Assembly interface physical field time sequence dynamic rapid prediction method and related device

Abstract

The invention belongs to the technical field of mechanical design, and discloses an assembly interface physical field time sequence dynamic rapid prediction method and related device.The prediction method comprises the steps that load-physical field feature mapping is conducted on a matching surface of an assembly structure, and a load sequence, a contact pressure field sequence and a temperature field sequence are obtained; performing transient rapid prediction on a physical field of the assembly interface to obtain a transient contact pressure field and a transient temperature field of the assembly interface; and according to the transient contact pressure field and the transient temperature field, performing dynamic rapid prediction to obtain a dynamic assembly interface physical field. According to the method, high-fidelity rapid prediction of the physical field of the assembly interface under the dynamic load can be realized, the efficiency of physical field prediction is greatly improved on the premise of ensuring high precision, and rapid prediction of the physical field of the assembly interface in a dynamic service state is realized; therefore, the dynamic connection performance of high-end equipment parts such as an electronic packaging chip, an aero-engine and a multi-axis machine tool and the health condition of a system are monitored on line in real time.

Description

Technical Field

[0001] The present invention belongs to the technical field of mechanical design, and relates to a method for rapid prediction of the physical field of a mechanical equipment assembly interface under dynamic load, and specifically to a method for rapid prediction of the physical field of an assembly interface in a time series dynamic manner and a related device. Background Art

[0002] In the design and analysis of high-end mechanical equipment such as electronic packaging chips, aircraft engines, and machine tools, the physical field behavior of the assembly interface (such as temperature field, stress field, etc.) has an important impact on the performance and stability of the system. Traditional physical field analysis usually relies on high-precision numerical simulation methods, such as finite element analysis (FEA) and computational fluid dynamics (CFD). Although these methods can provide accurate physical field distribution results, when dealing with large-scale and complex assembly structures, the amount of calculation is huge, the simulation process is cumbersome, and it takes a lot of computing resources and time. Therefore, how to improve the analysis efficiency while ensuring the accuracy of the analysis has become a difficult problem that needs to be solved urgently in the current engineering technology field. Summary of the invention

[0003] The present invention aims to address the problem in the prior art that it is impossible to balance the accuracy and efficiency of the physical field analysis of the assembly interface. The present invention aims to propose a method and related device for rapid dynamic time series prediction of the physical field of the assembly interface. The prediction method dynamically predicts the physical field of the assembly interface based on the time series load, and can realize real-time monitoring of the connection performance of the assembly interface under service status.

[0004] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0005] A first aspect of the present invention provides a method for rapid dynamic prediction of assembly interface physical field time series, comprising:

[0006] Perform characteristic mapping of load and physical field on the mating surface of the assembly structure to obtain load sequence, contact pressure field sequence and temperature field sequence;

[0007] According to the first load value in the load sequence, the first contact pressure field in the contact pressure field sequence, and the first temperature field in the temperature field sequence, a transient rapid prediction of the physical field of the assembly interface is performed to obtain the transient contact pressure field and transient temperature field of the assembly interface;

[0008] According to the transient contact pressure field and transient temperature field of the assembly interface, the physical field of the assembly interface is dynamically and rapidly predicted to obtain the dynamic assembly interface physical field.

[0009] Furthermore, characteristic mapping of load and physical field is performed on the mating surface of the assembly structure to obtain a load sequence, a contact pressure field sequence and a temperature field sequence, including the following steps:

[0010] (1.1) Divide the assembly structure into finite element meshes, construct a finite element model for assembly interface analysis, and apply boundary constraints;

[0011] (1.2) Initialize the load, perform transient finite element contact analysis, and calculate the contact pressure distribution and temperature distribution in the steady state;

[0012] (1.3) Randomly adjust the load parameters at the current moment, perform transient finite element contact analysis, select the sampling step t, and calculate the contact pressure field and temperature field at the next moment;

[0013] (1.4) Select a maximum time T. If t≥T is satisfied, the obtained load, contact pressure value and temperature value are integrated into load sequence, contact pressure field sequence and temperature field sequence according to the sampling time. Otherwise, execute step (1.3);

[0014] (1.5) Select a maximum number of rounds N. If the current number of rounds n ≥ N, then the process ends. Otherwise, execute steps (1.2) to (1.4).

