Target surface temperature inversion method and device based on PINN

Abstract

The application provides a PINN-based target surface temperature inversion method and device, wherein the method comprises the following steps: constructing a physical model of a target dynamic light-thermal response; constructing a fully connected neural network with time parameters, space parameters and target physical property parameters as inputs, and constructing a loss function of the neural network by using the physical model, and optimizing network parameters of the neural network through cyclic iteration to make the loss function tend to 0, so as to obtain a PINN model that can be used to predict a target surface temperature field; acquiring infrared radiance measured data of the target dynamic light-thermal response; the infrared radiance measured data is obtained under set time parameters and space parameters; and determining the target surface temperature according to the infrared radiance measured data and the PINN model. According to the scheme, the calculation amount in the target surface temperature inversion process can be reduced, and the speed of measuring the target surface temperature by using the infrared temperature measurement method can be improved.

Description

Technical Field

[0001] This invention relates to the field of infrared data processing technology, and in particular to a method and apparatus for inverting target surface temperature based on PINN (Physics-informed Neural Network). Background Technology

[0002] The surface temperature of a target can be obtained through infrared thermometry. During infrared thermometry, the detected infrared radiation is the result of the coupling effect between the target's surface emissivity and its temperature field. Therefore, the emissivity of the target needs to be measured beforehand, and then the surface temperature is obtained by subtracting the emissivity from the infrared radiation. However, when telemetry is performed on non-cooperative targets, it is impossible to decouple the emissivity or surface temperature for direct measurement.

[0003] Currently, the separation of emissivity and infrared signal in infrared thermometry can be achieved based on dynamic spectral evolution time series inversion. Specifically, firstly, a physical model of the target photothermal coupling and a numerical calculation model are established, and then the target photothermal properties are obtained based on experimental data using an inversion algorithm.

[0004] However, the inversion process typically involves calculating the sensitivity coefficient matrix in the input parameters of the forward problem. This requires continuously updating the input parameters and performing numerous numerical calculations for the forward problem. Due to the large computational load, it is impossible to quickly invert information such as the target surface emissivity and surface temperature, which is not conducive to rapid infrared thermometry. Summary of the Invention

[0005] This invention provides a target surface temperature inversion method and apparatus based on PINN, which can reduce the computational load in the target surface temperature inversion process and improve the speed of infrared thermometry for measuring target surface temperature.

[0006] In a first aspect, embodiments of the present invention provide a target surface temperature inversion method based on PINN, comprising:

[0007] Construct a physical model of the target's dynamic photothermal response;

[0008] A fully connected neural network is constructed with time parameters, spatial parameters, and target physical property parameters as inputs. The loss function of the neural network is constructed using the physical model. The network parameters of the neural network are optimized by iterative loops to make the loss function approach 0, thus obtaining a PINN model that can be used to predict the surface temperature field of the target.

[0009] Acquire measured infrared radiance data of the target's dynamic photothermal response; the measured infrared radiance data is obtained under set time and spatial parameters;

[0010] The target surface temperature is determined based on the measured infrared radiance data and the PINN model.

[0011] Secondly, embodiments of the present invention also provide a target surface temperature inversion device based on PINN, comprising:

[0012] The physical model building unit is used to build a physical model of the target's dynamic photothermal response;

[0013] The PINN model building unit is used to construct a fully connected neural network with time parameters, spatial parameters and target physical property parameters as inputs, and to construct the loss function of the neural network using the physical model. The network parameters of the neural network are optimized by iterative loops to make the loss function approach 0, thereby obtaining a PINN model that can be used to predict the temperature field of the target surface.

[0014] The measured data acquisition unit is used to acquire measured infrared radiance data of the target's dynamic photothermal response; the measured infrared radiance data is obtained under set time and space parameters;

[0015] The surface temperature determination unit is used to determine the target surface temperature based on the measured infrared radiance data and the PINN model.

[0016] Thirdly, embodiments of the present invention also provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the method described in any embodiment of this specification.

[0017] Fourthly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the methods described in any embodiment of this specification.

