RF front-end module thermal stress analysis method and system

Abstract

The present invention relates to the technical field of radio frequency front-end modules, and discloses a method and system for thermal stress analysis of radio frequency front-end modules. The method comprises: collecting temperature data, stress data, deformation data and heat distribution data of the radio frequency front-end module, and performing synchronous arrangement of timestamps to obtain a thermal stress monitoring data sequence; establishing a transfer equation containing thermal conductivity, density and heat capacity parameters of each microelement to obtain an initial thermal stress analysis model; establishing a thermal stress transfer equation and boundary constraints between adjacent microelement bodies, and solving to obtain a temperature field distribution matrix; calculating the normal stress and tangential stress of each microelement, and solving to obtain a stress field characteristic vector; inputting the stress field characteristic vector into a thermal stress time evolution model, mapping the feature to a latent space through an encoder in the thermal stress time evolution model and reconstructing it through a decoder, and outputting thermal stress distribution prediction data. The present invention improves the accuracy of thermal stress analysis of the radio frequency front-end module.

Description

Technical Field

[0001] The present invention relates to the technical field of radio frequency front-end modules, and in particular to a method and system for analyzing thermal stress of a radio frequency front-end module. Background Art

[0002] As the core component of wireless communication equipment, the performance and reliability of the RF front-end module directly affect the stable operation of the communication system. In high-frequency working environments, the RF front-end module faces complex thermal stress problems, which are mainly caused by the combined effects of working power, environmental conditions and structural constraints. Existing thermal stress analysis methods mainly focus on the optimization of module structure design parameters, but often ignore the adaptability of these parameters under different working conditions and ambient temperature changes.

[0003] In a wide frequency band, the thermal stress distribution of RF front-end modules shows significant nonlinear and time-varying characteristics. Traditional thermal stress analysis methods require over-discretization to reduce the estimation errors caused by assumptions and inappropriate boundary condition definitions caused by concentrated stress. This approach not only increases the computational burden, but also makes it difficult to accurately describe the spatial stress changes of complex RF structures. Summary of the invention

[0004] The present invention provides a method and system for analyzing thermal stress of a radio frequency front-end module, and the present invention improves the accuracy of thermal stress analysis of the radio frequency front-end module.

[0005] In a first aspect, the present invention provides a method for analyzing thermal stress of a radio frequency front-end module, the method comprising:

[0006] Collect the temperature data, stress data, deformation data and thermal distribution data of the RF front-end module, and synchronize and sort them with the timestamp to obtain the thermal stress monitoring data sequence;

[0007] Determine the spatial discretization parameters according to the thermal stress monitoring data sequence, divide the RF front-end module into N×M×L micro-elements, establish a transfer equation including the thermal conductivity, density and heat capacity parameters of each micro-element, and obtain an initial thermal stress analysis model;

[0008] Based on the distribution of stress compensation coefficients for each micro-element in the initial thermal stress analysis model, a thermal stress transfer equation and boundary constraints between adjacent micro-elements are established to obtain a temperature field distribution matrix;

[0009] Calculating the normal stress and tangential stress of each microelement according to the temperature field distribution matrix, establishing a high-order lumped parameter equation group, and solving to obtain the stress field eigenvector;

[0010] The stress field feature vector is input into the thermal stress time evolution model, the feature is mapped to the latent space by the encoder in the thermal stress time evolution model and reconstructed by the decoder, and the thermal stress distribution prediction data is output.

[0011] In a second aspect, the present invention provides a radio frequency front-end module thermal stress analysis system, the radio frequency front-end module thermal stress analysis system comprising:

[0012] The acquisition module is used to collect the temperature data, stress data, deformation data and thermal distribution data of the RF front-end module, and synchronize and sort the data with the timestamp to obtain the thermal stress monitoring data sequence;

[0013] Establish a module for determining the spatial discretization parameters according to the thermal stress monitoring data sequence, and dividing the RF front-end module into N×M×L micro-elements, establishing a transfer equation including the thermal conductivity, density and heat capacity parameters of each micro-element, and obtaining an initial thermal stress analysis model;

[0014] An allocation module is used to allocate stress compensation coefficients based on each microelement in the initial thermal stress analysis model, establish thermal stress transfer equations and boundary constraints between adjacent microelements, and solve to obtain a temperature field distribution matrix;

[0015] A solution module, used for calculating the normal stress and tangential stress of each microelement according to the temperature field distribution matrix, and establishing a high-order lumped parameter equation group to solve and obtain the stress field characteristic vector;

[0016] The output module is used to input the stress field feature vector into the thermal stress time evolution model, map the feature to the latent space through the encoder in the thermal stress time evolution model and reconstruct it through the decoder, and output the thermal stress distribution prediction data.

[0017] The third aspect of the present invention provides a computer device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the computer device executes the above-mentioned RF front-end module thermal stress analysis method.

[0018] A fourth aspect of the present invention provides a computer-readable storage medium, in which instructions are stored. When the computer-readable storage medium is run on a computer, the computer executes the above-mentioned RF front-end module thermal stress analysis method.

[0019] In the technical solution provided by the present invention, by establishing a multi-heat conduction path model and introducing a stress compensation coefficient, the problem of spatial stress variation in complex RF structures is solved, making the thermal stress analysis results more accurate; a high-order lumped parameter thermal stress model is adopted to avoid the computational burden caused by excessive discretization, while ensuring the estimation accuracy at different discretization levels; through the intelligent prediction method of a multi-scale inversion transform network, the thermal stress monitoring data with interference, multi-modal and large time span characteristics is effectively processed, and the prediction accuracy is improved; combined with the POI substrate material and the low-loss high-power electrode solution, an accurate description of the distributed heat source inside the module is achieved, ensuring the continuity of stress between adjacent discrete components; a complete thermal stress prediction model is established, which is suitable for thermal stress analysis of RF devices such as TF-SAW, filters, and duplexers, and has a good prediction effect under high-power working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0021] Figure 1 A schematic diagram of the steps of a method for analyzing thermal stress of a radio frequency front-end module in an embodiment of the present invention;

[0022] Figure 2 It is a schematic diagram of the structure of a radio frequency front-end module thermal stress analysis system in an embodiment of the present invention;

[0023] Figure 3 It is a schematic block diagram of the structure of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0024] Embodiments of the present invention provide a method and system for thermal stress analysis of a radio frequency front-end module. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0025] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of a method for analyzing thermal stress of a radio frequency front-end module in an embodiment of the present invention includes:

[0026] Step S1, collecting temperature data, stress data, deformation data and thermal distribution data of the RF front-end module, and arranging them synchronously with the timestamp to obtain a thermal stress monitoring data sequence;

[0027] It is understandable that the execution subject of the present invention may be a radio frequency front-end module thermal stress analysis system, or a terminal or a server, which is not specifically limited here. The embodiment of the present invention is described by taking a server as an execution subject as an example.

[0028] Specifically, the operating temperature of the RF front-end module in the frequency band range of 20MHz to 40GHz is monitored in real time. A temperature monitoring data set is obtained under multiple frequency band conditions through a high-precision temperature sensor. At the same time, in order to understand the stress distribution inside the module, especially the changes in the stress concentration area, stress monitoring data are collected in these areas. The stress state of the RF front-end module is detected in real time through devices such as pressure sensors or strain gauges. A thermal image sensor is aimed at the surface of the RF front-end module, and the heat distribution of its surface is recorded in real time to generate a heat distribution monitoring data set. This process can reflect the thermal stress phenomenon caused by uneven heat in different areas during the operation of the RF front-end module. In the thermal stress analysis, the deformation of the RF front-end module is dynamically monitored. Through devices such as laser displacement sensors or displacement sensors, the deformation of the module under different loads and temperature changes is monitored in real time. The sampling frequency is set according to the temperature monitoring data set, stress monitoring data set, heat distribution monitoring data set and deformation monitoring data set, and the original sampling data is obtained by continuous sampling. The sampling frequency is set according to the response speed and measurement accuracy requirements of the equipment, and a higher sampling frequency is selected to capture the rapidly changing temperature, stress, deformation and heat distribution information. The collected raw data contains noise and unnecessary frequency components. Signal processing is performed on these data, and bandpass filtering technology is applied to remove irrelevant frequency components to ensure that only the signal part of interest is in the data. Through amplitude normalization processing, all data are standardized to the same range, eliminating the influence of different sensor ranges and data units, so that various types of data can be compared and analyzed on the same scale to obtain a preprocessed data set. Data alignment is performed on the preprocessed data set, and data sets from different sources are synchronized to the same time point to form a multidimensional synchronized data matrix. The multidimensional synchronized data matrix is ​​processed by principal component analysis. Principal component analysis is a statistical technique for dimensionality reduction and feature extraction. The most important components in the data are extracted by eigenvalue decomposition method. Through principal component analysis, high-dimensional data is reduced to lower dimensions while retaining key information in the data, and target features of major physical fields such as temperature field, stress field, deformation field and heat distribution field are extracted. These target features can effectively describe the thermal stress performance of the RF front-end module under different working conditions and construct a thermal stress monitoring data sequence.