[0015] Further, according to the first load value in the load sequence, the first contact pressure field in the contact pressure field sequence, and the first temperature field in the temperature field sequence, a transient rapid prediction of the physical field of the assembly interface is performed to obtain the transient contact pressure field and transient temperature field of the assembly interface, including the following steps:

[0016] The first contact pressure value of the kth node in the i-th sampling round in the contact pressure field sequence, the first temperature value in the temperature field sequence and the load sequence are preprocessed as a physical field transient data set, and the preprocessed data are divided into a training set and a validation set;

[0017] The XGBoost model is trained using the training set data and the validation set data to obtain a trained XGBoost transient fast prediction model;

[0018] Input the set load and node coordinates into the trained XGBoost transient fast prediction model, predict the physical field of the transient assembly interface nodes, and obtain the contact pressure and temperature values ​​of each node on the transient assembly interface;

[0019] According to the contact pressure value, temperature value and node coordinates of each node on the transient assembly interface, the transient contact pressure field and transient temperature field of the assembly interface are obtained.

[0020] Furthermore, according to the transient contact pressure field and transient temperature field of the assembly interface, a dynamic rapid prediction of the physical field of the assembly interface is performed to obtain the dynamic physical field of the assembly interface, including the following steps:

[0021] Taking the Transformer network as the basic architecture, the Multi-DF-Trans model is established by using the image segmentation method, self-attention mechanism, interaction mechanism and decoding generation method in sequence.

[0022] Training the Multi-DF-Trans model to obtain the trained Multi-DF-Trans model;

[0023] The output of the XGBoost transient fast prediction model is used as the input of the trained Multi-DF-Trans model to establish the XG-Multi-TF model;

[0024] The dynamic temperature load sequence is input into the XG-Multi-TF model to perform dynamic and rapid prediction of the physical field of the assembly interface, and the dynamic contact pressure field and dynamic temperature field of the assembly interface are obtained.

[0025] Furthermore, the data used for training the Multi-DF-Trans model is obtained through the following process:

[0026] Integrate the load sequence, contact pressure field sequence and temperature field sequence into a physical field dynamic data set, perform preprocessing, and divide the preprocessed data into a training set and a validation set; and

[0027] The first contact pressure value of the kth node in the i-th sampling round in the contact pressure field sequence, the first temperature value in the temperature field sequence and the load sequence are preprocessed as a physical field transient data set, and the preprocessed data are divided into a training set and a validation set;

[0028] The load sequence, contact pressure field sequence and temperature field sequence are integrated into a physical field dynamic data set, including the following steps:

[0029] The contact pressure values ​​of each node at the jth sampling moment of the i-th sampling round in the contact pressure field sequence and the temperature values ​​of each node at the jth sampling moment of the i-th sampling round in the temperature field sequence are integrated into the contact pressure field and temperature field at the corresponding moment;

[0030] Integrate the contact pressure field and temperature field into contact pressure field sequence and temperature field sequence according to sampling rounds;

[0031] The contact pressure field sequence, temperature field sequence, transient contact pressure field and transient temperature field of the assembly interface and the load sequence are integrated into a physical field dynamic data set according to the sampling time.

[0032] Furthermore, the Multi-DF-Trans model includes:

[0033] The mutual attention module is used to predict the dynamic physical field and transmit the predicted dynamic physical field to the fusion output module;

[0034] A fusion output module is used to remove outliers and weightedly fuse the dynamic physical fields output by multiple mutual attention modules using the z-score method, and transmit the fused contact pressure field and dynamic temperature field to the smoothing output module;

[0035] A smoothing output module is used to smooth the fused contact pressure field and dynamic temperature field using an image filtering method, and output the final dynamic physical field;

[0036] Wherein, the mutual attention module includes:

[0037] An image segmentation module is used to segment the contact pressure field sequence and the temperature field sequence in step 1 into high-dimensional feature data and transmit them to the self-attention module;

[0038] The self-attention module is composed of multiple multi-head attention modules connected in sequence, which is used to extract the feature information within a single set of high-dimensional feature data based on the high-dimensional feature data, and output the high-dimensional feature information containing the self-attention feature to the cross-attention module;

[0039] The cross-attention module is composed of multiple multi-head attention modules connected in sequence, and is used to extract feature information between different groups of high-dimensional feature data based on high-dimensional feature information containing self-attention features, output high-dimensional information containing mutual attention information, and transmit it to the decoding generation module;

[0040] The decoding generation module is used to decode and generate images based on high-dimensional information containing mutual attention information.

