CN113158538A - Soft measurement method for heat flux density of complex-structure boiling surface - Google Patents
Soft measurement method for heat flux density of complex-structure boiling surface Download PDFInfo
- Publication number
- CN113158538A CN113158538A CN202110068179.4A CN202110068179A CN113158538A CN 113158538 A CN113158538 A CN 113158538A CN 202110068179 A CN202110068179 A CN 202110068179A CN 113158538 A CN113158538 A CN 113158538A
- Authority
- CN
- China
- Prior art keywords
- image
- convolution
- layer
- deconvolution
- temperature field
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000009835 boiling Methods 0.000 title claims abstract description 54
- 230000004907 flux Effects 0.000 title claims abstract description 23
- 238000000691 measurement method Methods 0.000 title claims abstract description 8
- 238000000034 method Methods 0.000 claims abstract description 54
- 238000009826 distribution Methods 0.000 claims abstract description 27
- 238000013528 artificial neural network Methods 0.000 claims abstract description 8
- 230000004913 activation Effects 0.000 claims abstract description 7
- 230000008569 process Effects 0.000 claims description 27
- 230000001052 transient effect Effects 0.000 claims description 13
- 238000012549 training Methods 0.000 claims description 12
- 230000006835 compression Effects 0.000 claims description 7
- 238000007906 compression Methods 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 7
- 239000000463 material Substances 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 abstract description 18
- 230000006870 function Effects 0.000 abstract description 16
- 238000004422 calculation algorithm Methods 0.000 abstract description 7
- 238000004088 simulation Methods 0.000 abstract description 6
- 238000013135 deep learning Methods 0.000 abstract description 4
- 238000013507 mapping Methods 0.000 abstract description 4
- 230000008901 benefit Effects 0.000 abstract description 3
- 238000010276 construction Methods 0.000 abstract description 3
- 238000005457 optimization Methods 0.000 abstract description 3
- 238000010438 heat treatment Methods 0.000 description 17
- 239000012071 phase Substances 0.000 description 11
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 8
- 238000006243 chemical reaction Methods 0.000 description 6
- 230000008859 change Effects 0.000 description 5
- 238000012546 transfer Methods 0.000 description 5
- 238000002939 conjugate gradient method Methods 0.000 description 4
- 238000005259 measurement Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- RYGMFSIKBFXOCR-UHFFFAOYSA-N Copper Chemical compound [Cu] RYGMFSIKBFXOCR-UHFFFAOYSA-N 0.000 description 2
- 238000005094 computer simulation Methods 0.000 description 2
- 229910052802 copper Inorganic materials 0.000 description 2
- 239000010949 copper Substances 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000008020 evaporation Effects 0.000 description 2
- 238000001704 evaporation Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 239000002086 nanomaterial Substances 0.000 description 2
- 230000006911 nucleation Effects 0.000 description 2
- 238000010899 nucleation Methods 0.000 description 2
- 238000003325 tomography Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000002591 computed tomography Methods 0.000 description 1
- 239000004020 conductor Substances 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011438 discrete method Methods 0.000 description 1
- 238000004070 electrodeposition Methods 0.000 description 1
- 239000012530 fluid Substances 0.000 description 1
- BTCSSZJGUNDROE-UHFFFAOYSA-N gamma-aminobutyric acid Chemical compound NCCCC(O)=O BTCSSZJGUNDROE-UHFFFAOYSA-N 0.000 description 1
- 230000017525 heat dissipation Effects 0.000 description 1
- 239000007791 liquid phase Substances 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000013021 overheating Methods 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 239000007790 solid phase Substances 0.000 description 1
- 238000002945 steepest descent method Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/28—Design optimisation, verification or simulation using fluid dynamics, e.g. using Navier-Stokes equations or computational fluid dynamics [CFD]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/08—Fluids
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/08—Thermal analysis or thermal optimisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/14—Force analysis or force optimisation, e.g. static or dynamic forces
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- Computer Hardware Design (AREA)
- Software Systems (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Algebra (AREA)
- Computing Systems (AREA)
- Fluid Mechanics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Mathematical Physics (AREA)
- Pure & Applied Mathematics (AREA)
- Investigating Or Analyzing Materials Using Thermal Means (AREA)
Abstract
The invention discloses a soft measurement method for heat flux density of a complex-structure boiling surface, wherein a developed artificial neural network comprises an input layer, a convolution layer, an activation layer, a deconvolution layer and an output layer, and the method comprises the following steps: obtaining a temperature field data set through forward simulation of a large number of heat conduction partial differential equations; convolving the data set; carrying out deconvolution on the obtained convolution characteristic graph; restoring the deconvolution feature map to obtain a high-resolution single-channel output image; a heat flow density distribution is obtained. The method has the advantages that the method needs to be reckoned for different practical problems in the traditional methodThe obtained mapping function F (q) is estimated based on a FluxNet deep learning algorithm according to different calculation conditionsB) Therefore, the method has the advantage that the overall calculation efficiency is greatly improved compared with the traditional inverse problem regularization method based on the construction of the adjoint operator and the solution of the partial differential equation constraint optimization problem.
