A laser loading parameter identification method based on temperature change rate

Abstract

A laser loading parameter identification method based on temperature change rate comprises: establishing a finite element transient thermal analysis model of a honeycomb sandwich structure, the honeycomb sandwich structure comprising a front panel, a honeycomb sandwich core and a back panel; loading a thermal load corresponding to a plurality of laser loading parameters on the front panel of the honeycomb sandwich structure, extracting the space-time characteristic sequence temperature response characteristic information of the front panel, performing data processing to obtain the temperature change rate of the whole front panel at different times, and then composing a plurality of training data sets; inputting the plurality of training data sets into a laser loading parameter identification training model, and when the iterative training reaches a set number of iteration steps, the laser loading parameter identification training model is used to obtain the optimal laser loading parameter identification training model after training. The present application solves the problem that the existing technology is complex in solving the non-steady-state heat conduction inverse problem and cannot identify the characteristic parameters of the thermal load.

Description

[0001] Field

[0002] The present invention belongs to the direction of solving the inverse problem of heat conduction by neural network, and specifically relates to a laser loading parameter identification method based on temperature change rate. Background Art

[0003] In recent years, with the continuous development of aerospace technology, the pursuit of the ultimate in aerospace has continued to increase. The damage and destructive behavior of thermal loads such as aerodynamic heating, radiation heating, and atmospheric reentry heating on aerospace device structures has attracted increasing attention, making it particularly important to identify the characteristic parameters of these thermal loads. Thermal loads can cause material degradation and generate thermal stresses, which in turn accelerate the rapid failure of structures. Accurately inverting the characteristic parameters of unknown thermal loads by extracting the thermal response of the structure is crucial for assessing the subsequent damage effects on the structure and making informed decisions.

[0004] Honeycomb sandwich structures, widely used in the aerospace field, are very complex to solve the inverse problem of unsteady-state heat conduction. The Sequence Function Method (SFSM) is one of the important methods for solving the inverse problem of unsteady-state heat conduction, but its inversion results are highly dependent on the selection of future time steps. It is necessary to select an appropriate future time step to accurately invert its surface heat flux. Moreover, this method cannot identify the characteristic parameters of the thermal load. Summary of the Invention

[0005] The present invention provides a laser loading parameter identification method based on temperature change rate, which solves the problem in the prior art that the characteristic parameters of the heat load are complex when solving the inverse problem of unsteady-state heat conduction and cannot be identified.

[0006] A laser loading parameter identification method based on temperature change rate, comprising:

[0007] Establishing a finite element transient thermal analysis model of a honeycomb sandwich structure for simulation, wherein the finite element transient thermal analysis model includes a front panel, a honeycomb sandwich core, and a back panel;

[0008] Simulating the loading of thermal loads corresponding to a plurality of laser loading parameters on the front panel of the finite element transient thermal analysis model, and extracting temperature response characteristic information of a spatiotemporal characteristic sequence corresponding to each laser loading parameter on the front panel;

[0009] Performing data processing on the temperature response characteristic information of the spatiotemporal characteristic sequence to obtain the temperature change rate of the entire front panel when the finite element transient thermal analysis model is at different times, thereby forming multiple sets of training data sets consisting of temperature change rates;

[0010] The temperature change rate changes dynamically with the continuous application of thermal load. A four-layer ConvLSTM is added to the basic network model to extract the spatial and temporal features of the three-dimensional spatiotemporal sequence temperature field to obtain a laser loading parameter identification training model for inferring the laser loading parameters based on the temperature response of the target object being laser loaded.

[0011] A plurality of training data sets are input into the laser loading parameter identification training model. When the iterative training reaches a set number of iteration steps, the laser loading parameter identification training model obtains the optimal laser loading parameter identification training model after training.

[0012] In some preferred embodiments, the temperature change rate processing formula is:

[0013]

[0014] where ξ i j is the temperature change rate of the i-th node at the j-th moment, T i,j+1 ,T i,j is the temperature response characteristic information of the spatiotemporal characteristic sequence at the j+1th moment and the jth moment of the i-th node, and Δt is the sampling time interval.

[0015] In some preferred embodiments, the optimal laser loading parameter identification training model is a 16-layer deep learning network with two output parameters.

[0016] In some preferred embodiments, the optimal laser loading parameter identification training model identifies and outputs a plurality of laser loading parameters corresponding to the heat load on the front panel of the finite element transient thermal analysis model: laser diameter parameter and laser power parameter.