[0041] Furthermore, according to the transient contact pressure field and transient temperature field of the assembly interface, a dynamic rapid prediction of the physical field of the assembly interface is performed to obtain the dynamic physical field of the assembly interface, including the following steps:

[0042] Input the current load into the trained XGBoost transient fast prediction model to obtain the transient physical field of the assembly interface at the current moment; the trained XGBoost transient fast prediction model is obtained by training the XGBoost model;

[0043] If the current moment t=1, the transient physical field of the assembly interface is used as the dynamic physical field of the assembly interface and as the sequential physical field of the assembly interface at the first moment;

[0044] If the current moment t>1, the transient physical field of the assembly interface and the sequential physical field of the assembly interface are input into the Multi-DF-Trans model to obtain the dynamic assembly interface physical field.

[0045] A second aspect of the present invention provides a fast prediction system for the time series dynamics of the physical field of an assembly interface, comprising:

[0046] A feature mapping module is used to perform feature mapping of loads and physical fields on the mating surface of the assembly structure to obtain a load sequence, a contact pressure field sequence, and a temperature field sequence;

[0047] A transient rapid prediction module is used to perform a transient rapid prediction of the physical field of the assembly interface according to the first load value in the load sequence, the first contact pressure field in the contact pressure field sequence, and the first temperature field in the temperature field sequence, so as to obtain the transient contact pressure field and transient temperature field of the assembly interface;

[0048] The dynamic rapid prediction module is used to perform dynamic rapid prediction of the physical field of the assembly interface according to the transient contact pressure field and transient temperature field of the assembly interface to obtain the dynamic assembly interface physical field.

[0049] The third aspect of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method for rapid dynamic prediction of the assembly interface physical field timing series when executing the computer program.

[0050] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for rapid dynamic prediction of the assembly interface physical field timing series is implemented.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] In the present invention, the characteristic mapping step of the assembly interface load-physical field outputs a characteristic information sequence of the physical field (such as temperature field, stress field, etc.) covering the assembly interface load domain through a parametric finite element analysis method. The present invention can achieve high-fidelity prediction of the assembly interface physical field under dynamic loads through the proposed assembly interface physical field time series dynamic rapid prediction method, solving the problem that the assembly interface physical field cannot be measured or measured accurately. And it can greatly improve the efficiency of physical field prediction under the premise of ensuring high precision, solve the problem that the assembly interface physical field cannot be measured quickly, and realize the rapid prediction of the assembly interface physical field under dynamic service state, so as to achieve the purpose of real-time online monitoring of the dynamic connection performance and system health status of high-end equipment parts such as electronic packaging chips, aircraft engines, and multi-axis machine tools.

[0053] Furthermore, machine learning and deep learning technologies based on image segmentation methods, self-attention mechanisms, interaction force mechanisms and decoding generation methods are used to build a Multi-DF-Trans model to explore the intrinsic connection between the dynamic service state physical field output by the finite element software and the dynamic load sequence. This can realize online monitoring of the physical field of the assembly interface under the dynamic service state, and the training and prediction efficiency of deep learning technology is much higher than that of traditional finite element calculations. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a schematic diagram of the structure of a flip-chip package;

[0055] Figure 2 It is a detailed finite element model of a flip-chip package;

[0056] Figure 3 It is the distribution diagram of the temperature of the flip-chip package;

[0057] Figure 4 It is the structural diagram of the time series dynamic rapid prediction model;

[0058] Figure 5 It is the structural diagram of the mutual attention module;

[0059] Figure 6 It is the prediction diagram and prediction accuracy of the dynamic assembly interface temperature field; among them, (a) is the preventive distribution, (b) is the true distribution, (c) is the relative error, and (d) is the absolute error;

[0060] Figure 7 It is a flow chart of a method for rapid prediction of the time series dynamics of the physical field of an assembly interface;

[0061] Figure 8 It is a flow chart of the fast prediction system of the time series dynamics of the physical field of the assembly interface;

[0062] Among them, 1 is a heat dissipation cover, 2 is a heat conductive sheet, 3 is a chip operator, 4 is a solder pad, 5 is a filler, 6 is a base, 7 is an upper assembly interface, and 8 is an assembly interface. DETAILED DESCRIPTION

[0063] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. The preferred embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thoroughly and comprehensively understood.