Description
Technical Field
The invention belongs to the technical field of research on heat flow density soft measurement of industrial equipment and precision instruments, and particularly relates to a heat flow density soft measurement method for a boiling surface with a complex structure.
Technical Field
Pool boiling is an important boiling mode, which refers to the process of heating fluid in a container with a large volume to realize boiling, and has important application value in the industrial field, and the important application value relates to how to improve the performance of a phase-change heat exchanger to the maximum extent and how to maintain a proper temperature interval to ensure that the reaction process is carried out stably. However, when the boiling state is changed from nucleate boiling to film boiling, the heat dissipation capacity of the apparatus reaches an upper limit, meaning that a maximum critical heat flux density is reached. How to increase the maximum critical heat flux density is one of the scientific and engineering technical problems that researchers have paid special attention to in recent years. In actual production, taking a reaction furnace as an example, the heat flow and the temperature of the heating surface of the reaction furnace are controllable, but the unknown high-transient heat flow density distribution of the inner wall of the reaction furnace is difficult to directly measure. If the change of the heat flux density distribution of the inner wall of the reaction furnace can be monitored in real time, the local overheating phenomenon can be better avoided, and therefore the process can be controlled more accurately and efficiently. The working surface is oriented to the industrial process of pool boiling, which is widely applied, a three-dimensional transient heat conduction partial differential equation is taken as a mathematical model, and the transient temperature distribution of the heating surface inverts the high transient heat flow density distribution of the non-contact surface in the furnace. The problem belongs to the inverse problem of the ill-defined partial differential equation for identifying Neumann boundary conditions, and the research on the boiling mechanism and the complex inverse partial differential equation problem calculation method has important research and application significance for controlling the reaction conditions and optimizing and designing product materials in the assistance industry.
However, since the boiling phenomenon involves several complex sub-processes (such as heat conduction, heat convection, and phase change heat transfer), the boiling phenomenon of the whole stage has been difficult to accurately model, calculate, and predict under a unified framework. In the past decades, researchers have conducted many studies on boiling phenomena on a macro, meso, micro, and molecular scale. For example, the V.K.Dhir establishes a unified framework of nucleate boiling and transition boiling on the basis of a macroscopic geometric model of a steam system. On a mesoscopic scale, r.mei et al studied the growth process of bubbles on plates and films during heating. At the microscopic scale, p.stephan and j.hammer propose a theoretical prediction of the micro boundary layer, i.e. "most of the heat during boiling is transferred from the microscopic region of the three-phase contact line by evaporation". However, due to lack of experimental and theoretical evidence, most of the traditional methods are not fully verified, and at present, the parameters dominating boiling heat transfer are not completely clear, so that the full-stage boiling phenomenon still needs to be further studied.