[0029] Step S2, determining the spatial discretization parameters according to the thermal stress monitoring data sequence, and dividing the RF front-end module into N×M×L micro-elements, establishing a transfer equation including the thermal conductivity, density and heat capacity parameters of each micro-element, and obtaining an initial thermal stress analysis model;

[0030] Specifically, the spatial distribution analysis is performed based on the temperature field data in the thermal stress monitoring data sequence. By analyzing the rate of change of temperature at different positions, the main trend of heat transfer in the RF front-end module is determined. By calculating the change of temperature gradient, the minimum spatial characteristic scale in the module is derived. This scale determines the accuracy of spatial discretization and the grid division standard. According to the minimum spatial characteristic scale, the benchmark data for grid division is set. According to the grid division benchmark data, the RF front-end module is divided into N×M×L micro-elements in the three directions of X, Y, and Z. These micro-elements are the most basic calculation units in thermal stress analysis. The morphology, thermodynamic properties and boundary conditions of each micro-element will affect the final analysis results. In this process, the geometric boundary coordinates of each micro-element are determined to ensure that the discretization process conforms to the actual structural characteristics of the RF front-end module and form an accurate spatial discretization structure. After constructing the spatial discretization structure of the micro-element, the material properties of different devices in the RF front-end module are calibrated, especially important devices such as TF-SAW (surface acoustic wave device), filter and duplexer. According to the dielectric constant and thermal conductivity of POI (lithium niobate) substrate materials, a material database is established to obtain the material mapping relationship of different devices. The thermal conductivity of each microelement is assigned based on the device material mapping relationship. The assignment of thermal conductivity depends on the physical properties of the device material, especially for the materials of different parts of the RF front-end module, the thermal conductivity will be different. By combining the material mapping relationship, an appropriate thermal conductivity value is assigned to each microelement to form a thermal conductivity distribution matrix. For the microelement located at the boundary of the module, its thermal conductivity is determined according to the parameters of the low-loss high-power electrode material, and the thermal conductivity of the boundary microelement is obtained. The density parameters of the microelement are spatially allocated. Density is a physical quantity that describes the mass per unit volume of a substance. In thermal stress analysis, density is closely related to heat capacity and directly affects the calculation of thermal conduction and thermal stress. In order to reasonably allocate density, the density distribution function is calculated according to the mass-volume relationship of each RF device. Through calculation, the density value of each microelement is obtained to form a density distribution matrix. Based on the thermal conductivity distribution matrix and the density distribution matrix, the heat capacity parameters of each microelement are calculated. Heat capacity is the ability of a substance to store thermal energy, and its value is closely related to density and specific heat capacity. By establishing a temperature-heat capacity mapping function, the heat capacity distribution matrix is ​​obtained. According to the thermal conductivity distribution matrix, density distribution matrix and heat capacity distribution matrix, a group of heat conduction equations for each microelement is constructed to describe the law of heat transfer within the module. At the same time, boundary conditions are set for the conduction equation, such as the heat exchange conditions between the module and the external environment, thermal coupling between different components, etc. The heat conduction equation system is subjected to thermomechanical coupling analysis. By establishing the correlation equation between the temperature field and the displacement field, the relationship between the deformation caused by temperature and the thermal stress is reflected. By solving these coupling equations, the initial thermal stress analysis model is obtained, which contains information such as temperature distribution, stress distribution and deformation caused by temperature changes.

[0031] Step S3, allocating stress compensation coefficients to each microelement in the initial thermal stress analysis model, establishing thermal stress transfer equations and boundary constraints between adjacent microelements, and solving to obtain a temperature field distribution matrix;

[0032] Specifically, the stress characteristic analysis is performed on the initial thermal stress analysis model. By analyzing the geometric structure and material properties of the device, the stress response data of each microelement is determined. When performing the stress characteristic analysis, the complex structure of the RF front-end module is considered, including the shape, size, material type and other factors of different components inside it, which directly affect the distribution and transmission of stress. Based on the stress response data, a thermodynamic analysis is performed on each microelement. Thermodynamic analysis not only focuses on the influence of temperature on the internal stress state of the module, but also takes into account the thermophysical properties of the material, especially the material parameters of the POI substrate material and the low-loss high-power electrode. On this basis, a calculation model for the stress compensation coefficient is established. The model obtains the initial stress compensation coefficient of each microelement by modeling the interaction between temperature change and material properties, and describes the correction amount required for different microelement due to thermal expansion and other effects. Based on the initial compensation coefficient, a stress compensation matrix is ​​constructed to describe the mutual influence between each microelement, especially the stress compensation relationship caused by temperature change. In order to accurately reflect the stress distribution under different working conditions, the stress distribution under the preset working state is corrected and calculated, and the correction compensation coefficient matrix is ​​obtained according to the initial compensation coefficient of each microelement and the specific working conditions. The modified compensation coefficient matrix is ​​input into the thermal stress transfer model between adjacent micro-elements. The stress transfer function is established by the principle of stress continuity to describe how stress between different micro-elements is transferred as the temperature changes. This transfer function helps to describe the propagation of thermal stress inside the module. By establishing the thermal stress transfer relationship, the interaction mode of stress between different micro-elements is clarified. The temperature boundary in the thermal stress transfer relationship is constrained. Since the module will be affected by the ambient temperature or external factors during operation, the boundary constraint conditions are set according to the thermal expansion characteristics within the target temperature range. The boundary constraint conditions are obtained by analyzing the thermal expansion coefficient of the material and the temperature change in the working environment, thereby limiting the displacement and stress of the micro-element on the boundary, making the thermal stress calculation more in line with the actual situation. Based on the thermal stress transfer relationship and boundary constraint conditions, a group of thermal stress control equations is constructed. This group of equations combines the thermophysical properties, stress compensation coefficients and heat transfer boundary conditions of each micro-element to provide a complete mathematical framework for the subsequent stress field solution. By solving these equations, the stress distribution function of each micro-element is obtained, thereby clarifying the thermal stress state of the RF front-end module under specific working conditions. The stress distribution function is solved by thermomechanical coupling. The temperature field calculation model is established through the mapping relationship between temperature and stress. The temperature-stress mapping relationship reflects the close connection between thermal stress and temperature, which enables the system to derive the initial value of the temperature field based on the known stress field information. The initial value of the temperature field is substituted into the heat conduction equation for iterative calculation. In each iteration, the value of the temperature field is updated according to the thermal stress distribution and boundary conditions. This process continuously corrects the predicted value of the temperature field until a stable temperature distribution matrix is ​​obtained.After multiple iterative calculations, the obtained temperature field distribution matrix can accurately reflect the temperature state of the RF front-end module under different working conditions.

[0033] Step S4, calculating the normal stress and tangential stress of each microelement according to the temperature field distribution matrix, and establishing a high-order lumped parameter equation group to obtain the stress field characteristic vector by solving;

[0034] Specifically, tensor analysis is performed on the thermal deformation of each microelement in the temperature field distribution matrix to describe the spatial deformation of the microelement after heating, especially how the deformation of each microelement propagates in space under the action of the temperature field. Through analysis, the thermal strain field equation is established to describe the strain behavior of microelement at different positions due to thermal expansion or thermal contraction, and the distribution data of thermal strain is obtained to reflect the deformation of each part of the RF front-end module due to temperature change during operation. Based on the thermal strain distribution data, the stress-strain relationship of each microelement is established. The core of the stress-strain relationship is to calculate the normal stress and tangential stress in the microelement according to Hooke's law. Hooke's law describes the linear relationship between stress and strain. Through this relationship, the normal stress (such as σx, σy, σz) and tangential stress (such as τxy, τyz, τxz) of each microelement under thermal deformation are calculated. These stress components constitute the initial stress state, reflecting the stress response of each microelement under the action of the temperature field. The initial stress components are corrected by boundary conditions. Since the elastic behavior of the material of the RF front-end module will change under different temperatures and external conditions, it is corrected according to the elastic coefficient of the material of the RF device. By calculating the correction coefficient, the initial stress component is adjusted to ensure that the stress calculation result can more accurately reflect the actual working state. Substitute the corrected stress component into the thermoelastic equation to optimize the calculation of the stress field. The thermoelastic equation is an equation that describes the relationship between stress and strain of a material under thermal action. By constructing the thermoelastic equation based on the elastic modulus and Poisson's ratio of the POI substrate material, a high-order lumped parameter equation is obtained to describe the stress distribution and interaction of each microelement in the RF front-end module. The high-order lumped parameter equation is corrected to take into account the heat source effect inside the module. Since the RF front-end module will have a certain heat source distribution during operation, especially under high power conditions, the heat source distribution is constructed according to the heat flux density function corresponding to the low-loss high-power electrode at the target power. The heat source distribution can reflect the generation and distribution of thermal energy inside the module. The heat source term matrix generated in this process will affect the solution of the thermoelastic equation. By introducing the heat source term matrix into the parametric equation group, the thermal behavior of the module under different working conditions can be effectively simulated. The parametric equation group is modified by the finite element method. The finite element method is a numerical calculation method for structural mechanics and thermodynamics problems. The stiffness matrix and mass matrix are constructed by this method to describe the stiffness characteristics and mass distribution inside the module. Based on the stiffness matrix and the mass matrix, a global equation group is obtained, which contains all the information of the thermoelastic equations, heat source terms and boundary conditions. By solving the global equation group, the principal stress components of each microelement are obtained. The principal stress refers to the maximum stress value that appears in the material along a specific direction, reflecting the stress state of the module under specific loading conditions. In this process, the principal stress and its direction cosine are calculated through the displacement-stress coordination condition. The characteristic stress data is subjected to eigenvalue decomposition and vector reconstruction.Eigenvalue decomposition is to decompose stress field data into a set of characteristic stresses and corresponding eigenvectors to find the main direction of the stress field. Through this process, the main direction of the stress field is determined, and characteristic basis functions are established based on these main directions. The characteristic basis function can convert the complex stress field into a simpler and easier to analyze mathematical model, providing a basis for the subsequent calculation of the stress field eigenvector. The stress field eigenvector is obtained.