[0064] The present invention proposes a dynamic rapid prediction method for the physical field of the assembly interface facing the time series load based on the direct prediction of the physical field of the assembly interface from the dynamic load. The dynamic load is the time-varying non-periodic temperature load, vibration load, impact load and their combined effects. The physical field is the temperature field and the stress field.

[0065] See also Figure 7 The present invention provides a time series dynamic rapid prediction method for the physical field of an assembly interface, which realizes the dynamic prediction of the physical quantity of the assembly interface, including the characteristic mapping of the dynamic load-assembly interface physical field, the transient rapid prediction of the physical field of the assembly interface, and the dynamic rapid prediction steps of the physical field of the assembly interface, which are specifically as follows:

[0066] 1. Perform load-physical field feature mapping on the mating surface of the assembly structure to obtain a load sequence, a contact pressure field sequence, and a temperature field sequence. The steps include:

[0067] (1.1) Divide the assembly structure into finite element meshes, construct a finite element model for assembly interface analysis, and apply boundary constraints;

[0068] (1.2) Initialize the load, perform transient finite element contact analysis, and calculate the contact pressure distribution and temperature distribution in the steady state;

[0069] (1.3) Randomly adjust the load parameters at the current moment, perform transient finite element contact analysis, select the sampling step t, and calculate the contact pressure field and temperature field at the next moment;

[0070] (1.4) Select a maximum time T. If t≥T is satisfied, the obtained load, contact pressure value and temperature value are integrated into load sequence, contact pressure field sequence and temperature field sequence according to the sampling time. Otherwise, execute step (1.3);

[0071] (1.5) Select a maximum number of rounds N. If the current number of rounds n ≥ N, then the process ends. Otherwise, execute steps (1.2) to (1.4).

[0072] The step of performing load-physical field feature mapping on the mating surface of the assembly structure can expand the data set to the maximum extent in the load domain and improve the generalization ability of the model.

[0073] Second, according to the first load value in the load sequence, the first contact pressure field in the contact pressure field sequence, and the first temperature field in the temperature field sequence, a transient rapid prediction of the physical field of the assembly interface is performed to obtain a transient contact pressure field and a transient temperature field of the assembly interface, the steps comprising:

[0074] (2.1) The first contact pressure value σ of the kth node in the i-th sampling round in the contact pressure field sequence is i,1,k , the first temperature value T in the temperature field sequencei,1,k The load sequence is preprocessed as a physical field transient data set, and the preprocessed data is divided into a training set and a validation set;

[0075] (2.2) Using the XGBoost model as the basic method, the training set data and the validation set data in step (2.1) are used for training to obtain a trained XGBoost transient fast prediction model;

[0076] (2.3) Inputting the set load and node coordinates into the trained XGBoost transient fast prediction model, predicting the physical field of the transient assembly interface node, and obtaining the contact pressure value and temperature value of each node of the transient assembly interface;

[0077] (2.4) Using the contact pressure value, temperature value and node coordinates of each node of the transient assembly interface, the transient contact pressure field {σ i,temp} and transient temperature field {T i,temp}.

[0078] Among them, the training parameters of the XGBoost transient fast prediction model are:

[0079] Maximum decision depth D max =200;

[0080] Iteration learning rate lr = 0.05;

[0081] Maximum number of iterations E max =100.

[0082] 3. According to the transient contact pressure field and transient temperature field of the assembly interface, a dynamic rapid prediction of the physical field of the assembly interface is performed to obtain the dynamic physical field of the assembly interface. The steps include:

[0083] (3.1) Integrate the load sequence, contact pressure field sequence and temperature field sequence into a physical field dynamic data set and preprocess it, and divide the preprocessed data into a training set and a validation set;

[0084] (3.2) Taking the Transformer network as the basic architecture, the Multi-DF-Trans model is established by sequentially using the image segmentation method, self-attention mechanism, interaction mechanism and decoding generation method;

[0085] (3.3) using the training set and the validation set divided after the preprocessing of the physical field dynamic data set in step (3.1) and the training set and the validation set divided after the preprocessing of the physical field transient data set in step (2.1) to train the Multi-DF-Trans model, and obtain the trained Multi-DF-Trans model;

[0086] (3.4) Using the output of the XGBoost transient fast prediction model in step (2.3) as the input of the trained Multi-DF-Trans model to establish an XG-Multi-TF model;

[0087] (3.5) The set dynamic temperature load sequence is input into the XG-Multi-TF model to perform dynamic and rapid prediction of the physical field of the assembly interface to obtain the dynamic assembly interface physical field.