The classical solving method of the class of standard three-dimensional transient heat conduction inverse problem in the pool boiling process comprises a Tikhonov regularization calculation method, a regularization calculation method based on the first class of Fredholm integral equation transformation, an iterative regularization calculation method based on a conjugate gradient method, a Landweber iterative regularization calculation method and a method based on L1And a norm variation calculation method and the like. When the method is used for solving the linear and semi-linear inverse problem with a small temperature change range in the boiling process, the conjugate gradient method has obvious solving efficiency advantage. Compared with a Newton iteration method which is commonly used for solving the nonlinear problem, the conjugate gradient method does not need to calculate a Hessian matrix and a corresponding inverse matrix, and the memory overhead in the calculation process is small; compared with the steepest descent method, the search direction selected in each iteration process of the conjugate gradient method is orthogonal to the previous direction, and the convergence process is greatly accelerated. In general, inverting the heat flow density distribution that is difficult to measure based on temperature information that is relatively easy to measure can be achieved by building a computational model facing the three-dimensional transient heat conduction inverse problem with the above algorithm.
However, conventional calculation methods such as regularization based on a conjugate iterative method require iterative calculation of a large number of positive partial differential equations, concomitant problems, and sensitivity problems over a continuous domain by discrete methods such as finite elements. The methods have the defects of overlarge calculation scale, low calculation precision, poor noise resistance and the like when the complicated heat transfer problem, particularly the nonlinear problem is solved, and the requirements of real-time measurement, control and optimization in the actual engineering problem are difficult to meet.
Disclosure of Invention
In view of the defects of the prior art, the invention aims to provide a soft measurement method for the heat flow density of a complex-structure boiling surface, which is based on a FluxNet deep learning algorithm to calculate the inverse problem of a partial differential equation of heat conduction. Aiming at the engineering problems, the provided method has the characteristics of higher calculation speed, strong nonlinear problem processing and good noise resistance under the condition of meeting the precision requirement, and is an efficient calculation model capable of replacing a complex partial differential equation abstract operator of the heat conduction positive/negative problem.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
a soft measurement method for heat flow density of a complex structure boiling surface comprises an artificial neural network, wherein the artificial neural network comprises an input layer, a convolution layer, an activation layer, an deconvolution layer and an output layer, and the method comprises the following steps:
s1, obtaining a temperature field data set through forward modeling of a large number of partial differential equations;
s2 convolving the data set;
s3 deconvoluting the convolution feature map obtained in step S2;
s4, restoring the deconvolution feature map to obtain a single-channel output image;
s5 obtains a heat flow density distribution.
In the input layer, the input is set to a time frame of tn，tn-1Temperature field and honeycomb material surface height information; in the output layer, the output is set to a time frame of tnThe heat flow density distribution at that moment.
It should be noted that, in the convolution, the convolution kernel weight can be obtained by error back propagation from the training data; defining a time frame tnThe temperature value of a data point on the time-temperature field is vx，yThe corresponding convolution kernel weight is wi，jConv () represents a convolution operator, the convolution process being shown as the formula: conv (v)x，y)＝∑j∑i wi，jvx-i，y-j (1)
Before deconvolution is introduced, the compression ratio between the input image and the feature image after the convolution operation needs to be determined, so that the resolution of the image is restored in the subsequent deconvolution process; the height of an image before convolution is defined as H, the width of the image is defined as W, the moving step length of a convolution kernel is defined as s, the boundary filling value of the image in the convolution process is defined as p, and the height of the image after convolution is defined as H 'and the width of the image is defined as W'. According to the moving principle of the convolution kernel, the relation of H, W and H ', W' is shown as the formula:
it should be noted that the temperature field information Θ at the observation boundary is constructed by the restored image size of the deconvolution layermTo unknown transient heat flow density distribution qBTo obtain a matrix theta with the temperature fieldmCorresponding heat flux q with uniform image resolutionBCorresponding images; the feature image is padded according to the image compression ratio in step S2, the padding value is set to 0, and the upsampled convolution kernel weight is obtained by error back propagation of the training data.