[0017] In some preferred embodiments, the deep learning network includes four layers of ConvLSTM for extracting spatial and temporal features in the three-dimensional spatiotemporal sequence temperature field, one layer of Conv3D for matching the three-dimensional tensor output by the ConvLSTM layer, a layer of BatchNormalization for standardizing the output after each layer of ConvLSTM and Conv3D, and three layers of Dropout for regularization, one layer of MaxPooling and one layer of GlobalAverangePooling3D for downsampling the three-dimensional data, and the last fully connected layer has two neurons to convert the output data into the required dimension.

[0018] Compared with the existing technology, the present invention has the following beneficial effects: the present invention proposes a laser loading parameter identification method based on temperature change rate, which is a thermal load characteristic parameter inversion method based on the temperature response of the spatiotemporal characteristic sequence and deep learning. It is based on the time change rate characteristic data of the temperature field response of the spatiotemporal characteristic sequence and combined with a deep network model of multi-parameter regression. It can accurately and quickly identify the various characteristic parameters of the thermal load, and provide effective data support and technical guidance for subsequent damage assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other implementation drawings based on the provided drawings without inventive effort.

[0020] Figure 1 Schematic diagram of the process of the laser parameter intelligent identification method of the present invention;

[0021] Figure 2 Schematic diagram of the laser loading parameter identification training model in the present invention;

[0022] Figure 3 Schematic diagram of a finite element transient thermal analysis model of a honeycomb sandwich structure according to an embodiment of the present invention;

[0023] Figure 4 Schematic diagram of a honeycomb sandwich core model according to an embodiment of the present invention;

[0024] Figure 5 This is a temperature field diagram showing the time variation after a heat load is applied to a finite element transient thermal analysis model of a honeycomb sandwich structure according to an embodiment of the present invention;

[0025] In the figure: 1-ConvLSTM layer, 2-BatchNormalization layer, 3-Dropout layer, 4-Conv3D layer, 5-MaxPooling layer, 6-GlobalAverangePooling3D layer, 7-Fully connected layer, 8-Output layer. DETAILED DESCRIPTION

[0026] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0027] The present invention provides a laser loading parameter identification method based on temperature change rate, comprising:

[0028] Step 1. Establish a finite element transient thermal analysis model of the honeycomb sandwich structure for simulation. The finite element transient thermal analysis model includes a front panel, a honeycomb sandwich core, and a back panel. The honeycomb sandwich core is located between the front panel and the back panel. The loading of random thermal loads and the extraction of temperature responses can be completed within the finite element transient thermal analysis model.

[0029] Thermal loads corresponding to a variety of laser loading parameters are simulated and loaded on the front panel of the finite element transient thermal analysis model, and temperature response characteristic information of a spatiotemporal characteristic sequence corresponding to each laser loading parameter on the front panel is extracted.

[0030] Step 2. Process the temperature response characteristic information of the spatiotemporal characteristic sequence to obtain the temperature change rate of the entire front panel when the finite element transient thermal analysis model is at different times, thereby forming multiple training data sets consisting of temperature change rates.

[0031] In order to improve the recognition accuracy of thermal load parameters, the present invention uses the temperature change rate to establish a training data set. The processing formula of the temperature change rate is:

[0032]

[0033] Where ξ i j is the temperature change rate of the i-th node at the j-th moment, T i,j+1 ,T i,j is the temperature response of the i-th node at time j+1 and time j, and Δt is the sampling time interval.

[0034] Step 3. The temperature change rate changes dynamically with the continuous loading of the thermal load. A four-layer ConvLSTM is added to the basic network model to extract the spatial and temporal features of the three-dimensional spatiotemporal sequence temperature field to obtain a laser loading parameter identification training model for inferring the laser loading parameters based on the temperature response of the target object loaded by the laser.

[0035] Step 4. Input multiple sets of training data sets into the laser loading parameter identification training model. When the iterative training reaches the set number of iteration steps, the laser loading parameter identification training model obtains the optimal laser loading parameter identification training model after training, and then identifies the laser loading parameters of the random thermal load on the front panel of the honeycomb sandwich structure.