[0088] Among them, the step (3.1) integrates the load sequence, contact pressure field sequence and temperature field sequence into a physical field dynamic data set and preprocesses it, and divides the preprocessed data into a training set and a verification set. The specific process is:

[0089] (3.1.1) The contact pressure values ​​of each node at the jth sampling moment of the i-th sampling round in the contact pressure field sequence and the temperature values ​​of each node at the jth sampling moment of the i-th sampling round in the temperature field sequence are integrated into the contact pressure field σ at the corresponding moment. i,j,dist and temperature field T i,j,dist ;

[0090] (3.1.2) The contact pressure field σ i,j,dist and temperature field T i,j,dist According to the sampling rounds, it is integrated into the contact pressure field sequence {σ i,dist} and the temperature field sequence {T i,dist};

[0091] (3.1.3) The contact pressure field sequence {σ i,dist}, the temperature field sequence {T i,dist}, the transient contact pressure field {σ i,temp}, the transient temperature field {T i,temp} and load sequence {P i}Integrate into physical field dynamic data sets according to sampling time.

[0092] The Multi-DF-Trans model includes:

[0093] The mutual attention module is used to predict the dynamic physical field and transmit the predicted dynamic physical field to the fusion output module;

[0094] The fusion output module is used to remove outliers and perform weighted fusion on the dynamic physical fields output by multiple mutual attention modules using the z-score method, and transmit the fused contact pressure field and dynamic temperature field to the smoothing output module; the outlier removal and weighted fusion process of a certain node is performed using the following formula:

[0095]

[0096]

[0097] In the formula, μ i,j represents the mean value of the physical field value of the node, N represents the number of mutual attention modules, k represents the sequence number of the mutual attention module, i represents the row number, j represents the column number, σ i,j Indicates the standard deviation of the node physical field value, Y k represents the dynamic physical field output by the kth mutual attention module, Y k [i,j] represents the physical field value at the node [i,j] in the physical field, V k,i,j Indicates whether the output of a certain interaction force module at this node is an abnormal value, Y valid,i,j Represents the output after outliers are removed and weighted fusion.

[0098] The smoothing output module is used to smooth the fused contact pressure field and dynamic temperature field using the image filtering method according to the following formula, and output the final dynamic assembly interface physical field.

[0099]

[0100] Where Y Conv (x, y) represents the smoothed dynamic physical field, K represents the convolution kernel, K(i, j) represents the weight value of the convolution kernel, and k w represents the width of the convolution kernel, k w Represents the height of the convolution kernel.

[0101] The mutual attention module comprises:

[0102] An image segmentation module is used to segment the contact pressure field sequence and the temperature field sequence in step 1 into high-dimensional feature data and transmit them to the self-attention module;

[0103] A multi-head attention module is used to extract feature information of high-dimensional feature data, including a multi-head attention layer, a normalization layer and a multi-layer perception layer. The multi-head attention layer is used to extract the sensitivity characteristics of high-dimensional features, the normalization layer is used to accelerate the convergence of the network, and the multi-layer perception layer is used for nonlinear transformation and feature extraction of high-dimensional vectors;

[0104] The self-attention module is composed of multiple multi-head attention modules connected in sequence, which is used to extract the feature information within a single set of high-dimensional feature data based on the high-dimensional feature data, and output the high-dimensional feature information containing the self-attention feature to the cross-attention module;

[0105] The cross-attention module is composed of multiple multi-head attention modules connected in sequence, and is used to extract feature information between different groups of high-dimensional feature data based on high-dimensional feature information containing self-attention features, output high-dimensional information containing mutual attention information, and transmit it to the decoding generation module;

[0106] The decoding generation module is used to decode and generate images based on high-dimensional information containing mutual attention information.

[0107] The multi-head attention module in the XG-Multi-TF model can capture the relationship between sequence information and is suitable for feature extraction of complex information; and the multi-path parallel approach can avoid overfitting or local optimal solutions that may be caused by single-head attention calculation, thereby improving the stability and robustness of the model.

