A dimension lifting method for nuclear power plant main pipeline temperature reconstruction based on ultrasonic array

Abstract

This invention relates to the field of temperature field reconstruction, specifically a method for enhancing the dimensionality of temperature reconstruction in the main pipeline of a nuclear power plant based on an ultrasonic array. It addresses the problem that existing technologies, with sparse topology structures, cannot meet the quantity and quality requirements of three-dimensional reconstruction in terms of reconstruction accuracy. The invention includes the following steps: temperature measurement and sub-temperature zone division based on an ultrasonic array; sparse dictionary training: solving for the time-of-flight (TOF) of acoustic rays in the sub-temperature zones based on physical equations; extending the measurement period; and building a neural network to enhance the dimensionality of temperature reconstruction. This invention utilizes a low-dimensional, limited number of "TOF-path" correspondences, trains a sparse dictionary, and introduces physical equations as prior information to achieve the inverse solution of nonlinear measurement equations, thereby achieving high-precision, refined reconstruction of the two-dimensional temperature field. By extending the measurement period and building a neural network, the correspondence between "A and B temperature surfaces" and the "three-dimensional pipeline" is established, thus enhancing the dimensionality of the temperature field reconstruction.

Description

Technical Field

[0001] This invention relates to the field of temperature field reconstruction, and in particular to a method for increasing the dimensionality of temperature reconstruction of the main pipeline of a nuclear power plant based on an ultrasonic array. Background Technology

[0002] A nuclear power reactor is a high-power-density, high-operation-parameter, and high-safety-requirement device that rapidly and efficiently converts nuclear, thermal, and kinetic energy. Timely and accurate monitoring of the nuclear energy release and heat transfer status of the reactor core is crucial for ensuring the safe and reliable operation of the reactor. Based on the structure and operating principle of pressurized water reactors, the temperature of the coolant in the primary loop of the nuclear reactor system directly reflects the nuclear power and core heat transfer status, making it a core parameter for nuclear reactor power control and safety protection.

[0003] Acoustic measurement, as a non-contact measurement method, has the advantages of fast response speed, high safety, and flexible layout, and can characterize the temperature and flow rate information on the acoustic channel line.

[0004] Ultrasonic array topology is a prerequisite for accurately reconstructing the spatial distribution of measurement parameters. By extracting information in an "array-like" manner, the spatial distribution of the heat flow field can be reconstructed, accurately characterizing the heat flow information of the primary loop, realizing integrated ultrasonic detection of the coolant heat flow field, and improving the operating efficiency of nuclear reactors.

[0005] The patent, titled "A Method for Reconstructing Temperature Field in Ultrasonic CT Based on Compressed Sensing," discloses a method that combines sparse reconstruction and time delay estimation to improve the accuracy of two-dimensional temperature field reconstruction. It calculates the temperature vector using acoustic-slow vectors and combines interpolation to obtain the temperature distribution and temperature field reconstruction, thereby achieving a significant improvement in the accuracy of two-dimensional temperature field reconstruction. Taking into account the influence of temperature on velocity, and given a certain scale of ultrasonic transducers, it can improve the reconstruction accuracy of the two-dimensional temperature field to a certain extent.

[0006] Due to the limited scale of ultrasonic arrays in the main pipelines of nuclear power plants, the path-time information extracted by the currently used ultrasonic arrays is limited. Only the average temperature of the measured section can be estimated using measured data, and the two-dimensional temperature distribution cannot be accurately reconstructed. Furthermore, the temperature field in the main pipelines of nuclear power plants is highly dynamic and non-uniform. The average temperature of the two-dimensional section cannot accurately characterize the dynamic evolution process of the space-time flow-thermal field coupling of the reactor primary coolant, and cannot meet the measurement requirements for reactor system control and protection.

[0007] In this sparse topology, in order to ensure the accuracy of temperature field reconstruction, the common method used in nuclear power plants is to arrange parallel sound rays on a two-dimensional plane. However, this arrangement method cannot meet the requirements of the number and quality of sound rays for three-dimensional temperature field reconstruction. The current reconstruction accuracy cannot accurately reconstruct the two-dimensional temperature field distribution, nor can it estimate the three-dimensional temperature field.