It should be noted that, the active layer adopts a ReLU function as an active function, and an expression formula of the ReLU function is shown as follows:
ReLU(x)＝max(x，0) (4)
the method has the beneficial effects that the FluxNet algorithm based on deep learning carries out model training once on a large amount of data obtained by forward modeling so as to construct the temperature field information theta on the boundary easy to observemReversely deducing unknown transient heat flow density distribution q on non-measurable boundaryBTo (3) is performed. Different from the situation that recalculation is needed for different practical problems in the traditional method, the mapping function F (q) estimated and obtained based on the FluxNet deep learning algorithmB) Can be repeatedly used. Accordingly, the invention is not limited to the specific embodiments describedCompared with the traditional inverse problem regularization method based on construction of adjoint operators and solving of partial differential equation constraint optimization problems, the volume calculation efficiency is greatly improved. The SIMilarity between the heat flow density distribution result obtained by the FluxNet algorithm inversion and the simulation test result is evaluated by using an SSIM (Structural SIMilarity) method, the average SIMilarity is 0.849, and the requirement of practical engineering application is met.
Drawings
FIG. 1 is a geometric model of a micro-nano porous structure reconstructed by CT scan data;
FIG. 2 is a schematic diagram of a FluxNet convolution operation and a convolution layer structure;
FIG. 3 is a schematic diagram of a FluxNet deconvolution operation and a deconvolution layer configuration;
FIG. 4 shows the result of FluxNet reverse predicting heat flux density distribution at several time points;
FIG. 5 is a flow chart of soft measurement of boiling surface heat flux density.
DETAILED DESCRIPTION OF EMBODIMENT (S) OF INVENTION
The present invention will be further described with reference to the accompanying drawings, and it should be noted that the present embodiment is based on the technical solution, and the detailed implementation and the specific operation process are provided, but the protection scope of the present invention is not limited to the present embodiment.
It should be noted that the physical model of the present invention is a three-dimensional transient heat conduction partial differential equation positive problem model and a three-dimensional transient heat conduction partial differential equation negative problem model, and the models are described as follows:
definition of thetamThe observed temperature of the heating surface during pool boiling is measured by the measured temperature thetamThe inverse of the contactless boiling surface heat flow density distribution can be regarded as a three-dimensional transient heat conduction partial differential equation inverse problem for identifying Neumann boundary conditions, and the control equation is as follows:
Θ(x，y，z，0)＝Θ0(x, y, z) in the three-dimensional solution domain omega (6)
As an initial condition, [ theta ]0(x, y, z) is the initial temperature distribution within the solution domain. Formula-is the Neumann boundary condition of the equation. In the above formula, p, CpAnd λ represents the density, heat capacity and heat conduction coefficient of the heat conductive material, respectively. Θ represents the temperature field in the solution domain as a function of the three-dimensional spatial coordinates x, y, z and time t in a cartesian rectangular coordinate system. Omega is a three-dimensional space solution domain, and n is a normal vector of a solution domain boundary. The heating boundary of the pool boiling heater is recorded as gammaHThe boiling surface is denoted by gammaBAnd the other boundaries are denoted as gammaR. qx denotes the constant heat flux density provided at the heating boundary, qBIs reported as the transient heat flux density distribution (unknown) on the boiling surface, qRThen the heat flow density information for the other boundaries is recorded. Observation duration of tf，T0TsideAnd h respectively represents the water temperature, the temperature of the side wall and the heat transfer coefficient, wherein the environment outside the boundary of the solution domain is constant to the boiling point.
The pool boiling inversion problem can be described as the thermophysical properties, initial conditions and the deflubution boundary Γ for a given material and water under the constraints of equation —BExternal Neumann boundary condition by heating the boundary ΓHThe instantaneous temperature data are observed to obtain thetamInversion of the mid-boiling boundary ΓBUpper unknown heat flow density distribution qB. The formula is implicit in terms of abstract operator equationsExpression equation-distribution of thermal current density qBWith measurable temperature field data thetamThe relationship between them.
F(qB)＝Θm (10)
Mathematical modelling from physical phenomena and assigning qBAnd carrying out simulation test on the function. And (4) carrying out forward modeling of a heat conduction equation based on the three-dimensional heat conduction models (5) - (9) to generate training data required by FluxNet.