[0036] Aiming at the problem of identifying and regressing the characteristic parameters of the thermal load loaded on the front panel of the honeycomb sandwich structure, based on the various characteristics of the data in the training data set, in order to solve the inverse problem of heat transfer, the network model in this application is a deep learning network with 16 intermediate layers. The optimal laser loading parameter identification training model includes two neurons, which output laser diameter parameters and laser power parameters. Specifically, it includes four layers of ConvLSTM, which are used to extract the spatial and temporal features in the temperature field of the three-dimensional spatiotemporal sequence, a layer of Conv3D to match the three-dimensional tensor output by the ConvLSTM layer, and a layer of BatchNormalization to standardize the output after each layer of ConvLSTM and Conv3D, and three layers of Dropout for regularization. A layer of MaxPooling and a layer of GlobalAverangePooling3D are used to downsample the three-dimensional data. Since the regressed values ​​are two values, the last fully connected layer is two neurons, which converts the output data into the required dimension.

[0037] In this application, a finite element transient thermal analysis model for honeycomb sandwich structures is proposed, taking into account the thermal loads of various laser loading parameters. In this model, the thermal load parameters (position, diameter, and power) are set as random numbers. After establishing the finite element transient thermal analysis model for the honeycomb sandwich structure, thermal loads corresponding to various laser loading parameters are simulated on the front panel. The temporal and spatial characteristic sequence temperature response information of the front panel corresponding to each laser loading parameter under the thermal load parameters is extracted.

[0038] To improve the accuracy of thermal load parameter identification, the temperature response characteristic information of the spatiotemporal feature sequence was processed to obtain the temperature change rate of the entire front panel at different times in the finite element transient thermal analysis model. This data set, consisting of temperature change rates, was then used to form multiple training data sets. These training data sets were then imported into the laser loading parameter identification training model for iterative training to identify the laser loading parameters for random thermal loads on the front panel of the honeycomb sandwich structure. Compared with the identification results obtained by directly iteratively using temperature response data, the data processed with temperature change rates showed more stable and faster model convergence in the deep learning training of the network model.

[0039] Because the data in the training data set is a three-dimensional tensor, this application adds a ConvLSTM (ConvolutionalLSTM) layer to extract the features of the spatiotemporal feature sequence data. One layer is often not enough to extract its response features well. Under the premise of ensuring the recognition accuracy of the model after training, the network calculation amount is optimized, so four layers of ConvLSTM are added; a layer of Conv3D is added to integrate the three-dimensional tensor; in the process of training the model, problems such as gradient disappearance often occur, so the BatchNormalization layer is added, which can not only solve the problem of gradient disappearance, but also standardize the weights, optimize the network gradient flow, speed up the training process and improve the accuracy. High performance: BatchNormalization is added after each ConvLSTM and Conv3D layer to standardize their output. To enhance the model's ability to generalize data and prevent overfitting, three layers of Dropout are added to the network for regularization. One layer of MaxPooling, one layer of GlobalAverangePooling3D, and one layer of fully connected layer can integrate the output into the desired dimension and shape. The activation function uses the advanced activation function LeakyRelu, the optimizer is Adadelta, and the loss function and regression evaluation indicator are both root mean square error (mse).

[0040] Figure 2 Schematic diagram of the ConvLSTM network designed and built for this application, where 1 represents the ConvLSTM layer, 2 represents the BatchNormalization layer, 3 represents the Dropout layer, 4 represents the Conv3D layer, 5 represents the MaxPooling layer, 6 represents the GlobalAverangePooling3D layer, 7 represents the fully connected layer, and 8 represents the output layer. Because the output is two values, there are two neurons.

[0041] In order to verify the effectiveness of the laser loading parameter identification method based on temperature change rate provided in this application, an embodiment is now provided.

[0042] Example 1

[0043] The thermal load was applied to the front panel of the honeycomb sandwich structure. To verify the content of this application, the calculations used three diameters (20mm, 25mm, and 30mm) and five power levels (60W, 70W, 80W, 90W, and 100W), totaling 15 different operating conditions (Table 1). The thermal load was applied randomly to the front panel of the honeycomb sandwich structure. The thermal load parameters were inverted based on the temperature field responses corresponding to different operating conditions.

[0044] Table 1 Working condition settings

[0045] Working condition number Spot diameter / mm Power / W Working condition number Spot diameter / mm Power / W Case 1 20 60 Case 9 25 90 Case 2 20 70 Case 10 25 100 Case 3 20 80 Case 11 30 60 Case 4 20 90 Case 12 30 70 Case 5 20 100 Case 13 30 80 Case 6 25 60 Case 14 30 90 Case 7 25 70 Case 15 30 100 Case 8 25 80

[0046] by Figure 3 The honeycomb sandwich structure shown is the research object, and its size is 150mm×150mm. The thickness of the upper and lower panels are both h f =1mm, cell height h c =18mm, regular hexagonal honeycomb sandwich core (such as Figure 4 As shown) the outer length l = 3.7mm, the wall thickness is t c =0.1mm; the upper and lower panels and the honeycomb sandwich core are made of aluminum alloy, and the material parameters are: elastic modulus E = 7.17×10 4 MPa, Poisson's ratio μ=0.33, material density ρ=2700kg / m 3 , surface emissivity ε = 0.3, the thermal conductivity k and specific heat capacity c of aluminum alloy are shown in Table 2.