[0035] The micro-element bodies of the RF front-end module are meshed, and the mesh nodes of key components such as TF-SAW, filters and duplexers are constructed. Through these mesh nodes, the finite element mesh of the RF front-end module is effectively divided, and the stiffness matrix and load vector are constructed on this basis. The stiffness matrix reflects the elastic response of each micro-element body under the action of external force, while the load vector describes the distribution of external heat sources or forces acting on the module. Through the construction of these matrices, discretized numerical equations are obtained. Based on the discretized numerical equations, a pre-processed conjugate gradient solver is constructed. This solver is an efficient numerical method for solving large-scale sparse linear equations. In the thermal stress analysis of RF front-end modules, especially when it comes to complex device structures and multi-dimensional stress fields, the use of the conjugate gradient method can effectively improve the solution efficiency. The conjugate gradient method iteratively approximates the true solution, and each iteration uses the current solution to adjust the next estimate, thereby gradually improving the accuracy of the solution. When using the conjugate gradient method, the boundary constraints of the POI substrate material and the low-loss high-power electrode are considered. These conditions can limit the displacement and stress state of the module and ensure the physical consistency of the solution process. Through iterative calculation, the numerical solution of the displacement field is obtained. The gradient calculation is performed on the numerical solution of the displacement field, and the strain of the microelement in space is obtained by solving the rate of change of the displacement field. The strain field is an important indicator for describing material deformation, which can reflect the degree of deformation of microelement at different positions under the action of temperature and external load. According to the displacement-strain relationship, the linear strain component and shear strain component of each microelement are obtained to form a strain field tensor, which reflects the deformation characteristics of the RF front-end module under thermal stress. The strain field tensor is substituted into the thermoelastic constitutive equation. The thermoelastic constitutive equation combines the relationship between temperature, strain and stress to describe the mechanical response of the material under thermal action. By substituting the strain field tensor into the equation, the stress component of each microelement is calculated. In the calculation process, the temperature field data under the preset power working state is combined to ensure that the stress calculation can take into account the influence of heat source and heat flux density on the stress state. Through calculation, a six-component stress tensor is obtained to describe the stress distribution in different directions. A characteristic equation is established for the six-component stress tensor. The characteristic equation obtains the corresponding eigenvalues ​​and eigenvectors by calculating the stress invariants of the stress tensor. The eigenvalues ​​represent the main stress components of the stress tensor, namely the principal stresses σ1, σ2, and σ3, while the eigenvectors describe the direction cosines of these principal stresses. The characteristic stress data is subjected to stress coordinate transformation to convert the distribution of the stress field into the principal coordinate system. Through the direction cosines, a coordinate rotation matrix is ​​constructed, which rotates the stress distribution in the original coordinate system to the principal coordinate system. The stress distribution in the principal coordinate system can reveal the stress state of the module in the principal direction. Based on the stress distribution in the principal coordinate system, singular value decomposition is performed to decompose a matrix into several orthogonal matrices and diagonal matrices. Through singular value decomposition, the main direction of the stress field is extracted, and the orthogonal basis vectors are selected according to the main direction of the stress field.These basis vectors constitute the characteristic basis functions of the stress field, describing the stress state at each position in the module. The stress characteristic basis functions are combined with the principal stress components to reconstruct the complete stress field distribution according to the stress superposition principle. The stress superposition principle states that the total stress field in the module is regarded as the superposition of the principal stress components, thereby obtaining a complete stress field distribution, reflecting the thermal stress state of the RF front-end module under different temperature and load conditions, and obtaining the stress field characteristic vector of the RF front-end module, revealing its mechanical properties under thermal stress.

[0036] Step S5: input the stress field feature vector into the thermal stress time evolution model, map the feature to the latent space through the encoder in the thermal stress time evolution model and reconstruct it through the decoder, and output the thermal stress distribution prediction data.

[0037] Specifically, the stress field feature vector is pre-processed to improve the availability and effectiveness of the data. The thermal stress evolution characteristics are decomposed in time and frequency through multi-scale wavelet transform, and feature subsets of different frequency bands are extracted. Multi-scale wavelet transform is a time-frequency analysis method that can capture the local characteristics of data in time and frequency at the same time, so as to help the system identify the dynamic changes of thermal stress in different frequency bands. Through time-frequency decomposition, the feature subset obtained contains information of different scales in the process of thermal stress evolution. The feature subset is input into the encoder network of the thermal stress time evolution model for processing. The encoder network maps the input high-dimensional feature data to a low-dimensional latent space. The encoder network reduces the feature dimension to the latent space through hierarchical convolution operation and downsampling mapping to obtain the latent feature representation. The encoder network includes a five-layer convolutional neural network structure. Each convolution layer uses the LeakyReLU activation function, which helps to avoid the gradient vanishing problem and improve the nonlinear modeling ability of the network. Each convolution layer is followed by an Instance Normalization layer, which can reduce the variability between different samples and make the network more stable during training. The size of the convolution kernel is 3×3, and the step size is set to 2. This setting effectively reduces the feature dimension while maintaining the spatial structure information. The latent feature representation is input into the time series feature extraction module in the thermal stress time evolution model for processing. The task of the time series feature extraction module is to capture the evolution law of thermal stress in time, so as to predict the future changes of thermal stress. This module adopts a three-layer bidirectional long short-term memory network (Bi-LSTM), and the hidden layer dimension of each layer is 256, so that the network can process time series data in a higher dimension and capture richer time series information. Bidirectional LSTM can take into account the positive and negative dependencies of the data at the same time, which helps to capture the long-term and short-term dependencies in the process of thermal stress evolution. In order to prevent overfitting, a Dropout layer is added between each LSTM layer with a deactivation rate of 0.2, which effectively improves the generalization ability of the model. The time series correlation features are input into the decoder network for processing. The decoder network maps the features in the latent space back to the original feature space of the thermal stress distribution to obtain the reconstructed thermal stress distribution feature data. The decoder network restores the feature dimension of thermal stress to the original space through deconvolution operation and upsampling technology. The decoder network also consists of five deconvolution layers, each of which uses the ReLU activation function, which helps the network to reconstruct the data nonlinearly. The Batch Normalization layer is used after the deconvolution layer to help speed up the training process and improve training stability. Adding a jump connection between the third and fourth layers of the decoder helps the network retain more low-level information, thereby improving reconstruction accuracy. The reconstructed feature data is transformed into a stress field mapping to restore it to the actual thermal stress distribution prediction data.By reconstructing the principal stress direction and synthesizing the stress components, the feature data output by the decoder is converted into a complete stress field distribution. The reconstructed features are reconstructed according to the principal stress direction to ensure that the prediction results are consistent with the actual physical scene. By synthesizing the stress components in various directions, the final thermal stress distribution prediction data is obtained.

[0038] In the embodiment of the present invention, by establishing a multi-heat conduction path model and introducing a stress compensation coefficient, the problem of spatial stress variation in complex RF structures is solved, making the thermal stress analysis results more accurate; a high-order lumped parameter thermal stress model is adopted to avoid the computational burden caused by excessive discretization, while ensuring the estimation accuracy at different discretization levels; through the intelligent prediction method of a multi-scale inversion transform network, the thermal stress monitoring data with interference, multi-modal and large time span characteristics is effectively processed, and the prediction accuracy is improved; combined with the POI substrate material and the low-loss high-power electrode solution, an accurate description of the distributed heat source inside the module is achieved, ensuring the continuity of stress between adjacent discrete components; a complete thermal stress prediction model is established, which is suitable for thermal stress analysis of RF devices such as TF-SAW, filters, and duplexers, and has a good prediction effect under high-power working conditions.

[0039] In a specific embodiment, the process of executing step S1 may specifically include the following steps:

[0040] Monitor the operating temperature of the RF front-end module in the frequency band of 20MHz to 40GHz to obtain a temperature monitoring data set, and collect data on the stress concentration area inside the RF front-end module to obtain a stress monitoring data set;

[0041] A thermal image sensor is aligned with the surface of the RF front-end module to obtain a heat distribution monitoring data set, and a deformation amount of the RF front-end module is dynamically monitored to obtain a deformation monitoring data set;

[0042] The sampling frequency is set according to the temperature monitoring data set, the stress monitoring data set, the heat distribution monitoring data set and the deformation monitoring data set, and continuous sampling is performed to obtain the original sampling data;

[0043] Performing bandpass filtering and amplitude normalization processing on the original sampling data to obtain a preprocessed data set, and performing data alignment on the preprocessed data set to obtain a multi-dimensional synchronous data matrix;

[0044] The principal component analysis is performed on the multi-dimensional synchronous data matrix, and the target characteristics of the temperature field, stress field, deformation field and heat distribution field are extracted through eigenvalue decomposition to construct a thermal stress monitoring data sequence.