[0108] The dynamic and rapid prediction of the physical field of the assembly interface includes the following steps:

[0109] (3.4.1) Input the current load into the XGBoost transient fast prediction model trained in step (2.2) to obtain the transient physical field of the assembly interface at the current moment;

[0110] (3.4.2) If the current moment t = 1, the transient physical field of the assembly interface at the current moment is taken as the dynamic physical field of the assembly interface and as the sequential physical field of the assembly interface at the first moment;

[0111] (3.4.3) If t>1 at the current moment, the transient physical field of the assembly interface and the sequential physical field of the assembly interface are input into the Multi-DF-Trans model in (3.2) to obtain the dynamic physical field of the assembly interface at the current moment, and added to the sequential physical field of the assembly interface to obtain the dynamic assembly interface physical field.

[0112] Flip-chip packaging (such as Figure 1 The beneficial effects of the present invention are further illustrated by taking the assembly structure of the present invention and the random time-series temperature load as an example.

[0113] See also Figure 1 and Figure 2 The assembly interface may be the upper assembly interface 7 or the lower assembly interface 8 .

[0114] The cross-sectional view of the quarter finite element model of the flip-chip package structure is shown in Figure 1. Figure 2 As shown, the assembly interface between the heat dissipation cover plate 1 and the heat conducting sheet 2 (upper assembly interface 7 and lower assembly interface 8) is selected as the physical field prediction object.

[0115] The initial ambient temperature of the structure is 20°C. Chip operator 3 is set as the heat source, and the temperature load range is 20-180°C. The overall structure temperature distribution under a certain temperature load is as follows: Figure 3 shown.

[0116] The structure of the XG-Multi-TF model is as follows Figure 4 As shown in Figure 2. The structure of the Multi-DF-Trans model is as follows: Figure 5 As shown in Figure 2, the time series temperature load is input into the time series dynamic rapid prediction model, and the dynamic temperature field of the assembly interface at the corresponding moment is output.

[0117] The comparison between the dynamic temperature field output by the prediction model and the corresponding temperature field in the data set is as follows: Figure 6 As shown in the figure, it can be seen that the fast prediction method of assembly interface physical field time series dynamics can effectively realize the fast prediction of assembly interface physical field under dynamic service state. 2 As the accuracy evaluation standard, the calculation formula is as follows:

[0118]

[0119] In the formula, y i Represents the predicted value of a pixel. Represents the true value of a pixel. Represents the mean of the true values.

[0120] The dynamic service state prediction accuracy of the present invention is 99.03%, and the maximum difference between the predicted value and the true value is 0.51°C, which fully meets the accuracy requirements in practical engineering applications. The average prediction time of this method under a single dynamic load is 10.18s, and the corresponding average simulation calculation time of a single load step for the same task under the same computer conditions is 262.6s, and the average prediction efficiency is improved by 96.12%.

[0121] See also Figure 8 Another embodiment of the present invention provides a fast prediction system for the timing dynamics of the physical field of an assembly interface, comprising:

[0122] A feature mapping module is used to perform feature mapping of loads and physical fields on the mating surface of the assembly structure to obtain a load sequence, a contact pressure field sequence, and a temperature field sequence;

[0123] A transient rapid prediction module is used to perform a transient rapid prediction of the physical field of the assembly interface according to the first load value in the load sequence, the first contact pressure field in the contact pressure field sequence, and the first temperature field in the temperature field sequence, so as to obtain the transient contact pressure field and transient temperature field of the assembly interface;

[0124] The dynamic rapid prediction module is used to perform dynamic rapid prediction of the physical field of the assembly interface according to the transient contact pressure field and transient temperature field of the assembly interface to obtain the dynamic assembly interface physical field.

[0125] Another embodiment of the present invention provides a computer program including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for rapid dynamic prediction of the assembly interface physical field timing sequence is implemented.

[0126] Another embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for rapid dynamic prediction of the assembly interface physical field timing series is implemented.

[0127] The present invention rapidly predicts the dynamic physical field of an assembly interface based on a dynamic service load, thereby realizing rapid dynamic monitoring of the physical field of an assembly interface of mechanical equipment.

[0128] The present invention adopts machine learning and deep learning technologies to build the XG-Multi-TF model to explore the intrinsic connection between the dynamic service state physical field output by the finite element software and the dynamic load sequence. It can realize the online monitoring of the physical field of the assembly interface under the dynamic service state, and the training and prediction efficiency of deep learning technology is much higher than that of traditional finite element calculations.