[0047] Table 2 Thermophysical properties of aluminum alloy

[0048]

[0049] For loading under different working conditions, the ABAQUS external subroutine can be used to set the required heat flux density and spot diameter. Since we know the loading power, the heat flux density can be calculated according to the formula To calculate. Figure 5 This is the temperature field diagram that changes with time when the power P is 100W, the heat flux diameter is D = 20mm, and the heat load loading time is 30s.

[0050] The temperature response characteristic information of the spatiotemporal feature sequence obtained using the temperature change rate proposed in this application was preprocessed and then trained in the laser loading parameter identification training model. The training and validation sets were randomly divided in the dataset at a ratio of 8:2. After training, the thermal load characteristic parameters were identified using data not in the original dataset and with the thermal load applied at random locations. The relative errors of the identification results are as follows (Table 3):

[0051] Table 3 Recognition results

[0052]

[0053]

[0054] It can be seen from Table 3 that the method proposed in this application can more accurately identify the parameters of unknown heat loads loaded at any position, and can quickly identify the parameters when the model is trained.

[0055] The above embodiments are merely exemplary embodiments of the present application and are not intended to limit the scope of the present application. The scope of protection of the present application is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present application within the essence and scope of protection of the present application, and such modifications or equivalent substitutions shall also be deemed to fall within the scope of protection of the present application.

Claims

1. A laser loading parameter identification method based on temperature change rate, characterized in that: include: Establishing a finite element transient thermal analysis model of a honeycomb sandwich structure for simulation, wherein the finite element transient thermal analysis model includes a front panel, a honeycomb sandwich core, and a back panel; Simulating the loading of thermal loads corresponding to a plurality of laser loading parameters on the front panel of the finite element transient thermal analysis model, and extracting temperature response characteristic information of a spatiotemporal characteristic sequence corresponding to each laser loading parameter on the front panel; Performing data processing on the temperature response characteristic information of the spatiotemporal characteristic sequence to obtain the temperature change rate of the entire front panel when the finite element transient thermal analysis model is at different times, thereby forming multiple sets of training data sets consisting of temperature change rates; The processing formula for the temperature change rate is: in is the temperature change rate of the i-th node at the j-th moment, T i,j+1 ,T i,j is the temperature response characteristic information of the spatiotemporal characteristic sequence at the j+1th moment and the jth moment of the i-th node, and Δt is the sampling time interval; The temperature change rate changes dynamically with the continuous application of thermal load. A four-layer ConvLSTM is added to the basic network model to extract the spatial and temporal features of the three-dimensional spatiotemporal sequence temperature field to obtain a laser loading parameter identification training model for inferring the laser loading parameters based on the temperature response of the target object being laser loaded. A plurality of training data sets are input into the laser loading parameter identification training model. When the iterative training reaches a set number of iteration steps, the laser loading parameter identification training model obtains the optimal laser loading parameter identification training model after training.

2. The laser loading parameter identification method based on temperature change rate according to claim 1, characterized in that: The optimal laser loading parameter identification training model is a 16-layer deep learning network with two output parameters.

3. The laser loading parameter identification method based on temperature change rate according to claim 2, characterized in that: The optimal laser loading parameter identification training model identifies and outputs a plurality of laser loading parameters corresponding to the heat load on the front panel of the finite element transient thermal analysis model: a laser diameter parameter and a laser power parameter.

4. The laser loading parameter identification method based on temperature change rate according to claim 2, characterized in that: The deep learning network includes four layers of ConvLSTM for extracting spatial and temporal features in the temperature field of a three-dimensional spatiotemporal sequence, a layer of Conv3D for matching the three-dimensional tensor output by the ConvLSTM layer, a layer of BatchNormalization for output standardization after each layer of ConvLSTM and Conv3D, three layers of Dropout for regularization, a layer of MaxPooling and a layer of GlobalAverangePooling3D for downsampling the three-dimensional data, and the last fully connected layer with two neurons to convert the output parameters into the required dimensions.

Images