[0008] There is an urgent need for a temperature reconstruction method based on ultrasonic arrays that can solve the above problems. Summary of the Invention

[0009] This invention proposes a method for enhancing the dimensionality of temperature reconstruction of the main pipeline of a nuclear power plant based on an ultrasonic array, which solves the problem that the reconstruction accuracy cannot meet the quantity and quality requirements of three-dimensional reconstruction under the sparse topology structure in the prior art.

[0010] The technical solution of this invention is implemented as follows: A method for dimensionality enhancement of temperature reconstruction in the main pipeline of a nuclear power plant based on an ultrasonic array, comprising the following steps:

[0011] Step 1: Temperature measurement and sub-temperature zone division of surface A and surface B based on ultrasonic array; temperature measurement and reconstruction are performed by arranging the ultrasonic array with parallel sound rays, and temperature measurement is performed by evenly distributing the transceiver transducers in two groups on two mutually perpendicular elliptical surfaces A and B; by arranging them in parallel, the direct sound rays are made to be on the same plane, and the direct wave sound rays on surface A and surface B are parallel to each other; the sub-temperature zone of each surface is divided into m×n sub-temperature zones;

[0012] Step Two: Sparse Dictionary Training:

[0013] temperature field Divided into M pixels, where ,dictionary It is an M×N dimensional matrix, where N is the number of atoms in the dictionary. arrive The sparse transformation process is as follows sparse vectors (N×1 dimension) is the dictionary field Sparse expressions in; The number of non-zero elements This represents its sparsity. The smaller the value, The stronger the sparsity.

[0014] Ansys numerical simulation was used to design several unimodal, bimodal, and multimodal time-varying temperature fields as training sets to establish a sparse dictionary. This allows for the calculation of the result based on the temperature field. Calculate the optimal sparse vector To satisfy ; Sparse dictionary The establishment process is as follows:

[0015] Set up a dictionary The two-dimensional mesh form of the i-th atom is as follows The two-dimensional wavelet decomposition process is defined as follows:

[0016] Step 1: Define a one-dimensional low-pass filter based on the selected wavelet basis type. K is the filter length; high-pass filter Its orthogonality is derived from the wavelet basis. Derivation of the low-pass filter operator It applies a low-pass filter to the signal. Perform downsampling with a step size of 2; high-pass filter operator Apply a high-pass filter to the signal And perform downsampling with a step size of 2; decompose the number of layers. ;

[0017] Step 2: [The sentence is incomplete and requires more context.] Apply each line separately and The row low-pass matrix is ​​obtained. High-pass matrix Two intermediate matrices;

[0018] Step 3: Apply the following to each column of the intermediate matrix from Step 2. and The four types of coefficients for word decomposition are obtained—low-frequency approximation coefficients. ,right Column filtering is used to obtain the overall temperature distribution trend of atoms; horizontal detail coefficients. ,right Column filtering is used to characterize the temperature gradient change in the vertical direction; vertical detail coefficients. ,right The high-pass filter is used to characterize the temperature gradient change in the horizontal direction; the diagonal detail coefficients are also used. ,right The temperature abrupt changes along the tilt direction are characterized by high-pass filtering.

[0019] Step 4: Repeat Step 3 until the Lth level of decomposition is completed; after the i-th atom undergoes L levels of wavelet decomposition, a set of multi-scale coefficients is obtained, which can be expressed as... , It is a wavelet basis type;

[0020] For each temperature field, the following wavelet bases are used: 'db4', 'db8', 'db12', 'db16', 'sym4', 'sym8', 'sym12', 'coif1', 'coif3', 'coif5', 'bior1.3', 'bior3.5', 'haar', with the mean of the L1 / L2 norm ratio of the wavelet coefficients. Mean values ​​of wavelet coefficients and Gini coefficients The average effective sparsity required to retain 99% of the energy The optimal wavelet basis is selected as an indicator; the calculation method is as follows:

[0021]

[0022] in, It is a one-dimensional vector obtained by flattening all coefficients after the i-th atom undergoes L-level wavelet decomposition. Describing the L1 norm, Describing the L2 norm, It is the wavelet coefficient vector of the i-th atom. The ascending order results It meets the conditions The smallest integer; It is the final scoring metric for determining the optimal wavelet basis. This is the penalty coefficient;

[0023] During the learning process, three loss functions are used as follows:

[0024]

[0025] Among them, the reconstruction loss term Used to measure the sparse dictionary Sparse vectors The difference between the constructed temperature field and the real temperature field; to enhance the sparse representation ability of each dictionary atom in its optimal transform domain, dictionary constraint terms are introduced. ;

[0026] Step 3: Solving for the ray-flight time in the sub-temperature region based on the physical equations

[0027] The relationship between temperature and sound velocity of liquid coolant was fitted using the IF97 formula based on steam and liquid water, yielding the following functional relationship:

[0028]

[0029] in, T represents the speed of sound; T represents the temperature.