[0018] This invention provides a target surface temperature inversion method and apparatus based on PINN. By using the Physical Information Neural Network (PINN), a PINN model that can be trained to predict the target surface temperature field can be generated without the need for training the neural network with tag data. The PINN model allows for rapid prediction of the target surface temperature without extensive computation. Therefore, this solution reduces the computational load in the target surface temperature inversion process and increases the speed of infrared thermometry for measuring target surface temperature. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of a target surface temperature inversion method based on PINN provided in an embodiment of the present invention;

[0021] Figure 2 This is a hardware architecture diagram of an electronic device provided in an embodiment of the present invention;

[0022] Figure 3 This is a structural diagram of a target surface temperature inversion device based on PINN provided in an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0024] Please refer to Figure 1 This invention provides a target surface temperature inversion method based on PINN, the method comprising:

[0025] Step 100: Construct a physical model of the target's dynamic photothermal response;

[0026] Step 102: Construct a fully connected neural network with time parameters, spatial parameters and target physical property parameters as inputs, and use the physical model to construct the loss function of the neural network. Optimize the network parameters of the neural network through iterative loops to make the loss function approach 0, thereby obtaining the PINN model that can be used to predict the temperature field of the target surface.

[0027] Step 104: Obtain measured infrared radiance data of the target's dynamic photothermal response; the measured infrared radiance data is obtained under set time and spatial parameters;

[0028] Step 106: Determine the target surface temperature based on the measured infrared radiance data and the PINN model.

[0029] In this embodiment of the invention, by using the Physical Information Neural Network (PINN), a PINN model that can be trained to predict the temperature field of a target surface can be generated without the need for training the neural network with tag data. The PINN model allows for rapid prediction of the target surface temperature without requiring extensive computation. Therefore, this solution reduces the computational load in the target surface temperature retrieval process and increases the speed of infrared thermometry for measuring target surface temperature.

[0030] The following description Figure 1 The execution method for each step is shown.

[0031] First, for step 100, a physical model of the target dynamic photothermal response is constructed.

[0032] In this embodiment of the invention, considering the target's photothermal coupling heat transfer under normal conditions, its temperature distribution is affected by solar radiation heat flow, ambient long-wave infrared radiation heat flow, internal heat sources, and material thermal conductivity. Considering unsteady-state conditions, the overall heat balance control equation and boundary conditions can be used as the physical model of the target's dynamic photothermal response. This physical model is as follows:

[0033]

[0034]

[0035] T(t,r)=T D (t,r)r∈Γ D

[0036] T(0,r)=T0(r)

[0037] Where t is the time parameter, r is the spatial parameter, T(t,r) is the target surface temperature field under the constraints of the time and spatial parameters, ρ is the material density, c is the specific heat capacity, λ is the thermal conductivity, and Q is the thermal conductivity. s (t,r) represents the internal heat source constrained by time and space parameters, and n w E is the wall normal vector. sun (t,r), E ref (t,r), E bgd (t,r), E w (t,r) and E oth (t, r) represent the absorbed solar irradiance heat flux, absorbed Earth albedo solar heat flux, absorbed background irradiance heat flux, heat flux radiated outward from the wall, and other radiative heat flux, respectively, under the constraints of time and space parameters. D (t,r) represents the temperature distribution given on the essential boundary, and T0(r) represents the initial temperature field; Γ N For Neumann type boundaries, Γ D For Dirichlet type boundaries.

[0038] Then, for step 102, a fully connected neural network is constructed with time parameters, spatial parameters and target physical property parameters as inputs, and the loss function of the neural network is constructed using the physical model. The network parameters of the neural network are optimized by iterative loops to make the loss function approach 0, thus obtaining the PINN model that can be used to predict the temperature field of the target surface.

[0039] Wherein, the spatial parameter r = (x,y,z), (x,y,z) are the spatial coordinates of the target being measured; the target characteristic parameters can be λ, α, ε, l, where α is the reflectivity of the target surface, ε is the emissivity of the target surface, and l is the geometric parameter.

[0040] In this embodiment of the invention, a fully connected neural network with (t,x,y,z,λ,α,ε,l) as input is constructed. This neural network can be regarded as a trial function representing the target surface temperature field T(t,x,y,z; λ,α,ε,l) under different parameters (time parameters, spatial parameters, and target physical property parameters), denoted as T. NN This refers to the target surface temperature field that can be predicted by a neural network. Specifically, this neural network is a Physical Information Neural Network (PINN). A key feature of PINN is that it can be trained without requiring labeled data.

[0041] To train a PINN model that can predict the temperature field of a target surface, a loss function (Loss) for the neural network needs to be constructed using the physical information contained in the physical model. This loss function serves as a constraint during the training process, ensuring that the constructed neural network satisfies physical interpretability. The loss function (Loss) is as follows:

[0042]

[0043]

[0044] In the above equation, the right side, from left to right, represents the domain loss term, the heat flux boundary loss term, the isothermal boundary loss term, and the initial loss term, N. r N b1 N b2 ω0 and N0 represent the number of spatiotemporal boundary sampling points within the solution domain, respectively. As the weight, T NN The target surface temperature field predicted by the neural network.