[0129] The present invention can achieve high-fidelity rapid prediction of the physical field of the assembly interface under dynamic load, can greatly improve the efficiency of physical field prediction while ensuring high precision, and achieve rapid prediction of the physical field of the assembly interface under dynamic service status, thereby achieving real-time online monitoring of the dynamic connection performance and system health status of high-end equipment components such as electronic packaging chips, aircraft engines, and multi-axis machine tools.

[0130] The above description is only for the best embodiment of the present invention, but it should not be understood as limiting the claims. The present invention is not limited to the above embodiments, and its specific structure is allowed to be changed. However, all changes made within the protection scope of the independent claims of the present invention are within the protection scope of the present invention.

[0131] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

Claims

1. A method for rapid prediction of assembly interface physical field time series dynamics, characterized in that: include: Perform characteristic mapping of load and physical field on the mating surface of the assembly structure to obtain load sequence, contact pressure field sequence and temperature field sequence; According to the first load value in the load sequence, the first contact pressure field in the contact pressure field sequence, and the first temperature field in the temperature field sequence, a transient rapid prediction of the physical field of the assembly interface is performed to obtain the transient contact pressure field and transient temperature field of the assembly interface; According to the transient contact pressure field and transient temperature field of the assembly interface, the physical field of the assembly interface is dynamically and rapidly predicted to obtain the dynamic assembly interface physical field.

2. The assembly interface physical field time series dynamic rapid prediction method according to claim 1 is characterized in that: The characteristic mapping of load and physical field is performed on the mating surface of the assembly structure to obtain the load sequence, contact pressure field sequence and temperature field sequence, including the following steps: (1.1) Divide the assembly structure into finite element meshes, construct a finite element model for assembly interface analysis, and apply boundary constraints; (1.2) Initialize the load, perform transient finite element contact analysis, and calculate the contact pressure distribution and temperature distribution in the steady state; (1.3) Randomly adjust the load parameters at the current moment, perform transient finite element contact analysis, select the sampling step t, and calculate the contact pressure field and temperature field at the next moment; (1.4) Select a maximum time T. If t≥T is satisfied, the obtained load, contact pressure value and temperature value are integrated into load sequence, contact pressure field sequence and temperature field sequence according to the sampling time. Otherwise, execute step (1.3); (1.5) Select a maximum number of rounds N. If the current number of rounds n ≥ N, then the process ends. Otherwise, execute steps (1.2) to (1.4).

3. The assembly interface physical field time series dynamic rapid prediction method according to claim 1 is characterized in that: According to the first load value in the load sequence, the first contact pressure field in the contact pressure field sequence, and the first temperature field in the temperature field sequence, a transient rapid prediction of the physical field of the assembly interface is performed to obtain the transient contact pressure field and transient temperature field of the assembly interface, including the following steps: The first contact pressure value of the kth node in the i-th sampling round in the contact pressure field sequence, the first temperature value in the temperature field sequence and the load sequence are preprocessed as a physical field transient data set, and the preprocessed data are divided into a training set and a validation set; The XGBoost model is trained using the training set data and the validation set data to obtain a trained XGBoost transient fast prediction model; Input the set load and node coordinates into the trained XGBoost transient fast prediction model, predict the physical field of the transient assembly interface nodes, and obtain the contact pressure and temperature values ​​of each node on the transient assembly interface; According to the contact pressure value, temperature value and node coordinates of each node on the transient assembly interface, the transient contact pressure field and transient temperature field of the assembly interface are obtained.

4. The assembly interface physical field time series dynamic rapid prediction method according to claim 1 is characterized in that: According to the transient contact pressure field and transient temperature field of the assembly interface, a dynamic rapid prediction of the physical field of the assembly interface is performed to obtain the dynamic assembly interface physical field, including the following steps: Taking the Transformer network as the basic architecture, the Multi-DF-Trans model is established by using the image segmentation method, self-attention mechanism, interaction mechanism and decoding generation method in sequence. Training the Multi-DF-Trans model to obtain the trained Multi-DF-Trans model; The output of the XGBoost transient fast prediction model is used as the input of the trained Multi-DF-Trans model to establish the XG-Multi-TF model; The dynamic temperature load sequence is input into the XG-Multi-TF model to perform dynamic and rapid prediction of the physical field of the assembly interface, and the dynamic contact pressure field and dynamic temperature field of the assembly interface are obtained.