[0030] Based on the m×n=N sub-temperature zones divided in step one, the flight time of the i-th sound ray can be determined. :

[0031]

[0032] in, It is the length of the i-th sound wave path in the j-th sub-temperature region. The acoustic slowness function of the j-th sub-temperature region;

[0033] Step 4: Expand the measurement cycle and establish a dataset of "A and B surface temperatures - pipe temperature";

[0034] Step 5: Build a neural network to establish the correspondence between "A and B surface temperatures and pipe temperatures" to achieve an increase in the dimensionality of temperature reconstruction.

[0035] Furthermore, step three specifically involves:

[0036] By introducing the physical equations of sound propagation as soft constraints, a loss function that integrates data-driven and physical information is constructed as follows:

[0037]

[0038] Definitions of each loss function:

[0039] Data fitting term This directly constrains the consistency between the predicted flight time and the measured data;

[0040] physical constraints Introducing the heat conduction equation as a physical prior; It represents the rate of change of temperature over time, approximated by the temperature difference between adjacent moments; Indicates the thermal diffusivity. It is the Laplace operator for temperature, representing the two-dimensional rate of change of temperature in space. It is a heat diffusion term;

[0041] vocal constraint terms Introducing acoustic constraint terms to force the temperature of non-acoustic regions. Average temperature around the adjacent sound ray Fluctuations occur.

[0042] Furthermore, step four specifically involves: the transducer transmission and reception strategy is as follows: the transducer group's transmission and reception are completed within 1 second, and the temperature field reconstruction of surfaces A and B within that second is completed based on the measured transit time; the 1-second measurement period is extended to s; based on the Ansys numerical simulation platform, an ultrasonic measurement experiment is conducted to simulate the time-varying temperature field under Num different temperature inlet conditions, obtaining the Num group transit time matrix:

[0043]

[0044] in, This represents the transit time of the i-th transducer at time t; the transit time matrix consists of T row vectors. Substituting the t-th row vector into step two yields the temperature distribution of surfaces A and B at time t. ;

[0045] Create a dataset with the following input features:

[0046]

[0047] The 3D pipe temperature field distribution from the CFD simulation is output as follows:

[0048] .

[0049] Furthermore, step five specifically involves:

[0050] After creating the dataset, build the neural network model;

[0051] The above equation applies between the flight time and the temperatures of surfaces A and B.

[0052] A multi-path convolutional neural network is used to simulate the data. The Num sets of data are divided into training and testing sets according to a certain ratio and fed into the neural network for learning. This allows us to obtain the three-dimensional pipe temperature field distribution results at each measurement time.

[0053] This invention discloses a method for enhancing the dimensionality of temperature reconstruction in the main pipeline of a nuclear power plant based on an ultrasonic array. It utilizes a low-dimensional, limited number of "TOF-path" correspondences, generating diverse temperature fields as prior information through simulation software. This is used to train a sparse dictionary capable of capturing the highly dynamic and non-uniform characteristics of the temperature field. Furthermore, physical equations are introduced as prior information to achieve the inverse solution of nonlinear measurement equations, thereby enabling high-precision, refined reconstruction of the two-dimensional temperature field and reflecting the true temperature of the primary circuit. This lays the foundation for improving the safety and economy of nuclear reactor operation. By extending the measurement cycle and building a neural network to establish the correspondence between "A and B temperature surfaces" and "three-dimensional pipeline," the dimensionality of the temperature field reconstruction is enhanced. Attached Figure Description

[0054] 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 only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 Schematic diagram of the main pipeline ultrasonic array;

[0056] Figure 2 : Schematic diagram of parallel arrangement;

[0057] Figure 3 Flowchart of sparse dictionary training;

[0058] Figure 4 Flowchart for calculating two-dimensional temperature field distribution;

[0059] Figure 5 Flowchart of multi-path convolutional neural network simulation. Detailed Implementation