[0045] Furthermore, the overall loss function can be optimized sequentially using the Adam optimization algorithm and L-BFGS.

[0046] In this embodiment of the invention, during the training of the neural network, the network parameters can be optimized iteratively to make the loss function (Loss) approach 0. When the loss function of the neural network approaches 0, the neural network can be considered an approximation of the true temperature of the target surface, i.e., T(t,x,y,z;λ,α,ε,l)≈T NN (t,x,y,z;λ,α,ε,l).

[0047] Thus, the neural network training is complete, and a PINN model that can be used to predict the temperature field of a target surface is obtained.

[0048] Finally, explanations are provided for step 104, "obtaining measured infrared radiance data of the target's dynamic photothermal response; the measured infrared radiance data is obtained under set time and space parameters," and step 106, "determining the target surface temperature based on the measured infrared radiance data and the PINN model."

[0049] In this embodiment of the invention, the photothermal response of the target can be dynamically monitored using detectors such as infrared thermal imagers to obtain measured infrared radiance data L of the target's dynamic photothermal response. λ (t,x,y,z).

[0050] In step 106, specifically, it may include:

[0051] S1, based on the measured infrared radiance data L λ The time and space parameters corresponding to (t,x,y,z) are input into the PINN model using assumed target material property parameters to obtain the predicted infrared radiance L′ output by the PINN model. λ (t,x,y,z); where r=(x,y,z); (x,y,z) are the spatial coordinates of the target being measured;

[0052] In this process, after assuming the target physical property parameters, the same time and space parameters as the measured data, along with the assumed target physical property parameters, can be input into the PINN model. In this way, the PINN can output the corresponding predicted infrared radiance.

[0053] S2, Define the error norm as The optimal target physical property parameters that minimize the error norm are obtained using an optimization algorithm;

[0054] The optimization algorithm can be gradient descent, particle swarm optimization, etc. It continuously optimizes to obtain the optimal target physical property parameters (λ, α, ε, l) that minimize the error norm.

[0055] S3. Calculate the target surface temperature based on the target surface emissivity in the optimal target physical property parameters.

[0056] The surface temperature of the target can be calculated based on the target surface emissivity ε and the blackbody radiation law.

[0057] This invention utilizes a Physical Information Neural Network (PINN) to train a numerical prediction model of the target's dynamic photothermal response without requiring tag data for training. This significantly improves the speed of predicting the target's dynamic photothermal response and solves the problem in traditional infrared thermometry where the unknown target surface emissivity necessitates extensive computation and is unsuitable for rapid infrared thermometry when using inverse problem methods to invert the target surface emissivity and temperature.

[0058] Compared with existing technologies, the technical solution proposed in this invention, where the photothermal response of the target is predicted by a trained neural network, allows for near-instantaneous calculation of the forward problem. This significantly accelerates the measurement of target surface temperature using infrared thermometry when the emission from an unknown target surface is unknown.

[0059] like Figure 2 , Figure 3 As shown, this embodiment of the invention provides a target surface temperature inversion device based on PINN. The device embodiment can be implemented through software, hardware, or a combination of both. From a hardware perspective, as... Figure 2 The diagram shown is a hardware architecture diagram of an electronic device for a target surface temperature inversion device based on PINN, provided in an embodiment of the present invention. (Except for...) Figure 2 In addition to the processor, memory, network interface, and non-volatile memory shown, the electronic device in the embodiment may also include other hardware, such as a forwarding chip responsible for processing packets. Taking software implementation as an example, such as... Figure 3 As shown, a device in a logical sense is formed by the CPU of its electronic device reading the corresponding computer program from non-volatile memory into memory and running it. This embodiment provides a target surface temperature inversion device based on PINN, comprising:

[0060] Physical model building unit 301 is used to build a physical model of the target's dynamic photothermal response;

[0061] The PINN model building unit 302 is used to build a fully connected neural network with time parameters, spatial parameters and target physical property parameters as inputs, and to build the loss function of the neural network using the physical model. The network parameters of the neural network are optimized by iterative loops to make the loss function approach 0, so as to obtain a PINN model that can be used to predict the temperature field of the target surface.

[0062] The measured data acquisition unit 303 is used to acquire measured infrared radiance data of the target's dynamic photothermal response; the measured infrared radiance data is obtained under set time and space parameters;

[0063] The surface temperature determination unit 304 is used to determine the target surface temperature based on the measured infrared radiance data and the PINN model.