5. The method for rapid dynamic prediction of assembly interface physical field time series according to claim 4, characterized in that: The data used to train the Multi-DF-Trans model is obtained through the following process: The load sequence, contact pressure field sequence and temperature field sequence are integrated into a physical field dynamic data set, preprocessed, and the preprocessed data is divided into a training set and a validation set; as well as The first contact pressure value of the kth node in the i-th sampling round in the contact pressure field sequence, the first temperature value in the temperature field sequence and the load sequence are preprocessed as a physical field transient data set, and the preprocessed data are divided into a training set and a validation set; The load sequence, contact pressure field sequence and temperature field sequence are integrated into a physical field dynamic data set, including the following steps: The contact pressure values ​​of each node at the jth sampling moment of the i-th sampling round in the contact pressure field sequence and the temperature values ​​of each node at the jth sampling moment of the i-th sampling round in the temperature field sequence are integrated into the contact pressure field and temperature field at the corresponding moment; Integrate the contact pressure field and temperature field into contact pressure field sequence and temperature field sequence according to sampling rounds; The contact pressure field sequence, temperature field sequence, transient contact pressure field and transient temperature field of the assembly interface and the load sequence are integrated into a physical field dynamic data set according to the sampling time.

6. The method for rapid dynamic prediction of assembly interface physical field time series according to claim 4, characterized in that: The Multi-DF-Trans model includes: The mutual attention module is used to predict the dynamic physical field and transmit the predicted dynamic physical field to the fusion output module; A fusion output module is used to remove outliers and weightedly fuse the dynamic physical fields output by multiple mutual attention modules using the z-score method, and transmit the fused contact pressure field and dynamic temperature field to the smoothing output module; A smoothing output module is used to smooth the fused contact pressure field and dynamic temperature field using an image filtering method, and output the final dynamic physical field; Wherein, the mutual attention module includes: An image segmentation module is used to segment the contact pressure field sequence and the temperature field sequence in step 1 into high-dimensional feature data and transmit them to the self-attention module; The self-attention module is composed of multiple multi-head attention modules connected in sequence, which is used to extract the feature information within a single set of high-dimensional feature data based on the high-dimensional feature data, and output the high-dimensional feature information containing the self-attention feature to the cross-attention module; The cross-attention module is composed of multiple multi-head attention modules connected in sequence, and is used to extract feature information between different groups of high-dimensional feature data based on high-dimensional feature information containing self-attention features, output high-dimensional information containing mutual attention information, and transmit it to the decoding generation module; The decoding generation module is used to decode and generate images based on high-dimensional information containing mutual attention information.

7. The method for rapid dynamic prediction of assembly interface physical field time series according to claim 1, characterized in that: According to the transient contact pressure field and transient temperature field of the assembly interface, a dynamic rapid prediction of the physical field of the assembly interface is performed to obtain the dynamic assembly interface physical field, including the following steps: Input the current load into the trained XGBoost transient fast prediction model to obtain the transient physical field of the assembly interface at the current moment; the trained XGBoost transient fast prediction model is obtained by training the XGBoost model; If the current moment t=1, the transient physical field of the assembly interface is used as the dynamic physical field of the assembly interface and as the sequential physical field of the assembly interface at the first moment; If the current moment t>1, the transient physical field of the assembly interface and the sequential physical field of the assembly interface are input into the Multi-DF-Trans model to obtain the dynamic assembly interface physical field.

8. A fast prediction system for the time series dynamics of the physical field of an assembly interface, characterized in that: include: A feature mapping module is used to perform feature mapping of loads and physical fields on the mating surface of the assembly structure to obtain a load sequence, a contact pressure field sequence, and a temperature field sequence; A transient rapid prediction module is used to perform a transient rapid prediction of the physical field of the assembly interface according to the first load value in the load sequence, the first contact pressure field in the contact pressure field sequence, and the first temperature field in the temperature field sequence, so as to obtain the transient contact pressure field and transient temperature field of the assembly interface; The dynamic rapid prediction module is used to perform dynamic rapid prediction of the physical field of the assembly interface according to the transient contact pressure field and transient temperature field of the assembly interface to obtain the dynamic assembly interface physical field.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method for rapid dynamic prediction of the assembly interface physical field time series as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for rapid dynamic prediction of the assembly interface physical field time series as described in any one of claims 1 to 7 is implemented.

Images