[0060] 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 only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] A method for dimensionality enhancement in temperature reconstruction of main pipelines in nuclear power plants based on ultrasonic arrays includes the following steps:

[0062] Step 1: As Figure 1 The schematic diagram of the main pipeline ultrasonic array shows temperature measurements and sub-temperature zone divisions based on the ultrasonic array on surfaces A and B. Due to the small radius of the main pipeline, the size and number of ultrasonic transducers are limited. Temperature measurement and reconstruction are performed using a parallel acoustic ray arrangement within the ultrasonic array. The transceiver transducers are evenly distributed in two groups on two mutually perpendicular elliptical surfaces, surfaces A and B, for temperature measurement. The green parallel lines constitute surface A, and the red parallel lines constitute surface B. Figure 2 As shown in the parallel arrangement diagram, the direct sound rays are arranged in parallel so that they lie on the same plane, with the four direct sound ray waves on planes A and B being parallel to each other; each plane is divided into m×n sub-temperature zones; to ensure the sound ray coverage of each sub-temperature zone, in Figure 1 The schematic diagram of the main pipeline ultrasonic array should ensure that there are ;

[0063] Step Two: Sparse Dictionary Training:

[0064] like Figure 3 As shown in the sparse dictionary training flowchart, common temperature fields such as single-peaked, double-peaked, and triple-peaked fields exhibit sparsity in the discrete Fourier transform domain, wavelet transform domain, and symmetric wavelet transform domain. Assume that the temperature field... Divided into M ( ) pixels, dictionary It is an M×N dimensional matrix, where N is the number of atoms in the dictionary. arrive The sparse transformation process is as follows sparse vectors (N×1 dimension) is the dictionary field Sparse expressions in [the context of sparse expressions]. The number of non-zero elements This represents its sparsity. The smaller the value, The stronger the sparsity of the dictionary; that is, the stronger the sparsity of the dictionary. right The better the sparsity of the dictionary, the more advantageous it is for high-precision sparse reconstruction. Compressed sensing theory shows that the stronger the sparsity of the dictionary, the more beneficial it is for high-precision sparse reconstruction.

[0065] The temperature field inside the reactor's main duct exhibits significant non-uniformity and high dynamic characteristics, with its spatial distribution constantly evolving over time, leading to marked differences in the temperature field structure at different times. Furthermore, various typical temperature distributions display different sparsity characteristics in different transform domains. To achieve high-precision reconstruction of the two-dimensional temperature field, an overcomplete dictionary needs to be constructed. This dictionary can learn and embed the sparse representation characteristics of common temperature distribution features in their respective optimal transform domains. Ansys numerical simulations were used to design several unimodal, bimodal, and multimodal time-varying temperature fields as training sets to establish the sparse dictionary. This allows for the calculation of the result based on the temperature field. Calculate the optimal sparse vector To satisfy ; Sparse dictionary The establishment process is as follows:

[0066] Set up a dictionary The two-dimensional mesh form of the i-th atom is as follows The two-dimensional wavelet decomposition process is defined as follows:

[0067] Step 1: Define a one-dimensional low-pass filter based on the selected wavelet basis type. K is the filter length; high-pass filter Its orthogonality is derived from the wavelet basis. Derivation of the low-pass filter operator It applies a low-pass filter to the signal. Perform downsampling with a step size of 2; high-pass filter operator Apply a high-pass filter to the signal And perform downsampling with a step size of 2; decompose the number of layers. ;

[0068] Step 2: [The sentence is incomplete and requires more context.] Apply each line separately and The row low-pass matrix is ​​obtained. High-pass matrix Two intermediate matrices;

[0069] Step 3: Apply the following to each column of the intermediate matrix from Step 2. and The four types of coefficients for word decomposition are obtained—low-frequency approximation coefficients. ,right Column filtering is used to obtain the overall temperature distribution trend of atoms; horizontal detail coefficients. ,right Column filtering is used to characterize the temperature gradient change in the vertical direction; vertical detail coefficients. ,right The high-pass filter is used to characterize the temperature gradient change in the horizontal direction; the diagonal detail coefficients are also used. ,right The temperature abrupt changes along the tilt direction are characterized by high-pass filtering.