[0045] Specifically, the operating temperature of the RF front-end module in the frequency band range of 20 MHz to 40 GHz is monitored to obtain a temperature monitoring data set. Through a high-precision temperature sensor, the temperature data of the module surface and each key point inside the module are recorded in real time when the RF front-end module is working, reflecting the thermal behavior of the RF front-end module under different frequency band excitations. Data is collected from the stress concentration area inside the RF front-end module, and stress sensors or stress testing technologies, such as surface strain gauges or piezoelectric stress sensors, are used to measure the distribution of stress inside the module. The obtained stress monitoring data set contains stress values ​​in different areas, which helps the system identify which areas have high stress concentration phenomena. At the same time, a thermal image sensor is used to monitor the surface of the RF front-end module. The thermal image sensor can capture the heat distribution on the module surface in real time, output a heat distribution monitoring data set, and reveal the heat distribution of the RF front-end module during operation. At the same time, the deformation of the RF front-end module is dynamically monitored. The deformation monitoring data set records the deformation of the module under heat in real time by using tools such as displacement sensors or strain gauges, helping to analyze the generation and evolution of thermal stress. Set a suitable sampling frequency and perform continuous sampling to ensure that each data set has a corresponding data value at the same time node and avoid data deviation caused by inconsistent sampling intervals. The original sampled data obtained contains multi-dimensional information such as temperature, stress, thermal distribution and deformation. The original data is bandpass filtered and amplitude normalized. The bandpass filter effectively removes noise outside the frequency band and retains the frequency components in the target signal. Amplitude normalization ensures that the signal strength between different sensors is consistent, so that data from different sources are comparable in numerical range, eliminating the impact caused by differences in sensor accuracy or range, and ensuring data consistency. Align the data sets after bandpass filtering and normalization, align different types of data (temperature, stress, deformation, etc.) in the time dimension, and construct a multi-dimensional synchronous data matrix, in which each row represents a time point and each column represents a specific sensor data. The dimensions of the data matrix are ,in is the number of sampling time points, is the dimension of the data, which depends on the number and type of sensors. The target features in the synchronized data matrix are extracted by principal component analysis. Principal component analysis maps data from a high-dimensional space to a low-dimensional space by eigenvalue decomposition. The covariance matrix of the data matrix is ​​calculated :

[0046] ;

[0047] in, is the matrix containing the original data, is the number of samples of the data. Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalue and the corresponding eigenvector :

[0048] ;

[0049] Through eigenvalue decomposition, the most important components in the data are found, and several main characteristic directions are selected. These directions correspond to the maximum variance of the data, which means that they have the most important information in the data. By selecting the eigenvector corresponding to the maximum eigenvalue, a low-dimensional representation of the data is constructed to obtain the thermal stress monitoring data sequence.

[0050] In a specific embodiment, the process of executing step S2 may specifically include the following steps:

[0051] The spatial distribution of the temperature field data in the thermal stress monitoring data sequence is analyzed, and the minimum spatial characteristic scale is calculated according to the temperature gradient change to obtain the grid division benchmark data;

[0052] According to the grid division benchmark data, the RF front-end module is divided into N×M×L micro-elements in the X, Y, and Z directions, and the geometric boundary coordinates of each micro-element are determined to obtain a spatial discretization structure;

[0053] Calibrate the material properties of TF-SAW, filters and duplexers in the spatial discretization structure, establish a material database based on the dielectric constant and thermal conductivity of the POI substrate material, and obtain the device material mapping relationship;

[0054] The thermal conductivity of each microelement is assigned based on the device material mapping relationship, and the thermal conductivity of the boundary microelement is determined according to the material parameters of the low-loss high-power electrode to obtain the thermal conductivity distribution matrix;

[0055] The density parameters of the micro-element are spatially allocated, and the density distribution function is calculated according to the mass-volume relationship of each RF device to obtain a density distribution matrix. The heat capacity parameters of each micro-element are calculated based on the thermal conductivity distribution matrix and the density distribution matrix, and a temperature-heat capacity mapping function is established to obtain a heat capacity distribution matrix.

[0056] According to the thermal conductivity distribution matrix, density distribution matrix and heat capacity distribution matrix, the heat conduction equations of each microelement are constructed, and the heat transfer boundary conditions are set to obtain the transfer equation system. The transfer equation system is subjected to thermal-mechanical coupling analysis, and the initial thermal stress analysis model is obtained by establishing the correlation equation between the temperature field and the displacement field.

[0057] Specifically, the temperature field data in the thermal stress monitoring data sequence is spatially analyzed to obtain the temperature gradient change information. Through the spatial analysis of the temperature field data, the area with more drastic temperature changes is identified. The temperature gradient is an important indicator to describe the temperature change rate. The temperature gradient is calculated using the partial derivative form, namely:

[0058] ;

[0059] in, , and The temperature along , , The gradient in the axial direction. These gradient values ​​help determine the areas with the fastest temperature changes, which are usually areas where stress and deformation are concentrated, that is, thermal stress affected areas. According to the temperature gradient change, the minimum spatial characteristic scale (i.e., grid division benchmark data) is calculated. This scale determines the granularity of spatial discretization in thermal stress analysis, affecting the accuracy of grid division and the complexity of calculation. According to the spatial characteristic scale, the RF front-end module is placed in , , Divided into three directions Each microelement is a small three-dimensional cube or cuboid, and its geometric boundary coordinates are determined by the spatial discretization structure. This structure defines the geometric grid of the entire module, so that each microelement can be effectively connected to the surrounding microelement for heat conduction and mechanical analysis. The result of spatial discretization is a A grid of small units, each of which corresponds to a small part of the RF front-end module. Through discretization, thermal conduction, stress and other physical effects are simulated and analyzed at a finer scale. The material properties of each component in the RF front-end module are calibrated, especially for components such as TF-SAW (thin film surface acoustic wave devices), filters and duplexers. The material calibration process is based on the materials used in the module, such as the dielectric constant and thermal conductivity of the POI (ferroelectric potassium niobate) substrate material. The properties of these materials determine how heat is transferred between different components. Establish a material database to record the key physical properties of various materials, including the dielectric constant , thermal conductivity , elastic modulus Etc. Through the database of these material parameters, the thermal conduction and stress transfer relationship between different materials in the RF front-end module is described. According to the material mapping relationship provided in the material database, the thermal conductivity of each microelement is assigned. These thermal conductivity values ​​determine the efficiency of heat transfer in different microelements. Especially between different RF devices, the difference in thermal conductivity will significantly affect the distribution of thermal stress. For example, for low-loss high-power electrodes, the thermal conductivity of their materials is usually high, and special attention should be paid to the thermal conductivity distribution of boundary microelements. On this basis, the thermal conductivity distribution matrix is ​​obtained, which represents the thermal conductivity distribution of each microelement in the RF front-end module. The density of the microelement is spatially allocated. Density It is an important parameter for heat capacity calculation. The density distribution function is derived from the mass-volume relationship of the RF device. For example, for the metal part of the RF front-end module, its density is calculated through specific material data, and this value is distributed to each microelement to obtain the density distribution matrix. The heat capacity of each microelement is calculated based on the thermal conductivity distribution matrix and the density distribution matrix. Heat capacity It is a physical quantity that describes the ability of an object to absorb heat, and is related to the density of the material. Specific heat capacity and volume For each microelement, its heat capacity is calculated by the following formula:

[0060] ;

[0061] in, is the volume of the microelement, is the density, is the specific heat capacity. By calculating the heat capacity of each microelement, the heat capacity distribution matrix is ​​obtained. The heat conduction equations of each microelement are constructed according to the thermal conductivity distribution matrix, density distribution matrix and heat capacity distribution matrix. The heat conduction equation is described by Fourier's heat conduction law, and the basic form is:

[0062] ;

[0063] in, is the temperature field, It's time. is the thermal diffusivity, It is the Laplace operator of temperature, which describes the spatial distribution of temperature. By solving the heat conduction equations of all microelements, a complete heat transfer equation system is obtained, which reflects the temperature changes of the entire RF front-end module at different time and space points. On this basis, heat transfer boundary conditions are set for the heat conduction equation. Boundary conditions include boundary conditions for heat flow, such as specifying the temperature of the module surface or interface, or specifying the transfer direction of heat flow. Through thermal-mechanical coupling analysis, combined with the correlation equations of temperature field and displacement field, the initial thermal stress analysis model is obtained. By establishing the relationship between temperature field and displacement field, the thermal stress of each microelement is solved, and the stress concentration areas in the module and the stress evolution trend of these areas are further analyzed.

[0064] In a specific embodiment, the process of executing step S3 may specifically include the following steps:

[0065] Perform stress characteristic analysis on the initial thermal stress analysis model and obtain stress response data based on the geometric parameters and material properties of the device structure;

[0066] Based on the stress response data, a thermodynamic analysis is performed on each microelement, and a stress compensation coefficient calculation model is established according to the material parameters of the POI substrate material and the low-loss high-power electrode to obtain the initial compensation coefficient.