In the pool boiling single bubble nucleation growth process, the annular region where the bubble surface is simultaneously in contact with the liquid phase and solid phase surfaces is called Three-phase contact lines (TPCL). According to the theory of micro boundary layers proposed by p.stephan, most of the heat is transferred from the three-phase contact line area (annular area) by evaporation during boiling heat exchange. Thus, it is assumed that the heat dissipated by the phase change is entirely from the three-phase contact line area. In the computational model in this work, the width of the high heat flux annular region was set to 12 μm (small enough), and bubble nucleation was cavity dependent and randomly distributed within the microcavity. The real computational domain geometric model is a real three-dimensional model with a micro-nano porous structure and is derived from CT tomography data of a subject group which deposits a honeycomb porous structure on the surface of smooth copper by using an electrodeposition method. The heating surface 1, boiling surface 2, side surface 3 and three-phase contact line 4 of the geometric model are shown in fig. 1. The heating surface 1 is heated by a stable heat source, and the upper surface of the boiling surface 2 is water which is being heated and boiled. In the geometric structure, the heating heat flux density of the heating surface 1 is set to be a constant value, and a three-phase contact line 4 passes through a heat flux density function qBThe component Q in the time domain (as shown in the formula) simulates the bubble generation rate and the heat flux density of the area under different working conditions.
Q＝A sin[kπ(t-θ)]+B (11)
Where t is the time in seconds(s) for applying the heat flow, and the heat flow density Q (W/m) is established according to the periodic characteristics of the pool boiling process2) The output power model per unit area is used for determining q in conjunction with a plurality of independent annular areasBThe unknown function, A, B, k, θ, is the heat flow density function model parameter. The side surface 3 and the boiling surface 2 in the model are setThe initial temperature of each geometry is the same as the boiling point of the surface-heated water. By changing the parameter of Q, a heat conduction calculation model of the micro-nano structure surface boiling process under a plurality of different working conditions can be obtained, and the heat conduction calculation model is further used for inversion simulation test.
As shown in fig. 5, the present invention is a soft measurement method for heat flux density of a complex-structured boiling surface, including an artificial neural network, wherein the artificial neural network includes an input layer, a convolution layer, an activation layer, a deconvolution layer, and an output layer, and the method includes the following steps:
s1, obtaining a temperature field data set through forward modeling of a large number of partial differential equations;
s2 convolving the data set;
s3 deconvoluting the convolution feature map obtained in step S2;
s4, restoring the deconvolution feature map to obtain a single-channel output image;
s5 obtains a heat flow density distribution.
In the input layer, the input is set to a time frame of tn，tn-1Temperature field and honeycomb material surface height information; in the output layer, the output is set to a time frame of tnHeat flow density distribution.
It should be noted that, in the convolution, the convolution kernel weight can be obtained by error back propagation from the training data; defining a time frame tnThe temperature value of a data point on the time-temperature field is vx，yThe corresponding convolution kernel weight is wi，jConv () represents the convolution operator, X, Y is the convolution kernel size (X ═ 2a +1, Y ═ 2b +1 or X ═ 2a, Y ═ 2b, where a, b are positive integers), and the convolution process is shown in the formula:
before deconvolution is introduced, the compression relationship between the input image and the feature image after the convolution operation needs to be determined, so that the resolution of the image is restored in the subsequent deconvolution process; defining a graph before convolutionThe image height is H, the width is W, the convolution kernel moving step length is s, the boundary filling value of the image in the convolution process is p, the height of the image after convolution is H ', and the width is W'. According to the moving principle of the convolution kernel, the relationship between H, W and H ', W' is shown in the formulas (2), (3):
it should be noted that the temperature field information Θ on the observation boundary is constructed by restoring the image size by the deconvolution layermTo unknown transient heat flow density distribution qBMapping relation is obtained to obtain a matrix theta with the temperature fieldmCorresponding heat flux q with uniform image resolutionBCorresponding images; the feature image is padded according to the image compression ratio in step S2, the padding value is set to 0, and the upsampled convolution kernel weight is obtained by error back propagation of the training data.