[0070] Step 4: Repeat Step 3 until the Lth level of decomposition is completed; after the i-th atom undergoes L levels of wavelet decomposition, a set of multi-scale coefficients is obtained, which can be expressed as... , It is a wavelet basis type;

[0071] During dictionary learning, considering the different sparsity characteristics of different feature temperature fields, the following wavelet bases are used for each temperature field: 'db4', 'db8', 'db12', 'db16', 'sym4', 'sym8', 'sym12', 'coif1', 'coif3', 'coif5', 'bior1.3', 'bior3.5', and 'haar', based on the mean of the L1 / L2 norm ratio of the wavelet coefficients. Mean values ​​of wavelet coefficients and Gini coefficients The average effective sparsity required to retain 99% of the energy The optimal wavelet basis is selected as an indicator; the calculation method is as follows:

[0072]

[0073] in, It is a one-dimensional vector obtained by flattening all coefficients after the i-th atom undergoes L-level wavelet decomposition. Describing the L1 norm, Describing the L2 norm, It is the wavelet coefficient vector of the i-th atom. The ascending order results It meets the conditions The smallest integer; It is the final scoring metric for determining the optimal wavelet basis. This is the penalty coefficient;

[0074] During the learning process, three loss functions are used as follows:

[0075]

[0076] Among them, the reconstruction loss term Used to measure the sparse dictionary Sparse vectors The difference between the constructed temperature field and the real temperature field; to enhance the sparse representation ability of each dictionary atom in its optimal transform domain, dictionary constraint terms are introduced. ;

[0077] Specifically, by performing wavelet decomposition on each dictionary atom, the sum of the L1 norms of all wavelet coefficients is calculated and normalized. This constraint helps the dictionary atom maintain sparsity in the multi-scale transform domain, thus better capturing the structural features of the temperature field; on the other hand, the sparsity constraint term... In order to ensure coefficient The sparsity of the data means that each temperature field sample is represented by a linear combination of only a few dictionary atoms. This mechanism can activate key dictionary atoms with a small number of non-zero coefficients, which not only improves the accuracy and robustness of the temperature field representation and effectively reduces reconstruction error, but also significantly reduces the number of model optimization parameters, thereby improving training efficiency and suppressing overfitting.

[0078] Step 3: Solving for the ray-flight time in the sub-temperature region based on the physical equations

[0079] Based on the sparse dictionary constructed in step two This invention describes the temperature field reconstruction problem as follows: for measured flight time data... Assuming the existence of a temperature field Sparse vectors Satisfying the linear representation relation In this case, the speed of sound corresponding to the coordinates in space can be further calculated, thereby deriving the theoretical flight time for each sound path. Ideally, it should match the measured value, that is... .

[0080] Since temperature affects the speed of sound, the relationship between temperature and sound speed in the coolant of the heat pipe section of a nuclear reactor is fitted using the IF97 formula for steam and liquid water, yielding the following functional relationship:

[0081]

[0082] in, T represents the speed of sound; T represents the temperature. Once the speed of sound of the coolant is calculated, the temperature can be determined by inversion.

[0083] To solve the above problem, since the location of the acoustic transceiver is already determined, and based on the m×n=N sub-temperature zones divided in step one, the length of each acoustic path in each sub-temperature zone is known; therefore, the flight time of the i-th acoustic ray can be determined. :

[0084]

[0085] in, It is the length of the i-th sound wave path in the j-th sub-temperature region. The acoustic slowness function of the j-th sub-temperature region is the reciprocal of the speed of sound;

[0086] Step 4: Expand the measurement cycle and establish a dataset of "A and B surface temperatures - pipe temperature";

[0087] Step 5: Build a neural network to establish the correspondence between "A and B surface temperatures and pipe temperatures" to achieve an increase in the dimensionality of temperature reconstruction.

[0088] Furthermore, step three specifically involves:

[0089] Under the constraints of limited sensor deployment in practice, the amount of available flight time data is relatively small, leading to an ill-posed solution to the inverse problem and limiting reconstruction accuracy. To enhance the physical rationality and numerical stability of the reconstruction process, such as... Figure 4 The flowchart for calculating the two-dimensional temperature field distribution is shown. Introducing the physical equation of sound propagation as a soft constraint, the following loss function, which integrates data-driven and physical information, is constructed:

[0090]

[0091] The definitions and functions of each loss function are as follows:

[0092] Data fitting term This directly constrains the consistency between the predicted flight time and the measured data, ensuring the reconstruction accuracy within the sensor deployment area;

[0093] physical constraints Introducing the heat conduction equation as a physical prior to enhance the rationality of the temperature field in the spatiotemporal evolution process; It represents the rate of change of temperature over time, approximated by the temperature difference between adjacent moments, reflecting the temporal continuity of temperature evolution; Indicates the thermal diffusivity. It is the Laplace operator for temperature, representing the two-dimensional rate of change of temperature in space. It is a heat diffusion term, reflecting the second-order change of temperature in space, that is, the driving effect of the surrounding temperature distribution on the local temperature change; through this term constraint, the temperature field is forced to follow the heat diffusion law, avoiding physical inconsistencies in fluctuations.