[0067] The stress compensation matrix of each microelement is constructed according to the initial compensation coefficient, and the stress distribution under the preset working state is corrected and calculated to obtain a corrected compensation coefficient matrix;

[0068] Input the modified compensation coefficient matrix into the thermal stress transfer model between adjacent micro-elements, establish the stress transfer function according to the stress continuity principle, and obtain the thermal stress transfer relationship;

[0069] Constrain the temperature boundary in the thermal stress transfer relationship and obtain the boundary constraint conditions according to the thermal expansion characteristics within the target temperature range;

[0070] Based on the thermal stress transfer relationship and boundary constraints, the thermal stress control equations are constructed, and the stress field is solved according to the stress equilibrium principle to obtain the stress distribution function.

[0071] The stress distribution function is solved by thermal-mechanical coupling, and a temperature field calculation model is established according to the temperature-stress mapping relationship to obtain the initial value of the temperature field. The initial value of the temperature field is substituted into the heat conduction equation for iterative calculation. The temperature field value is updated according to the thermal stress distribution and boundary conditions to obtain the temperature field distribution matrix.

[0072] Specifically, the stress characteristics of the initial thermal stress analysis model are analyzed. According to the geometric parameters and material properties of the device structure, the stress response data is obtained. The stress response describes the stress reaction of an object under external load or temperature change, which depends on the mechanical properties and geometric shape of the material. For example, for each microelement in the RF front-end module, assuming that its material follows Hooke's law, the stress and strain The relationship between them is expressed as:

[0073] ;

[0074] in, is the elastic modulus of the material, is strain. The stress response data is used to simulate the entire module through numerical methods (such as the finite element method) to calculate the stress distribution in different areas. After obtaining the stress response data, each microelement is subjected to thermodynamic analysis to establish a stress compensation coefficient calculation model. The calculation of the stress compensation coefficient mainly takes into account the thermal expansion differences between different materials and the deformation effects of these materials after heating. Taking POI (ferroelectric potassium niobate) substrate material and low-loss high-power electrode as an example, the thermal expansion coefficient and thermal conductivity of the POI substrate material are different from those of the electrode material, and a compensation coefficient model is established for each material. Assume is the thermal expansion coefficient of the material, is the temperature change, compensation coefficient Calculated according to the following formula:

[0075] ;

[0076] in, is the coefficient of thermal expansion, in K, and is the temperature change in K. Through the compensation coefficient model, each microelement is thermodynamically compensated to estimate the stress change caused by temperature change. The stress compensation matrix of each microelement is constructed according to the initial compensation coefficient, and the original stress distribution is corrected so that it can more accurately reflect the actual situation. The calculation process of the corrected compensation coefficient matrix is ​​usually based on the following two steps: The original stress distribution is corrected by considering the interaction between different microelements. According to the calculation results of the compensation coefficient, the stress is adjusted to obtain a more accurate stress distribution. Corrected compensation coefficient matrix The calculation formula is expressed as:

[0077] ;

[0078] in, is the initial stress compensation coefficient matrix, is the change of the compensation coefficient after correction. The correction compensation coefficient matrix is ​​input into the thermal stress transfer model. The core of the thermal stress transfer model is to establish the stress transfer relationship between adjacent micro-elements according to the principle of stress continuity. The principle of stress continuity states that the stress should remain continuous between two adjacent micro-elements. Assume that there is a stress transfer relationship between the contact surface of the two micro-elements , the stress transfer is expressed by the following equation:

[0079] ;

[0080] in, and are the stresses of adjacent micro-elements. Through the stress transfer function, the thermal stress transfer relationship is constructed to solve the thermal stress of the entire module. After obtaining the thermal stress transfer relationship, the temperature boundary is constrained, especially the thermal expansion characteristics within the temperature range must be considered. Thermal expansion characteristics refer to the changes in the size and shape of the material when the temperature changes. The temperature behavior of the system is controlled by introducing boundary conditions. For example, assuming that within the target temperature range arrive The thermal expansion coefficient of the material is known, the temperature field is constrained by the following boundary conditions:

[0081] ;

[0082] in, is the dimensional change caused by temperature change, is the initial length of the material, is the coefficient of thermal expansion, is the temperature change. Through these boundary conditions, the temperature field changes in the system are effectively controlled and constrained, thereby affecting the distribution of the stress field. Based on the thermal stress transfer relationship and boundary constraints, the thermal stress control equations are constructed. The thermal stress control equations mainly describe the stress distribution in the entire system through the principle of stress balance, combined with the heat transfer equation and the mechanical equation. The thermal stress control equations are:

[0083] ;

[0084] ;

[0085] in, is stress, is the elastic modulus, It's strain. is the temperature, It's time. is the thermal diffusion coefficient. By solving this set of equations, we get the stress distribution function of the whole system. Based on the stress distribution function, we solve the thermal-mechanical coupling to further solve the temperature field. Temperature and stress are highly coupled. Changes in temperature will cause changes in stress, and stress will also affect the distribution of the temperature field. Solve the temperature field through iterative calculation. Based on the temperature-stress mapping relationship, establish a calculation model for the temperature field:

[0086] ;

[0087] in, is the initial value of the temperature field, is the temperature change. During the calculation process, the initial temperature field is substituted into the heat conduction equation and the temperature field is updated through iterative solution. In each iteration, the stress distribution is updated according to the current temperature distribution and boundary conditions, thereby correcting the calculation results of the temperature field until convergence. The temperature field distribution matrix of the entire module is obtained.

[0088] In a specific embodiment, the process of executing step S4 may specifically include the following steps:

[0089] The thermal deformation of each microelement in the temperature field distribution matrix is ​​subjected to tensor analysis, and the thermal strain field equation is established according to the spatial deformation of the microelement to obtain the thermal strain distribution data;

[0090] Based on the thermal strain distribution data, the stress-strain relationship of each microelement is established, and the normal stress σx, σy, σz and the tangential stress τxy, τyz, τxz are calculated according to Hooke's law to obtain the initial stress components;

[0091] Perform boundary condition correction on the initial stress component, calculate the correction coefficient according to the material elastic coefficient of the radio frequency device, and obtain the corrected stress component;

[0092] Substituting the modified stress component into the thermoelastic equation, a high-order lumped parameter equation is established according to the elastic modulus and Poisson's ratio of the POI substrate material to obtain a set of parameter equations;

[0093] A distributed heat source term is introduced into the parametric equation group, and the heat source distribution is constructed according to the heat flux density function corresponding to the target power of the low-loss high-power electrode to obtain the heat source term matrix;

[0094] The parametric equations are modified based on the heat source matrix, and the stiffness matrix and mass matrix are constructed according to the finite element method to obtain the global equations.

[0095] The global equations are numerically solved, and the principal stresses σ1, σ2, σ3 and their direction cosines are calculated according to the displacement-stress coordination conditions to obtain characteristic stress data. The characteristic stress data are then subjected to eigenvalue decomposition and vector reconstruction, and characteristic basis functions are established according to the main directions of the stress field to obtain the characteristic vectors of the stress field.

[0096] Specifically, tensor analysis is performed on the thermal deformation of each microelement to quantify the deformation of each microelement caused by the change in temperature field. In the tensor analysis process, the thermal deformation is expressed as the displacement field caused by thermal expansion, and the relationship between temperature change and thermal strain is described by the thermal expansion coefficient. Assume is the coefficient of thermal expansion, is the temperature change, the thermal deformation of the microelement The calculation is done by the following formula:

[0097] ;

[0098] in, is the length of the microelement. By performing tensor analysis on the spatial deformation of each microelement, the thermal strain field equation is established to describe the thermal strain distribution of each microelement. Assume that the thermal strain tensor There is a linear relationship between it and the displacement field, which is described by the definition of the strain tensor:

[0099] ;

[0100] in, are the components of the displacement vector, are spatial coordinates, Represents the thermal strain component of the microelement in the spatial direction. The stress-strain relationship of the microelement is established based on the thermal strain distribution data. The relationship between stress and strain is given by Hooke's law, which describes the behavior of the material within the elastic range. Hooke's law is expressed by the following formula:

[0101] ;

[0102] in, is stress, is the elastic modulus, is the strain. Normal stress , , and tangential stress , , The calculation uses the following formula:

[0103] ;

[0104] ;

[0105] in, , , is the shear strain component, is the shear modulus of the material, , , is along , , The thermal strain component in the direction of the initial stress component is corrected by boundary conditions to ensure that the stress calculation of the material under the external environment and load can meet the actual boundary constraints. According to the elastic coefficient of the RF device material, the correction coefficient is calculated by the following formula:

[0106] ;

[0107] in, is the elastic modulus corrected according to the boundary conditions, is the elastic modulus of the original material. Through the corrected coefficient, the corrected stress component is obtained ,Right now:

[0108] Correction factor;

[0109] Substitute the corrected stress components into the thermoelastic equation for solution. The thermoelastic equation describes the behavior of materials under thermal stress, combining the basic principles of heat conduction and elasticity. The thermoelastic behavior of different materials is described by the following thermoelastic equation:

[0110] ;

[0111] in, is the coefficient of thermal expansion, is the temperature change, is the thermal strain. Through this equation, a high-order lumped parameter equation group is established to describe the stress state of each microelement in the RF front-end module. The high-order lumped parameter equation group contains parameters such as elastic modulus and thermal expansion coefficient of different material regions, which can comprehensively consider the stress distribution in different regions. After introducing the distributed heat source term, the parameter equation group is modified according to the distribution of the heat source. The distributed heat source term is related to the heat flux density of the electrode. For example, the heat flux density generated by a low-loss high-power electrode during operation Calculated by the following formula:

[0112] ;

[0113] in, is the power generated by the electrode, is the area of ​​the electrode. By calculating the heat source distribution of the electrode, the heat source term matrix is ​​obtained , and substitute it into the parametric equations for correction. After the parametric equations are corrected based on the heat source matrix, the stiffness matrix and mass matrix are constructed according to the finite element method. These two matrices are basic tools for describing the mechanical behavior and thermal conductivity of materials. and the mass matrix They are closely related to the elastic modulus, density and other parameters of the material. The formula for constructing the stiffness matrix is:

[0114] ;

[0115] in, is the deformation gradient matrix, is the material stiffness matrix, is the volume of the object. The mass matrix is ​​constructed as:

[0116] ;

[0117] in, is the density of the material, is the shape function. The global equations are numerically solved to obtain the displacement and stress distribution of each microelement under different loads and temperature fields. In this process, the displacement-stress coordination condition ensures the coupling between the stress field and the displacement field. Based on the coordination condition between the displacement field and the stress field, the principal stress is calculated by numerical methods. , , and its direction cosines. The calculation of the principal stress is performed by eigenvalue decomposition, and the eigenvalue decomposition formula is:

[0118] ;

[0119] in, is the eigenvalue, is the identity matrix, is the stress matrix. By solving the characteristic equation, the principal stress values ​​and their direction cosines are obtained. After the principal stress data is decomposed by eigenvalues, the characteristic basis functions of the stress field are constructed to describe the main directions and distributions of the stress field, and to help further analyze the thermal stress behavior in the RF front-end module. Based on the direction cosines of the principal stresses, combined with numerical reconstruction technology, the stress field eigenvector is calculated.

[0120] In a specific embodiment, the execution step numerically solves the global equation group, calculates the principal stresses σ1, σ2, σ3 and their direction cosines according to the displacement-stress coordination condition, obtains characteristic stress data, performs eigenvalue decomposition and vector reconstruction on the characteristic stress data, establishes characteristic basis functions according to the main directions of the stress field, and obtains the characteristic vector of the stress field. The process can specifically include the following steps:

[0121] The global equations are discretized by finite element method, and the stiffness matrix and load vector are constructed according to the grid nodes of TF-SAW, filter and duplexer to obtain the discretized numerical equations.

[0122] A pre-processed conjugate gradient solver is constructed based on the discretized numerical equations, and iterative calculations are performed according to the boundary constraints of the POI substrate material and the low-loss high-power electrode to obtain the numerical solution of the displacement field.

[0123] The gradient of the numerical solution of the displacement field is calculated, and the linear strain and shear strain components of each microelement are obtained according to the displacement-strain relationship to obtain the strain field tensor;

[0124] Substitute the strain field tensor into the thermoelastic constitutive equation, calculate the stress components according to the temperature field under the preset power working state, and obtain a six-component stress tensor;

[0125] The characteristic equation is established for the six-component stress tensor, and the eigenvalue and eigenvector are calculated according to the stress invariant to obtain the characteristic stress data, which includes the principal stresses σ1, σ2, σ3 and their direction cosines.

[0126] Based on the characteristic stress data, stress coordinate transformation is performed, and the coordinate rotation matrix is ​​constructed according to the direction cosine to obtain the stress distribution in the principal coordinate system. The stress distribution in the principal coordinate system is decomposed by singular value, and the orthogonal basis vectors are selected according to the main direction of the stress field to obtain the stress characteristic basis function.

[0127] The stress characteristic basis functions are combined with the principal stress components, and the complete stress field distribution is reconstructed according to the stress superposition principle to obtain the stress field characteristic vector.

[0128] Specifically, the global equations are discretized by finite element method. The finite element method transforms continuous physical field problems into discrete numerical equations. Based on the geometric structure of the RF front-end module, especially the grid nodes of TF-SAW (thin film surface acoustic wave), filter and duplexer, the stiffness matrix and load vector of each microelement are constructed by dividing these nodes. Stiffness matrix It is a matrix that describes the rigid response of a material in all directions, reflecting how an object deforms when subjected to external forces. It describes the influence of external forces or heat sources on the object. The stiffness matrix and load vector are constructed using the basic formulas in the finite element method:

[0129] ;

[0130] in, is the strain-displacement matrix, is the material stiffness matrix, is a volume element. The load vector It is expressed as:

[0131] ;

[0132] in, is the shape function, is the external force density or heat source term per unit volume. The numerical equation obtained by discretization is ,in is a displacement vector that describes the deformation of an object under external force. A preconditioned conjugate gradient solver is constructed based on the discretized numerical equations. The conjugate gradient method is an effective method for solving linear equations and is applicable to sparse matrices. In order to improve the efficiency of the solution, a preconditioning technique is used to convert the original system into a system that is easier to solve by preconditioning the original system. The preconditioning process includes the stiffness matrix Preconditioning is performed, and the preconditioning methods include incomplete LU decomposition or Jacobi preconditioner. The iterative formula of the preconditioned conjugate gradient method is expressed as:

[0133] ;

[0134] ;

[0135] ;

[0136] in, is the residual vector, is the initial displacement estimate, is the inverse of the preconditioning matrix, is the residual vector after preprocessing, is the search direction. Through iterative calculation, the exact solution of the system, that is, the numerical solution of the displacement field, is gradually approached. The numerical solution of the displacement field is gradient calculated to determine the strain of each microelement, especially the linear strain and shear strain components. According to the relationship between the displacement field and the strain, the linear strain and shear strain are calculated using the following formula:

[0137] ;

[0138] ;

[0139] These strain components constitute the strain field tensor , describing the degree of deformation of each microelement in each direction. Substitute the strain field tensor into the thermoelastic constitutive equation to calculate the stress component. The thermoelastic constitutive equation reflects the mechanical behavior of the material under thermal stress and is expressed in the following form:

[0140] ;

[0141] in, is stress, is the elastic modulus, is the coefficient of thermal expansion, is the temperature change, is the strain. From this equation, we can calculate the normal stress , , and tangential stress , , , and obtain the six-component stress tensor. Establish the characteristic equation for the six-component stress tensor, calculate the eigenvalue and eigenvector of stress, and obtain the characteristic stress data. The characteristic equation of stress is composed of the determinant of the stress tensor, and the calculation process is as follows:

[0142] ;

[0143] in, is the eigenvalue, is the identity matrix, is the stress matrix. Solving the characteristic equation, we get the principal stress , , and its direction cosines, which represent the maximum, minimum, and intermediate stress values, respectively, and provide information about the main direction of the stress field. Based on the characteristic stress data, stress coordinate transformation is performed. The stress distribution is transformed into the principal coordinate system to observe the stress distribution more clearly. In order to perform coordinate transformation, a coordinate rotation matrix is ​​constructed. , whose elements are determined by the direction cosines of the stress field:

[0144] ;

[0145] After substituting the stress data into the rotation matrix, the stress distribution in the principal coordinate system is obtained. The stress distribution in the principal coordinate system is subjected to singular value decomposition, and a matrix is ​​decomposed into the product of three matrices, so as to facilitate the extraction of its main features. Through singular value decomposition, the main direction of the stress field is extracted, and the orthogonal basis vectors are selected according to the main direction of the stress field. These orthogonal basis vectors constitute the stress characteristic basis function. The stress characteristic basis function describes the distribution of the stress field in different directions. Using the principle of stress superposition, the stress characteristic basis function is combined with the principal stress component to reconstruct the complete stress field distribution.

[0146] In a specific embodiment, the process of executing step S5 may specifically include the following steps:

[0147] The stress field feature vectors are pre-processed, and the time-frequency decomposition of the thermal stress evolution characteristics is performed through multi-scale wavelet transform to obtain feature subsets containing different frequency bands;

[0148] The feature subset is input into the encoder network of the thermal stress time evolution model. The feature is reduced to the latent space through hierarchical convolution operation and downsampling mapping to obtain the latent feature representation. The encoder network includes a five-layer convolutional neural network structure. Each convolution layer uses the LeakyReLU activation function and the Instance Normalization normalization layer. The convolution kernel size is 3×3 and the step size is 2.

[0149] The latent feature representation is input into the temporal feature extraction module of the thermal stress time evolution model. The temporal evolution law of thermal stress is captured through bidirectional sequence learning to obtain the temporal correlation features. The temporal feature extraction module includes a three-layer bidirectional long short-term memory network with a hidden layer dimension of 256. A Dropout layer is added between each layer with a deactivation rate of 0.2.

[0150] The temporal correlation features are input into the decoder network of the thermal stress time evolution model. The characteristic dimensions of the thermal stress distribution are restored through deconvolution operations and upsampling to obtain reconstructed feature data. The decoder network includes a five-layer deconvolution layer structure. Each deconvolution layer uses the ReLU activation function and the Batch Normalization layer, and a jump connection is added between the third and fourth layers.

[0151] The reconstructed feature data are transformed into stress field mapping, and the thermal stress distribution prediction data are obtained through principal stress direction reconstruction and stress component synthesis.