It should be noted that the active layer adopts a ReLU function as an active function, as shown in equation (4):
ReLU(x)＝max(x，0) (4)
examples
This example reconstructs CT tomography data of honeycomb porous copper samples and builds a simulated model of pool boiling by COMSOL Multiphysics software. The model obtains the temperature distribution of the heating surface and the heat flow density distribution of the boiling surface by simulating the nucleate boiling process of water at the concave cavity. As shown in the figure, the heating surface 1 of the model is in direct contact with a heat source, the boiling surface 2 and the side surface 3 are completely immersed in boiling water, and the three-phase contact line 4 of the model is a phase-change heat exchange area of a flat plate and the boiling water. Setting the heat exchange coefficient of the boiling surface and water to be 1000W/m2K, heating heat flux density of the heating surface of 5X 105W/m2. The dynamics of the change in time of the heat flux density component Q at the three-phase contact line 4 is determined by the formula, where the parameter C is 1, 2, 3, …, 100, and each model simulation time is 0.5 seconds(s). The high heat flux of the Q function in space corresponds in part to that of FIG. 1The three-phase contact line 4 is an annular area. The output simulated by this case is the temperature distribution Θ of the boiling surface 2m. In the embodiment, the case of the micro-nano structure surface pool boiling under 100 different working conditions is simulated as the data set of the convolutional neural network.
Inverse algorithm based on FluxNet network
This example performed 100 sets of positive problem simulations, where 80% of the data was used to train the FluxNet neural network model and 20% of the data was used to verify parameter accuracy. The FluxNet image recurrent neural network is constructed as follows:
input layer and output layer: for a pool boiling forward scenario, its input is set to time frame tn，tn-1The temperature field (t is more than or equal to 0 and less than or equal to 0.05s) and the surface height information of the honeycomb material are output and set as a time frame tnThe heat flux density distribution (as in fig. 4) to obtain a three-channel input, single-channel output network.
And (3) rolling layers: since the input layer is a three-channel image, the dimension of the convolution kernel is X × Y × 3, where X, Y is the size of the convolution kernel (X ═ 2a +1, Y ═ 2b +1 or X ═ 2a, Y ═ 2b, where a and b are positive integers), that is, the temperature field matrix Θ output in forward modelingmAll of the data in (1). The convolution process is shown in figure 2. The temperature value of a data point on the temperature field is vx，yThe corresponding convolution kernel weight is wi，jThe convolution process is shown in the formula
An active layer: the present embodiment selects the ReLU function as the activation function.
And (3) deconvolution layer: construction of Θ by restoring image size through deconvolutionm→qBMapping to obtain a matrix theta with the temperature fieldmCorresponding heat flux q with uniform image resolutionBCorresponding to the image. The feature image is padded according to the image compression ratio in step S2, the padding value is set to 0, and the upsampled convolution kernel weight is obtained by error back propagation of the training data. The deconvolution layer structure of the FluxNet network is shown in fig. 3.
The training process of the FluxNet network can normalize the input and output results by a formula, thereby accelerating convergence. In the FluxNet training process, the learning step length of error back propagation in the training process is set to be 2 x 10-6The loss function is L2Norm with regularization coefficient of 10-4。
Various modifications may be made by those skilled in the art based on the above teachings and concepts, and all such modifications are intended to be included within the scope of the present invention as defined in the appended claims.
Claims (6)
1. A soft measurement method for heat flow density of a complex-structure boiling surface comprises an artificial neural network, and is characterized in that the artificial neural network comprises an input layer, a convolution layer, an activation layer, a deconvolution layer and an output layer, and the method comprises the following steps:
s1, obtaining a temperature field data set through forward modeling of a large number of partial differential equations;
s2 convolving the data set;
s3 deconvoluting the convolution feature map obtained in step S2;
s4, restoring the deconvolution feature map to obtain a single-channel output image;
s5 obtains a heat flow density distribution.
2. The method of claim 1, wherein the input is set to a time frame t in the input layern，tn-1Temperature field and honeycomb material surface height information; in the output layer, the first and second layers are arranged in a common plane,the output is set to a time frame of tnThe heat flow density distribution at that moment.