[0094] vocal constraint terms When sensors are sparsely distributed, such as when the sound ray coverage is less than 10%, the non-sound ray regions lack direct observation constraints, and relying solely on physical equations may still lead to non-physical jumps or abnormal fluctuations in these regions. Therefore, a sound ray constraint term is introduced to constrain the temperature in the non-sound ray regions. Average temperature around the adjacent sound ray This constraint allows for fluctuations, preventing deviations from the actual physical context. It also helps smooth the temperature distribution at the boundary between ray and non-ray regions, enhancing overall spatial continuity.

[0095] Furthermore, step four specifically involves:

[0096] like Figure 1 As shown in the schematic diagram of the main pipeline ultrasonic array, the transducer transmit and receive strategy is to complete the transmit and receive of the transducer group within 1 second, and to reconstruct the temperature field of surfaces A and B within that second based on the measured transit time.

[0097] Since we need to reconstruct the temperature field of the three-dimensional pipeline based on the two-dimensional cross section, and the temperature field is continuous in both the time domain and the spatial domain, the temperature field of the two-dimensional cross section at different times can actually be used as prior information of the temperature field of the three-dimensional pipeline.

[0098] Extend the measurement period from 1 second to s;

[0099] Based on the Ansys numerical simulation platform, ultrasonic measurement experiments were conducted under Nm different temperature inlet conditions to simulate time-varying temperature fields, obtaining the Nm group of flight time matrices:

[0100]

[0101] in, This represents the transit time of the i-th transducer at time t; the transit time matrix consists of T row vectors. Substituting the t-th row vector into step two yields the temperature distribution of surfaces A and B at time t. ;

[0102] Create a dataset with the following input features:

[0103]

[0104] The 3D pipe temperature field distribution from the CFD simulation is output as follows:

[0105] .

[0106] Furthermore, step five specifically involves:

[0107] like Figure 5 The flowchart for simulating a multipath convolutional neural network shows how a neural network model is built after the dataset is created.

[0108] A multimodal temporal feature hierarchical fusion neural network is proposed to perform cross-modal and cross-time scale feature extraction from time-of-flight (TOF) signals and two-dimensional temperature distribution signals (Temp2D), solving the problem that traditional single-modal models cannot simultaneously achieve both "spatial distribution accuracy" and "temporal evolution efficiency".

[0109] 1. Multimodal input hierarchical preprocessing, dual-modal parallel feature channels;

[0110] Input layer: Simultaneously receives two types of heterogeneous data

[0111] Mode 1: TOF signal: a time sequence reflecting the propagation time of the medium in the pipeline;

[0112] o-mode 2: Temp2D signal: a two-dimensional temperature distribution matrix of the pipe cross-section.

[0113] Parallel branch construction: Independent feature extraction subnetworks are designed for the two modalities to avoid interference from modal information.

[0114] oTOF branch: Extracts local temporal features of the TOF signal through convolution 1 + pooling 1;

[0115] oTemp2D branch: Extracts the spatial distribution features of Temp2D through convolution 2 + pooling 2;

[0116] o Cross-modal native channels: Directly map the original TOF and Temp2D to T feature channels according to the time scale, preserving the temporal correlation of the original modality and avoiding feature compression loss.

[0117] 2. Cross-scale feature fusion

[0118] The feature fusion module enables the synergistic enhancement of three types of features:

[0119] • It integrates the temporal features of the TOF branch, the spatial features of the Temp2D branch, and the temporal correlation features of the cross-modal native channels;

[0120] A channel attention weighting mechanism is adopted: dynamic weights are assigned to the features of T channels to highlight key modal information at different time scales.