[0152] Specifically, the stress field feature vector is pre-processed. The original stress feature data is converted into a format suitable for input into the deep learning model. The evolution characteristics of thermal stress are decomposed into time and frequency by multi-scale wavelet transform. Wavelet transform extracts feature subsets of different frequency bands by performing multi-scale analysis on the signal in time and frequency, helping to capture the changes of thermal stress on different time scales. The change characteristics of thermal stress are expressed by the following formula:

[0153] ;

[0154] in, is the signal after wavelet transformation, is the original stress field characteristic signal, is a wavelet function. Through analysis at different scales, feature subsets containing multiple frequency bands are obtained, and these subsets are passed as input data to the subsequent deep learning model. The feature subset is input into the encoder network of the thermal stress time evolution model. The encoder network reduces the high-dimensional feature data to the latent space through hierarchical convolution operations and downsampling mapping to obtain the latent feature representation. Use convolutional neural networks for feature extraction and dimensionality reduction. The structure of the encoder network includes five layers of convolutional neural networks, and the convolution operation of each layer uses the LeakyReLU activation function and the Instance Normalization normalization layer. The size of the convolution kernel is 3×3, and the step size is 2, which means that each layer is extracting higher-level features while reducing the spatial resolution by half. The LeakyReLU activation function can avoid the "dead neuron" problem that occurs in traditional ReLU, and its formula is:

[0155] ;

[0156] in, is a small constant, usually set to 0.01. Instance Normalization can effectively avoid interference between different features by independently normalizing each batch of inputs, making network training more stable. The potential feature representation after dimensionality reduction is input into the time series feature extraction module of the thermal stress time evolution model. This module captures the evolution of thermal stress over time through a bidirectional long short-term memory network (Bi-LSTM). Bi-LSTM can process time series data from both forward and backward directions to better capture time correlation. The time series feature extraction module consists of three layers of Bi-LSTM, the hidden layer dimension of each layer is set to 256, and a Dropout layer is added between each layer with a deactivation rate of 0.2 to prevent overfitting. The output of Bi-LSTM is expressed by the following formula:

[0157] ;

[0158] in, is the time step The hidden state of and are the weight matrices of the previous hidden state and the current input, respectively. is the bias term, is the input feature. Bi-LSTM can capture richer temporal dependencies in time series by passing information in both the forward and reverse channels. The time series correlation features are input into the decoder network of the thermal stress time evolution model. The task of the decoder network is to map the features in the latent space back to the original thermal stress distribution feature dimension through deconvolution operations and upsampling. The deconvolution layer (also called the transposed convolution layer) is used for upsampling of images or feature maps. The formula for the deconvolution operation is:

[0159] ;

[0160] in, is the upsampled feature map, is the deconvolution kernel, is the input feature map. The decoder network consists of five deconvolution layers, each of which uses the ReLU activation function and the Batch Normalization layer to restore the high-dimensional thermal stress distribution characteristics. Between the third and fourth layers, a jump connection is added to help retain low-level feature information during the decoding process, thereby enhancing the expressive power of the decoder. At the output stage of the decoder network, the reconstructed feature data is transformed through the stress field mapping. Mapping the features output by the decoder back to the actual thermal stress distribution depends on the reconstruction of the principal stress direction and the synthesis of the stress components. The principal stress direction is usually defined by the maximum, minimum, and intermediate principal stress values, so it is expressed by the following formula:

[0161] ;

[0162] in, , , They are principal stresses respectively. By reconstructing the principal stress directions, the stress distribution in the principal coordinate system is obtained, and these components are combined into the final stress field distribution through stress component synthesis. According to the above steps, the predicted data of thermal stress distribution is finally obtained.

[0163] The above describes the RF front-end module thermal stress analysis method in the embodiment of the present invention. The following describes the RF front-end module thermal stress analysis system in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, a radio frequency front-end module thermal stress analysis system includes:

[0164] The acquisition module is used to collect the temperature data, stress data, deformation data and thermal distribution data of the RF front-end module, and synchronize and sort the data with the timestamp to obtain the thermal stress monitoring data sequence;

[0165] Establish a module for determining the spatial discretization parameters according to the thermal stress monitoring data sequence, and divide the RF front-end module into N×M×L micro-elements, establish a transfer equation containing the thermal conductivity, density and heat capacity parameters of each micro-element, and obtain an initial thermal stress analysis model;

[0166] The allocation module is used to allocate stress compensation coefficients based on each micro-element in the initial thermal stress analysis model, establish thermal stress transfer equations and boundary constraints between adjacent micro-elements, and solve to obtain the temperature field distribution matrix;

[0167] The solution module is used to calculate the normal stress and tangential stress of each microelement according to the temperature field distribution matrix, and establish a high-order lumped parameter equation group to solve the stress field eigenvector;

[0168] The output module is used to input the stress field feature vector into the thermal stress time evolution model, map the feature to the latent space through the encoder in the thermal stress time evolution model and reconstruct it through the decoder, and output the thermal stress distribution prediction data.

[0169] Through the collaborative cooperation of the above-mentioned components, by establishing a multi-heat conduction path model and introducing a stress compensation coefficient, the problem of spatial stress variation in complex RF structures is solved, making the thermal stress analysis results more accurate; a high-order lumped parameter thermal stress model is adopted to avoid the computational burden caused by excessive discretization, while ensuring the estimation accuracy at different discretization levels; through the intelligent prediction method of the multi-scale inversion transform network, the thermal stress monitoring data with interference, multi-modal and large time span characteristics is effectively processed, and the prediction accuracy is improved; combined with the POI substrate material and the low-loss high-power electrode solution, an accurate description of the distributed heat source inside the module is achieved, ensuring the continuity of stress between adjacent discrete components; a complete thermal stress prediction model is established, which is suitable for thermal stress analysis of RF devices such as TF-SAW, filters, and duplexers, and has good prediction effects under high-power working conditions.

[0170] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0171] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0172] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0173] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0174] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0175] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or 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 to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0176] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for analyzing thermal stress of a radio frequency front-end module, characterized in that: The method comprises: Collect temperature data, stress data, deformation data and thermal distribution data of the RF front-end module, synchronize and organize them with timestamps, and obtain a thermal stress monitoring data sequence; Determine the spatial discretization parameters according to the thermal stress monitoring data sequence, divide the RF front-end module into N×M×L micro-elements, establish a transfer equation including the thermal conductivity, density and heat capacity parameters of each micro-element, and obtain an initial thermal stress analysis model; Based on the distribution of stress compensation coefficients for each micro-element in the initial thermal stress analysis model, a thermal stress transfer equation and boundary constraints between adjacent micro-elements are established to obtain a temperature field distribution matrix; The normal stress and tangential stress of each microelement are calculated according to the temperature field distribution matrix, and a high-order lumped parameter equation group is established to solve the stress field characteristic vector; specifically, the method comprises: performing tensor analysis on the thermal deformation of each microelement in the temperature field distribution matrix, establishing a thermal strain field equation according to the spatial deformation of the microelement, and obtaining thermal strain distribution data; establishing a stress-strain relationship of each microelement based on the thermal strain distribution data, calculating the normal stress σx, σy, σz and the tangential stress τxy, τyz, τxz according to Hooke's law, and obtaining the initial stress component; performing boundary condition correction on the initial stress component, calculating the correction coefficient according to the material elastic coefficient of the radio frequency device, and obtaining the corrected stress component; substituting the corrected stress component into the thermal elastic coefficient. Equation, establish a high-order lumped parameter equation according to the elastic modulus and Poisson's ratio of the POI substrate material, and obtain a group of parameter equations; introduce a distributed heat source term into the group of parameter equations, and construct a heat source distribution according to the heat flux density function corresponding to the target power of the low-loss high-power electrode to obtain a heat source term matrix; based on the heat source term matrix, modify the group of parameter equations, and construct a stiffness matrix and a mass matrix according to the finite element method to obtain a global group of equations; numerically solve the global group of equations, calculate the principal stresses σ1, σ2, σ3 and their direction cosines according to the displacement-stress coordination conditions, and obtain characteristic stress data, and perform eigenvalue decomposition and vector reconstruction on the characteristic stress data, establish characteristic basis functions according to the main directions of the stress field, and obtain characteristic vectors of the stress field; The stress field feature vector is input into the thermal stress time evolution model, the feature is mapped to the latent space by the encoder in the thermal stress time evolution model and reconstructed by the decoder, and the thermal stress distribution prediction data is output.

2. The method for analyzing thermal stress of a radio frequency front-end module according to claim 1, characterized in that: The temperature data, stress data, deformation data and thermal distribution data of the RF front-end module are collected, and the timestamps are synchronized and sorted to obtain a thermal stress monitoring data sequence, including: Monitor the operating temperature of the RF front-end module within the frequency band of 20 MHz to 40 GHz to obtain a temperature monitoring data set, and collect data on the stress concentration area inside the RF front-end module to obtain a stress monitoring data set; A thermal image sensor is aligned with the surface of the RF front-end module to obtain a heat distribution monitoring data set, and a deformation amount of the RF front-end module is dynamically monitored to obtain a deformation monitoring data set; Setting a sampling frequency according to the temperature monitoring data set, the stress monitoring data set, the heat distribution monitoring data set, and the deformation monitoring data set, and performing continuous sampling to obtain original sampling data; Performing bandpass filtering and amplitude normalization processing on the original sampled data to obtain a preprocessed data set, and performing data alignment on the preprocessed data set to obtain a multi-dimensional synchronous data matrix; The multi-dimensional synchronous data matrix is ​​subjected to principal component analysis, and target features of the temperature field, stress field, deformation field and heat distribution field are extracted by eigenvalue decomposition to construct a thermal stress monitoring data sequence.