[0064] In one embodiment of the present invention, the physical model is:

[0065]

[0066]

[0067] T(t,r)=T D (t,r)r∈Γ D

[0068] T(0,r)=T0(r)

[0069] Where t is the time parameter, r is the spatial parameter, T(t,r) is the target surface temperature field under the constraints of the time and spatial parameters, ρ is the material density, c is the specific heat capacity, λ is the thermal conductivity, and Q is the thermal conductivity. s (t,r) represents the internal heat source constrained by time and space parameters, and n w E is the wall normal vector. sun (t,r), E ref (t,r), E bgd (t,r), E w (t,r) and E oth (t, r) represent the absorbed solar irradiance heat flux, absorbed Earth albedo solar heat flux, absorbed background irradiance heat flux, heat flux radiated outward from the wall, and other radiative heat flux, respectively, under the constraints of time and space parameters. D (t,r) represents the temperature distribution given on the essential boundary, and T0(r) represents the initial temperature field; Γ N For Neumann type boundaries, Γ D For Dirichlet type boundaries.

[0070] In one embodiment of the present invention, the loss function Loss is:

[0071]

[0072]

[0073] Where, N r N b1 N b2ω0 and N0 represent the number of spatiotemporal boundary sampling points within the solution domain, respectively. As the weight, T NN The target surface temperature field predicted by the neural network.

[0074] In one embodiment of the present invention, the surface temperature determining unit is specifically used for: determining the surface temperature based on the measured infrared radiance data L. λ The time and space parameters corresponding to (t,x,y,z) are input into the PINN model using assumed target material property parameters to obtain the predicted infrared radiance L′ output by the PINN model. λ (t,x,y,z); where r=(x,y,z); (x,y,z) are the spatial coordinates of the target being measured; the error norm is defined as The optimal target physical property parameters that minimize the error norm are obtained using an optimization algorithm; the target surface temperature is calculated based on the target surface emissivity in the optimal target physical property parameters.

[0075] It is understood that the structures illustrated in the embodiments of the present invention do not constitute a specific limitation on a PINN-based target surface temperature inversion device. In other embodiments of the present invention, a PINN-based target surface temperature inversion device may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0076] The information interaction and execution process between the modules in the above-mentioned device are based on the same concept as the method embodiment of the present invention, and the specific details can be found in the description of the method embodiment of the present invention, and will not be repeated here.

[0077] This invention also provides an electronic device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements a target surface temperature inversion method based on PINN according to any embodiment of this invention.

[0078] This invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform a target surface temperature inversion method based on PINN according to any embodiment of this invention.

[0079] Specifically, a system or apparatus equipped with a storage medium may be provided, on which software program code implementing the functions of any of the embodiments described above is stored, and the computer (or CPU or MPU) of the system or apparatus may read and execute the program code stored in the storage medium.

[0080] In this case, the program code read from the storage medium can itself implement the function of any of the above embodiments, and therefore the program code and the storage medium storing the program code constitute part of the present invention.

[0081] Examples of storage media used to provide program code include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, program code can be downloaded from a server computer via a communication network.

[0082] Furthermore, it should be clear that not only can the program code read by the computer be executed, but also the operating system or other components operating on the computer can be instructed based on the program code to perform some or all of the actual operations, thereby realizing the function of any of the embodiments described above.

[0083] Furthermore, it is understood that the program code read from the storage medium is written to the memory set in the expansion board inserted into the computer or to the memory set in the expansion module connected to the computer. Then, based on the instructions of the program code, the CPU or other components installed on the expansion board or expansion module execute some and all of the actual operations, thereby realizing the function of any of the above embodiments.

[0084] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0085] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as ROM, RAM, magnetic disk, or optical disk.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A target surface temperature inversion method based on PINN, characterized in that, include: Construct a physical model of the target's dynamic photothermal response; A fully connected neural network is constructed with time parameters, spatial parameters, and target physical property parameters as inputs. The loss function of the neural network is constructed using the physical model. The network parameters of the neural network are optimized by iterative loops to make the loss function approach 0, thus obtaining a PINN model that can be used to predict the surface temperature field of the target. Acquire measured infrared radiance data of the target's dynamic photothermal response; the measured infrared radiance data is obtained under set time and spatial parameters; The target surface temperature is determined based on the measured infrared radiance data and the PINN model.