[0121] Since the temperature field changes at every moment, the flight time of the same path is different, thus it carries the characteristics of the temperature field changing over time; due to the existence of flow velocity, the temperature of surfaces A and B at the previous moment will affect the temperature characteristics of their adjacent surfaces, thus it carries the characteristics of the temperature field changing in space; the flight time and the temperature of surfaces A and B satisfy the above equation.

[0122] Considering the correlation within the input features, in order to better capture the features, a multi-path convolutional neural network is proposed to simulate it. The Num groups of data are divided into a training set of 8: test set of 2 and fed into the neural network for learning, so as to obtain the three-dimensional pipe temperature field distribution results at each measurement time.

[0123] This invention discloses a method for increasing the dimensionality of temperature reconstruction in the main pipeline of a nuclear power plant based on an ultrasonic array:

[0124] Given the limited scale and number of ultrasonic transducers in the main pipeline of a nuclear power plant, this method first uses measured sparse low-dimensional data to finely reconstruct the temperature distribution of the low-dimensional cross-section, and then uses a neural network to achieve high-quality reconstruction of the temperature field of the high-dimensional pipeline, thus solving the current problem of not being able to achieve high-quality three-dimensional reconstruction under sparse deployment.

[0125] By combining sparse dictionary training with the solution of nonlinear measurement equations incorporating physical equations, and utilizing the Ansys simulation platform to provide a large number of temperature fields with different time-varying characteristics and spatial properties for the training of the sparse dictionary, the dictionary exhibits excellent robustness and generalization. At the same time, the introduction of physical equations and the design of multi-objective loss terms improve the convergence speed and accuracy of the solution of nonlinear measurement equations, solving the problem that only the average temperature of two-dimensional cross sections can be estimated under the existing sparse, parallel sound ray layout, and realizing the fine reconstruction of two-dimensional temperature fields under sparse data.

[0126] Of course, those skilled in the art should be able to make various corresponding changes and modifications based on the present invention without departing from its spirit and essence, but all such changes and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for dimensionality enhancement in temperature reconstruction of main pipelines in nuclear power plants based on ultrasonic arrays, characterized in that: Includes the following steps: Step 1: Temperature measurement and sub-temperature zone division of surface A and surface B based on ultrasonic array; Temperature measurement and reconstruction are carried out by arranging the ultrasonic array with parallel sound rays, and temperature measurement is carried out by evenly distributing the transceiver transducers in two groups on two mutually perpendicular elliptical surfaces A and B. By arranging them in parallel, the direct sound rays are made to be on the same plane, and the direct sound rays on plane A and plane B are parallel to each other; the sub-temperature region of each plane is divided into m×n sub-temperature regions; Step Two: Sparse Dictionary Training: The temperature field T is divided into M pixels, where D is an M×N dimensional matrix, where N is the number of atoms in the dictionary. The sparse transformation process from D to T is as follows: A sparse vector x is a sparse expression in the dictionary field D. The number of non-zero elements in x, k, represents its sparsity. Ansys numerical simulations were used to design several unimodal, bimodal, and multimodal time-varying temperature fields as training sets to establish a sparse dictionary D. This ensured that the optimal sparse vector could be calculated based on D for each temperature field. To satisfy ; The process of constructing the sparse dictionary D is as follows: Set up a dictionary The two-dimensional mesh form of the i-th atom is as follows ; The two-dimensional wavelet decomposition process is defined as follows: Step 1: Define a one-dimensional low-pass filter based on the selected wavelet basis type. K is the filter length; high-pass filter It is derived from h by the orthogonality of the wavelet basis; low-pass filter operator It applies a low-pass filter h to the signal and performs downsampling with a step size of 2; the high-pass filter operator Apply a high-pass filter g to the signal and perform downsampling with a step size of 2; Number of decomposition layers L; Step 2: [The sentence is incomplete and requires more context.] Apply each line separately and The row low-pass matrix is ​​obtained. High-pass matrix Two intermediate matrices; Step 3: Apply the following to each column of the intermediate matrix from Step 2. and The four types of coefficients for word decomposition are obtained—low-frequency approximation coefficients. ,right Column filtering is used to obtain the overall temperature distribution trend of atoms; horizontal detail coefficients. ,right Column filtering is used to characterize the temperature gradient change in the vertical direction; vertical detail coefficients. ,right The high-pass filter is used to characterize the temperature gradient change in the horizontal direction; the diagonal detail coefficients are also used. ,right The temperature abrupt changes along the tilt direction are characterized by high-pass filtering. Step 4: Repeat Step 3 until the Lth level of decomposition is completed; after the i-th atom undergoes L levels of wavelet decomposition, a set of multi-scale coefficients is obtained, which can be expressed as... , It is a wavelet basis type; For each temperature field, the following wavelet bases are used: The mean of the L1 / L2 norm ratio of the wavelet coefficients Mean values ​​of wavelet coefficients and Gini coefficients The average effective sparsity required to retain 99% of the energy The optimal wavelet basis is selected as an indicator; the calculation method is as follows: ; in, It is a one-dimensional vector obtained by flattening all coefficients after the i-th atom undergoes L-level wavelet decomposition. Describing the L1 norm, Represents the L2 norm. It is the wavelet coefficient vector of the i-th atom. The ascending order results It meets the conditions The smallest integer; It is the final scoring metric for determining the optimal wavelet basis. This is the penalty coefficient; During the learning process, three loss functions are used as follows: ; Among them, the reconstruction loss term This is used to measure the difference between the temperature field constructed from the sparse dictionary D and the sparse vector x and the true temperature field; to enhance the sparse representation ability of each dictionary atom in its optimal transformation domain, a dictionary constraint term is introduced. ; Step 3: Solving for the ray-flight time in the sub-temperature region based on the physical equations The relationship between temperature and sound velocity of liquid coolant was fitted using the IF97 formula based on steam and liquid water, yielding the following functional relationship: ; Where c is the speed of sound; T is the temperature. Based on the m×n=N sub-temperature zones divided in step one, the flight time of the i-th sound ray can be determined. : ; in, It is the length of the i-th sound wave path in the j-th sub-temperature region. The acoustic slowness function of the j-th sub-temperature region; Step 4: Expand the measurement cycle and establish a dataset of "A and B surface temperatures - pipe temperature"; Step 5: Build a neural network to establish the correspondence between "A and B surface temperatures and pipe temperatures" to achieve an increase in the dimensionality of temperature reconstruction.