3. The method for analyzing thermal stress of a radio frequency front-end module according to claim 2, characterized in that: The method of determining the spatial discretization parameters according to the thermal stress monitoring data sequence, dividing the RF front-end module into N×M×L micro-elements, establishing a transfer equation including the thermal conductivity, density and heat capacity parameters of each micro-element, and obtaining an initial thermal stress analysis model includes: Performing spatial distribution analysis on the temperature field data in the thermal stress monitoring data sequence, calculating the minimum spatial characteristic scale according to the temperature gradient change, and obtaining grid division benchmark data; Divide the RF front-end module into N×M×L micro-elements in the X, Y, and Z directions according to the grid division benchmark data, and determine the geometric boundary coordinates of each micro-element to obtain a spatial discretization structure; Calibrate the material properties of the filter and duplexer in the spatial discretization structure, establish a material database according to the dielectric constant and thermal conductivity of the POI substrate material, and obtain a device material mapping relationship; Assigning a value to the thermal conductivity of each microelement based on the device material mapping relationship, and determining the thermal conductivity of the boundary microelement according to the material parameters of the low-loss high-power electrode to obtain a thermal conductivity distribution matrix; The density parameters of the micro-element are spatially allocated, and the density distribution function is calculated according to the mass-volume relationship of each radio frequency device to obtain a density distribution matrix, and the heat capacity parameters of each micro-element are calculated based on the thermal conductivity distribution matrix and the density distribution matrix, and a temperature-heat capacity mapping function is established to obtain a heat capacity distribution matrix; A group of heat conduction equations for each microelement is constructed according to the thermal conductivity distribution matrix, the density distribution matrix and the heat capacity distribution matrix, and heat transfer boundary conditions are set to obtain a transfer equation system. A thermal-mechanical coupling analysis is performed on the transfer equation system, and an initial thermal stress analysis model is obtained by establishing a correlation equation between the temperature field and the displacement field.

4. The method for analyzing thermal stress of a radio frequency front-end module according to claim 3, characterized in that: The method allocates stress compensation coefficients to each microelement in the initial thermal stress analysis model, establishes thermal stress transfer equations and boundary constraints between adjacent microelements, and solves the temperature field distribution matrix, including: Performing stress characteristic analysis on the initial thermal stress analysis model, and obtaining stress response data according to geometric parameters and material properties of the device structure; Based on the stress response data, a thermodynamic analysis is performed on each microelement, and a stress compensation coefficient calculation model is established according to the material parameters of the POI substrate material and the low-loss high-power electrode to obtain an initial compensation coefficient; Constructing a stress compensation matrix of each microelement according to the initial compensation coefficient, and performing correction calculation on the stress distribution under a preset working state to obtain a correction compensation coefficient matrix; Inputting the correction compensation coefficient matrix into the thermal stress transfer equation between adjacent micro-elements, establishing a stress transfer function according to the stress continuity principle, and obtaining a thermal stress transfer relationship; Constraining the temperature boundary in the thermal stress transfer relationship, and obtaining boundary constraint conditions according to the thermal expansion characteristics within the target temperature range; Based on the thermal stress transfer relationship and the boundary constraint conditions, a thermal stress control equation group is constructed, and the stress field is solved according to the stress equilibrium principle to obtain a stress distribution function; The stress distribution function is thermally coupled and solved, a temperature field calculation model is established according to the temperature-stress mapping relationship, the initial value of the temperature field is obtained, and the initial value of the temperature field is substituted into the heat conduction equation for iterative calculation, the temperature field value is updated according to the thermal stress distribution and boundary constraints, and the temperature field distribution matrix is ​​obtained.

5. The method for analyzing thermal stress of a radio frequency front-end module according to claim 1, characterized in that: The global equations are numerically solved, the principal stresses σ1, σ2, σ3 and their direction cosines are calculated according to the displacement-stress coordination conditions to obtain characteristic stress data, and the characteristic stress data are subjected to eigenvalue decomposition and vector reconstruction, and characteristic basis functions are established according to the main directions of the stress field to obtain the characteristic vectors of the stress field, including: Performing finite element discretization processing on the global equation group, constructing a stiffness matrix and a load vector according to the grid nodes of the filter and the duplexer, and obtaining a discretized numerical equation; A pre-processed conjugate gradient solver is constructed based on the discretized numerical equation, and iterative calculation is performed according to the boundary constraints of the POI substrate material and the low-loss high-power electrode to obtain a numerical solution of the displacement field; Performing gradient calculation on the numerical solution of the displacement field, obtaining the linear strain and shear strain components of each microelement according to the displacement-strain relationship, and obtaining the strain field tensor; Substituting the strain field tensor into the thermoelastic constitutive equation, calculating the stress components according to the temperature field under the preset power working state, and obtaining a six-component stress tensor; Establishing a characteristic equation for the six-component stress tensor, calculating eigenvalues ​​and eigenvectors according to stress invariants, and obtaining characteristic stress data, wherein the characteristic stress data includes principal stresses σ1, σ2, σ3 and direction cosines thereof; Perform stress coordinate transformation based on the characteristic stress data, construct a coordinate rotation matrix according to the direction cosines, obtain the stress distribution of the principal coordinate system, perform singular value decomposition on the stress distribution of the principal coordinate system, select orthogonal basis vectors according to the main direction of the stress field, and obtain the stress characteristic basis function; The stress characteristic basis function is combined with the principal stress component, and the complete stress field distribution is reconstructed according to the stress superposition principle to obtain the stress field characteristic vector.

6. The method for analyzing thermal stress of a radio frequency front-end module according to claim 5, characterized in that: The step of inputting the stress field feature vector into the thermal stress time evolution model, mapping the feature to a latent space through an encoder in the thermal stress time evolution model and reconstructing the feature through a decoder, and outputting thermal stress distribution prediction data comprises: Performing data pre-processing on the stress field feature vector, performing time-frequency decomposition on the thermal stress evolution feature by multi-scale wavelet transform, and obtaining feature subsets containing different frequency bands; Input the feature subset into the encoder network of the thermal stress time evolution model, reduce the feature dimension to the latent space through hierarchical convolution operation and downsampling mapping, and obtain the latent feature representation, wherein the encoder network includes a five-layer convolutional neural network structure, each convolution layer uses a LeakyReLU activation function and an Instance Normalization normalization layer, the convolution kernel size is 3×3, and the step size is 2; Input the potential feature representation into the time series feature extraction module of the thermal stress time evolution model, capture the time evolution law of thermal stress through bidirectional sequence learning, and obtain the time series correlation feature, the time series feature extraction module includes a three-layer bidirectional long short-term memory network, the hidden layer dimension is 256, and a Dropout layer is added between each layer, and the inactivation rate is 0.2; Inputting the temporal correlation features into the decoder network of the thermal stress time evolution model, and restoring the thermal stress distribution feature dimension through deconvolution operation and upsampling to obtain reconstructed feature data, wherein the decoder network includes a five-layer deconvolution layer structure, each deconvolution layer uses a ReLU activation function and a Batch Normalization normalization layer, and a jump connection is added between the third layer and the fourth layer; The reconstructed characteristic data is subjected to stress field mapping conversion, and thermal stress distribution prediction data is obtained through principal stress direction reconstruction and stress component synthesis.

7. A radio frequency front-end module thermal stress analysis system, characterized in that: Used to perform the RF front-end module thermal stress analysis method according to any one of claims 1 to 6, the RF front-end module thermal stress analysis system comprising: The acquisition module is used to collect the temperature data, stress data, deformation data and thermal distribution data of the RF front-end module, and synchronize and sort the data with the timestamp to obtain the thermal stress monitoring data sequence; Establish a module for determining the spatial discretization parameters according to the thermal stress monitoring data sequence, and dividing the RF front-end module into N×M×L micro-elements, establishing a transfer equation including the thermal conductivity, density and heat capacity parameters of each micro-element, and obtaining an initial thermal stress analysis model; An allocation module is used to allocate stress compensation coefficients based on each microelement in the initial thermal stress analysis model, establish thermal stress transfer equations and boundary constraints between adjacent microelements, and solve to obtain a temperature field distribution matrix; A solution module, used for calculating the normal stress and tangential stress of each microelement according to the temperature field distribution matrix, and establishing a high-order lumped parameter equation group to solve and obtain the stress field characteristic vector; The output module is used to input the stress field feature vector into the thermal stress time evolution model, map the feature to the latent space through the encoder in the thermal stress time evolution model and reconstruct it through the decoder, and output the thermal stress distribution prediction data.

8. A computer device, characterized in that: It comprises a memory and a processor, the memory stores a computer program that can be run on the processor, and the processor implements the RF front-end module thermal stress analysis method described in any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the processor executes the RF front-end module thermal stress analysis method as described in any one of claims 1 to 6.

Images