high temperature.

[0143] 4. The embodiment of the present invention only uses displacement, velocity and force data, and data acquisition is convenient and efficient. In addition, the bolt hysteresis system prediction model of the embodiment of the present invention has high robustness, good recognition effect and strong model versatility.

[0144] In some selectable embodiments, the function / operation mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the function / operation involved, the two boxes shown in succession can actually be executed substantially simultaneously or the boxes can sometimes be executed in reverse order. In addition, the embodiment presented and described in the flow chart of the present invention is provided by way of example, for the purpose of providing a more comprehensive understanding of technology. The disclosed method is not limited to the operation and logic flow presented herein. Selectable embodiments are expected, wherein the order of various operations is changed and the sub-operation of a part for which is described as a larger operation is performed independently.

[0145] In addition, although the present invention is described in the context of functional modules, it should be understood that, unless otherwise specified, one or more of the functions and / or features described may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in separate physical devices or software modules. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the present invention. More specifically, in view of the properties, functions, and internal relationships of the various functional modules in the device disclosed herein, the actual implementation of the module will be understood within the conventional skills of the engineer. Therefore, those skilled in the art can implement the present invention set forth in the claims without excessive experimentation using ordinary techniques. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present invention, which is determined by the full scope of the appended claims and their equivalents.

[0146] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0147] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0148] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0149] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0150] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0151] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

[0152] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the described embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present invention.

Claims

1. A method for constructing a bolt connection model integrating physical information and deep learning, characterized in that: Through the bolt connection excitation, the hysteresis response data set of the bolt connection interface is obtained according to the Ivan model; Performing labeling processing on the hysteresis response data set to obtain a target data set; Constructing a target loss function according to the hysteresis response data set; Construct an initial training model based on convolutional neural network and long short-term memory network; The initial training model is trained according to the target loss function and the target data set to obtain a bolt hysteresis system prediction model.

2. A method for constructing a bolt connection model integrating physical information and deep learning according to claim 1, characterized in that: The following steps are also included: The bolt hysteresis system prediction model is verified according to the verification data set in the target data set.

3. The method for constructing a bolt connection model integrating physical information and deep learning according to claim 1, characterized in that: The method of obtaining a hysteresis response data set of a bolt connection interface according to the Ivan model through bolt connection excitation includes the following steps: According to the Ivan model, a bolt connection structure is simulated and modeled to obtain a bolt connection simulation model; The bolt connection excitation is applied to the bolt connection simulation model to generate a hysteresis response data set of the bolt connection interface.

4. The method for constructing a bolt connection model integrating physical information and deep learning according to claim 1, characterized in that: The step of labeling the hysteresis response data set to obtain a target data set includes the following steps: Randomly selecting a number of the hysteresis response data sets as known data sets, and dividing the known data sets according to a preset ratio to obtain a training data set and a verification data set; Using the remaining hysteresis response data set as a test data set; The target data set includes the training data set, the verification data set and the test data set.

5. The method for constructing a bolt connection model integrating physical information and deep learning according to claim 1, characterized in that: The step of constructing a target loss function according to the hysteresis response data set comprises the following steps: Obtaining a normalized hysteresis force in the hysteresis response data set, and obtaining a displacement of the bolt connection structure in the hysteresis response data set; constructing a first loss function according to the normalized hysteresis force; constructing a second loss function according to the smoothness between the displacement and the normalized hysteresis force; The target loss function is obtained according to the first loss function and the second loss function.

6. The method for constructing a bolt connection model integrating physical information and deep learning according to claim 1, characterized in that: The initial training model is trained according to the target loss function and the target data set to obtain a bolt hysteresis system prediction model, comprising the following steps: Classifying the training data set in the target data set according to the speed condition to obtain a first data set and a second data set; The first data set and the second data set are respectively input into the first channel and the second channel of the initial training model, and the initial training model is trained in combination with the constraints of the target loss function to obtain the bolt hysteresis system prediction model.

7. A bolt connection model construction device integrating physical information and deep learning, characterized in that: include: The first module is used to obtain the hysteresis response data set of the bolt connection interface according to the Ivan model through bolt connection excitation; The second module is used to label the hysteresis response data set to obtain a target data set; A third module is used to construct a target loss function according to the hysteresis response data set; The fourth module is used to build an initial training model based on convolutional neural networks and long short-term memory networks; The fifth module is used to train the initial training model according to the target loss function and the target data set to obtain a bolt hysteresis system prediction model.

8. An electronic device, characterized in that: including a processor and a memory; The memory is used to store programs; The processor executes the program to implement the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The storage medium stores a program, and the program is executed by a processor to implement the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images