3. The method of claim 1, wherein the convolution kernel weights are derived from training data by error back propagation when performing the convolution; defining a time frame tnThe temperature value of a data point on the time-temperature field is vx,yThe corresponding convolution kernel weight is wi,jConv (g) represents a convolution operator, the convolution process being represented by the following formula:
4. the method of claim 3, wherein before deconvolution is introduced, the compression relationship between the input image and the convolved feature image is determined to restore the resolution of the image in a subsequent deconvolution process; the height of an image before convolution is defined as H, the width of the image is defined as W, the moving step length of a convolution kernel is defined as s, the boundary filling value of the image in the convolution process is defined as p, and the height of the image after convolution is defined as H 'and the width of the image is defined as W'. According to the moving principle of the convolution kernel, the relationship between H, W and H ', W' is expressed by the following formula:
5. the method of claim 1, wherein the image size is recovered by the deconvolution layer to construct the temperature field information Θ at the observation boundarymTo unknown transient heatDensity of flow qBTo obtain a matrix theta with the temperature fieldmCorresponding heat flux density distribution q with uniform image resolutionBCorresponding images; the feature image is padded according to the image compression ratio in step S2, the padding value is set to 0, and the upsampled convolution kernel weight is obtained by error back propagation of the training data.
6. The method of claim 1, wherein the activation layer employs a ReLU function as the activation function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110068179.4A CN113158538A (en) | 2021-01-19 | 2021-01-19 | Soft measurement method for heat flux density of complex-structure boiling surface |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110068179.4A CN113158538A (en) | 2021-01-19 | 2021-01-19 | Soft measurement method for heat flux density of complex-structure boiling surface |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113158538A true CN113158538A (en) | 2021-07-23 |
Family
ID=76878499
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110068179.4A Pending CN113158538A (en) | 2021-01-19 | 2021-01-19 | Soft measurement method for heat flux density of complex-structure boiling surface |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113158538A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114091372A (en) * | 2021-11-23 | 2022-02-25 | 西安交通大学 | Method for back-calculating inner boundary heat flow density of two-dimensional annular region |
| CN115983137A (en) * | 2023-01-31 | 2023-04-18 | 西安交通大学 | Turbine flow field prediction method based on similarity principle and deep learning and related device |
| CN116910428A (en) * | 2023-08-03 | 2023-10-20 | 大连理工大学 | Space-time dynamic system soft measurement method for automatically determining partial differential equation structure |
| CN115983137B (en) * | 2023-01-31 | 2024-05-31 | 西安交通大学 | Turbine flow field prediction method and related device based on similarity principle and deep learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2006242923A (en) * | 2005-03-07 | 2006-09-14 | Central Res Inst Of Electric Power Ind | Co2 flux measuring method for forest |
| US20170323694A1 (en) * | 2014-12-16 | 2017-11-09 | Joint Stock Company "Atomenergoproekt" | Water-Cooled Water-Moderated Nuclear Reactor Core Melt Cooling and Confinement System |
| CN110097519A (en) * | 2019-04-28 | 2019-08-06 | 暨南大学 | Double supervision image defogging methods, system, medium and equipment based on deep learning |
| CN110956111A (en) * | 2019-11-22 | 2020-04-03 | 苏州闪驰数控系统集成有限公司 | Artificial intelligence CNN, LSTM neural network gait recognition system |
-
2021
- 2021-01-19 CN CN202110068179.4A patent/CN113158538A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2006242923A (en) * | 2005-03-07 | 2006-09-14 | Central Res Inst Of Electric Power Ind | Co2 flux measuring method for forest |
| US20170323694A1 (en) * | 2014-12-16 | 2017-11-09 | Joint Stock Company "Atomenergoproekt" | Water-Cooled Water-Moderated Nuclear Reactor Core Melt Cooling and Confinement System |
| CN110097519A (en) * | 2019-04-28 | 2019-08-06 | 暨南大学 | Double supervision image defogging methods, system, medium and equipment based on deep learning |
| CN110956111A (en) * | 2019-11-22 | 2020-04-03 | 苏州闪驰数控系统集成有限公司 | Artificial intelligence CNN, LSTM neural network gait recognition system |