2. The method according to claim 1, characterized in that, The physical model is as follows: T(t,r)=T D (t,r)r∈Γ D T(0,r)=T0(r) Where t is the time parameter, r is the spatial parameter, T(t,r) is the target surface temperature field under the constraints of the time and spatial parameters, ρ is the material density, c is the specific heat capacity, λ is the thermal conductivity, and Q is the thermal conductivity. s (t,r) represents the internal heat source constrained by time and space parameters, and n w E is the wall normal vector. sun (t,r), E ref (t,r), E bgd (t,r), E w (t,r) and E oth (t, r) represent the absorbed solar irradiance heat flux, absorbed Earth albedo solar heat flux, absorbed background irradiance heat flux, heat flux radiated outward from the wall, and other radiative heat flux, respectively, under the constraints of time and space parameters. D (t,r) represents the temperature distribution given on the essential boundary, and T0(r) represents the initial temperature field; Γ N For Neumann type boundaries, Γ D For Dirichlet type boundaries.

3. The method according to claim 2, characterized in that, The loss function Loss is: Where, N r N b1 N b2 ω0 and N0 represent the number of spatiotemporal boundary sampling points within the solution domain, respectively. As the weight, T NN The target surface temperature field predicted by the neural network.

4. The method according to claim 3, characterized in that, The step of determining the target surface temperature based on the measured infrared radiance data and the PINN model includes: Based on the measured infrared radiance data L λ The time and space parameters corresponding to (t,x,y,z) are input into the PINN model using assumed target material property parameters to obtain the predicted infrared radiance L′ output by the PINN model. λ (t,x,y,z); where r=(x,y,z); (x,y,z) are the spatial coordinates of the target being measured; Define the error norm as The optimal target physical property parameters that minimize the error norm are obtained using an optimization algorithm; The target surface temperature is calculated based on the target surface emissivity in the optimal target physical property parameters.

5. A target surface temperature inversion device based on PINN, characterized in that, include: The physical model building unit is used to build a physical model of the target's dynamic photothermal response; The PINN model building unit is used to construct a fully connected neural network with time parameters, spatial parameters and target physical property parameters as inputs, and to construct the loss function of the neural network using the physical model. The network parameters of the neural network are optimized by iterative loops to make the loss function approach 0, thereby obtaining a PINN model that can be used to predict the temperature field of the target surface. The measured data acquisition unit is used to acquire measured infrared radiance data of the target's dynamic photothermal response; the measured infrared radiance data is obtained under set time and space parameters; The surface temperature determination unit is used to determine the target surface temperature based on the measured infrared radiance data and the PINN model.

6. The apparatus according to claim 5, characterized in that, The physical model is as follows: T(t,r)=T D (t,r)r∈Γ D T(0,r)=T0(r) Where t is the time parameter, r is the spatial parameter, T(t,r) is the target surface temperature field under the constraints of the time and spatial parameters, ρ is the material density, c is the specific heat capacity, λ is the thermal conductivity, and Q is the thermal conductivity. s (t,r) represents the internal heat source constrained by time and space parameters, and n w E is the wall normal vector. sun (t,r), E ref (t,r), E bgd (t,r), E w (t,r) and E oth (t, r) represent the absorbed solar irradiance heat flux, absorbed Earth albedo solar heat flux, absorbed background irradiance heat flux, heat flux radiated outward from the wall, and other radiative heat flux, respectively, under the constraints of time and space parameters. D (t,r) represents the temperature distribution given on the essential boundary, and T0(r) represents the initial temperature field; Γ N For Neumann type boundaries, Γ D For Dirichlet type boundaries.

7. The apparatus according to claim 6, characterized in that, The loss function Loss is: Where, N r N b1 N b2 ω0 and N0 represent the number of spatiotemporal boundary sampling points within the solution domain, respectively. As the weight, T NN The target surface temperature field predicted by the neural network.

8. The apparatus according to claim 7, characterized in that, The surface temperature determination unit is specifically used for: determining the surface temperature based on the measured infrared radiance data L. λ The time and space parameters corresponding to (t,x,y,z) are input into the PINN model using assumed target material property parameters to obtain the predicted infrared radiance L′ output by the PINN model. λ (t,x,y,z); where r=(x,y,z); (x,y,z) are the spatial coordinates of the target being measured; the error norm is defined as The optimal target physical property parameters that minimize the error norm are obtained using an optimization algorithm; the target surface temperature is calculated based on the target surface emissivity in the optimal target physical property parameters.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor, when executing the computer program, implements the method as described in any one of claims 1-4.

10. A computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the method of any one of claims 1-4.

Images