2. The method for dimensionality enhancement of temperature reconstruction in nuclear power plant main pipelines based on ultrasonic arrays according to claim 1, characterized in that: The specific details of step three are as follows: By introducing the physical equations of sound propagation as soft constraints, a loss function that integrates data-driven and physical information is constructed as follows: ; Definitions of each loss function: Data fitting term : Directly constrain the consistency between predicted flight time and measured data; physical constraints Introducing the heat conduction equation as a physical prior; It represents the rate of change of temperature over time, approximated by the temperature difference between adjacent moments; Indicates the thermal diffusivity. It is the Laplace operator for temperature, representing the two-dimensional rate of change of temperature in space. It is a heat diffusion term; vocal constraint terms Introducing acoustic constraint terms to force the temperature of non-acoustic regions. Average temperature around the adjacent sound ray Fluctuations occur.

3. The method for dimensionality enhancement of temperature reconstruction in the main pipeline of a nuclear power plant based on an ultrasonic array, as described in claim 1 or 2, is characterized in that: The specific details of step four are as follows: The transducer transmit / receive strategy is to complete the transmit / receive of the transducer group within 1 second, and to reconstruct the temperature field of surfaces A and B within that second based on the measured transit time. Extend the measurement period from 1 second to s; Based on the Ansys numerical simulation platform, ultrasonic measurement experiments were conducted under Nm different temperature inlet conditions to simulate time-varying temperature fields, obtaining the Nm group of flight time matrices: ; in, This represents the transit time of the i-th transducer at time t; the transit time matrix consists of T row vectors. Substituting the t-th row vector into step two yields the temperature distribution of surfaces A and B at time t. ; Create a dataset with the following input features: ; The 3D pipe temperature field distribution from the CFD simulation is output as follows: 。 4. The method for increasing the dimension of temperature reconstruction in the main pipeline of a nuclear power plant based on an ultrasonic array, as described in claim 3, is characterized in that: Step five specifically refers to: After creating the dataset, build the neural network model; The above equation applies to the relationship between the flight time and the temperatures of surfaces A and B. ; A multi-path convolutional neural network is used to simulate the data. The Num sets of data are divided into training and testing sets according to a certain ratio and fed into the neural network for learning. This allows us to obtain the three-dimensional pipe temperature field distribution results at each measurement time.

Images