Non-Patent Citations (2)
| Title |
|---|
| 娄继琳: "大功率真空电子器件内部温度推算及测量技术", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 衡益 等: "池沸腾强化传热中的三维瞬态热传导反问题", 《科学通报》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114091372A (en) * | 2021-11-23 | 2022-02-25 | 西安交通大学 | Method for back-calculating inner boundary heat flow density of two-dimensional annular region |
| CN114091372B (en) * | 2021-11-23 | 2024-02-02 | 西安交通大学 | Method for back-calculating boundary heat flux density in two-dimensional annular region |
| CN115983137A (en) * | 2023-01-31 | 2023-04-18 | 西安交通大学 | Turbine flow field prediction method based on similarity principle and deep learning and related device |
| CN115983137B (en) * | 2023-01-31 | 2024-05-31 | 西安交通大学 | Turbine flow field prediction method and related device based on similarity principle and deep learning |
| CN116910428A (en) * | 2023-08-03 | 2023-10-20 | 大连理工大学 | Space-time dynamic system soft measurement method for automatically determining partial differential equation structure |
| CN116910428B (en) * | 2023-08-03 | 2024-03-22 | 大连理工大学 | Space-time dynamic system soft measurement method for automatically determining partial differential equation structure |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| De Boer et al. | Mesh deformation based on radial basis function interpolation | |
| CN113051831B (en) | Modeling method and thermal error control method for thermal error self-learning prediction model of machine tool | |
| Wu | An approach combining body-fitted grid generation and conjugate gradient methods for shape design in heat conduction problems | |
| CN113158538A (en) | Soft measurement method for heat flux density of complex-structure boiling surface | |
| Liu et al. | Acoustic tomography reconstruction method for the temperature distribution measurement | |
| Jahangiry et al. | Combination of Isogeometric analysis and level-set method in topology optimization of heat-conduction systems | |
| Stavropoulos et al. | Warping in SLM additive manufacturing processes: estimation through thermo-mechanical analysis | |
| Wang et al. | Solving of two-dimensional unsteady-state heat-transfer inverse problem using finite difference method and model prediction control method | |
| Yoshida et al. | Lattice Boltzmann method for the convection–diffusion equation in curvilinear coordinate systems | |
| Aquino et al. | Self-learning finite elements for inverse estimation of thermal constitutive models | |
| Chen et al. | Identification of transient boundary conditions with improved cuckoo search algorithm and polynomial approximation | |
| CN116451584A (en) | Thermal stress prediction method and system based on neural network | |
| Edalatifar et al. | A deep learning approach to predict the flow field and thermal patterns of nonencapsulated phase change materials suspensions in an enclosure | |
| Hughes et al. | An adaptive reduced order model for the angular discretization of the Boltzmann transport equation using independent basis sets over a partitioning of the space‐angle domain | |
| Lu et al. | Three-dimensional temperature field inversion calculation based on an artificial intelligence algorithm | |
| Xu et al. | Physics-informed neural networks for studying heat transfer in porous media | |
| Du et al. | Super Resolution Generative Adversarial Networks for Multi-Fidelity Pressure Distribution Prediction | |
| Kumar et al. | Modeling moving-boundary problems of solidification and melting adopting an arbitrary Lagrangian-Eulerian approach | |
| Kumar et al. | Evaluation of artificial neural network in data reduction for a natural convection conjugate heat transfer problem in an inverse approach: experiments combined with CFD solutions | |
| Gu et al. | A fast inversion approach for the identification of highly transient surface heat flux based on the generative adversarial network | |
| Rahideh et al. | Two-dimensional inverse transient heat conduction analysis of laminated functionally graded circular plates | |
| Nokhosteen et al. | Melting behavior prediction of latent heat storage materials: A multi-pronged solution | |
| Mozafari-Shamsi et al. | Application of the ghost fluid lattice Boltzmann method to moving curved boundaries with constant temperature or heat flux conditions | |
| Heng et al. | Efficient reconstruction of local heat fluxes in pool boiling experiments by goal-oriented adaptive mesh refinement | |
| Akbari et al. | Blending machine learning and sequential data assimilation over latent spaces for surrogate modeling of Boussinesq systems |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210723 |