Method and device for reconstructing physical field of turbine rotor of gas turbine

Abstract

The invention provides a physical field reconstruction method and device for a turbine rotor of a gas turbine, and belongs to the field of gas engine control. The method comprises the steps that measuring point temperature data of a preset node of the turbine rotor are acquired, and a physical field of the turbine rotor is reconstructed based on the temperature data of the preset node and the node position; predicting temperature distribution data of all nodes of the turbine rotor through the first reconstruction model; predicting stress, strain and displacement distribution data of the turbine rotor through a second reconstruction model based on the temperature distribution data of all nodes of the turbine rotor; and respectively carrying out linear fitting on the temperature distribution data, the stress distribution data, the strain distribution data and the displacement distribution data to obtain a temperature field, a stress field, a strain field and a displacement field of turbine rotor reconstruction. In this way, the global physical field is predicted in real time through the local temperature data, complex numerical calculation is avoided, and real-time prediction of the physical field of the turbine rotor is guaranteed.

Description

Technical Field

[0001] The present invention belongs to the field of gas turbine control, and particularly relates to a method and device for reconstructing the physical field of a gas turbine turbine rotor. Background Art

[0002] As an efficient power device, gas turbines are of great significance for ensuring national energy security and achieving the "dual carbon" goal. The turbine rotor of a gas turbine is an important high-temperature rotating component of the generator set. During operation, it not only bears its own gravity and the centrifugal force brought by rotation, but also bears the tensile force brought by the blade centrifugal force at the wheel groove part, as well as the thermal stress brought by the internal and external temperature difference under a large volume. Compared with the mode of long-term stable full-load operation, gas turbine generator sets participating in peak shaving must be able to cope with frequent start-stop and large load changes during flexible operation, which means that the rotor has to bear more severe alternating mechanical loads and thermal loads. During peak shaving, gas turbines have higher efficiency within a shorter start-up time, can reduce pollutant emissions, and have higher economic and environmental benefits; however, a shorter start-up time often means more severe load changes, which will bring a greater temperature difference between the inside and outside of the rotor, causing the thermal stress level to rise sharply and seriously damaging the safety and stability of the unit operation.

[0003] At present, it is extremely difficult to monitor the real-time state of a gas turbine turbine rotor, which poses a huge challenge to its safe operation and life assessment. When the gas turbine is in operation, the turbine rotor is in a complex working condition of high temperature and high-speed rotation, which limits the direct measurement technical means of temperature and stress, making it difficult to obtain the temperature distribution and stress state inside the rotor in real time and accurately. In the prior art, if local physical fields are obtained through sensors, complex numerical calculation and analysis are required to obtain the complete physical field information of the turbine rotor, which has time lag and cannot meet the requirements of real-time monitoring of the turbine rotor state during peak shaving.

[0004] Therefore, in order to improve the start-up speed of the unit's flexible operation on the premise of ensuring the safe and stable operation of the rotor, it is urgent to establish a method for reconstructing the physical field of a gas turbine turbine rotor with real-time efficacy. Summary of the Invention

[0005] In order to solve the problem that it is difficult to obtain the internal physical field of a gas turbine turbine rotor in real time, the present invention provides a method and device for reconstructing the physical field of a gas turbine turbine rotor.

[0006] To achieve the above object, the present invention provides the following technical solutions: A method for reconstructing the physical field of a gas turbine turbine rotor, the method comprising: Divide the turbine rotor into multiple nodes based on a grid of preset dimensions, and obtain the temperature data and node positions of the preset nodes of the turbine rotor; Based on the temperature data and node positions of the preset nodes, predict the temperature distribution data of all nodes of the turbine rotor through a first reconstruction model; furthermore, based on the temperature distribution data of all nodes of the turbine rotor, predict the stress, strain, and displacement distribution data of the turbine rotor through a second reconstruction model; Perform linear fitting on the temperature, stress, strain, and displacement distribution data respectively to obtain the reconstructed temperature field, stress field, strain field, and displacement field of the turbine rotor.

[0007] Optionally, the first reconstruction model and the second reconstruction model include: Construct a first reconstruction model T-GCN by cascading a convolutional neural network CNN and a graph convolutional neural network GCN, and train T-GCN; Adopt the same network structure as GCN in T-GCN as the second reconstruction model S-GCN for stress-strain-displacement prediction, and train S-GCN.

[0008] Optionally, construct a data sample set for starting a gas turbine to train T-GCN and S-GCN. The construction of the data sample set for starting a gas turbine includes: Arrange temperature sensors in the area with a significant surface temperature gradient of the turbine rotor to obtain real-time temperature data; Based on the real-time temperature data, perform parametric modeling on the average exhaust temperature curve during the load increase stage of the gas turbine startup process to generate a sample set including two startup types: cold startup and warm startup; Using the transient thermo-mechanical coupling finite element method, combined with the sample set generated by the parametric modeling, calculate the temperature, stress, strain, and displacement distribution data of the turbine rotor, and correct the numerical model through the measured data of the corresponding physical parameter sensors; Based on the corrected numerical model, extract the three-dimensional coordinates, temperature, stress, strain, and displacement distribution data of the preset nodes of the turbine rotor and the node temperature data corresponding to the positions of the temperature sensors to construct a training data set.

[0009] Optionally, the parametric modeling uses the Latin hypercube sampling method to sample the average exhaust temperature curves during the load increase stage of cold startup and warm startup to generate a sample set of startup curves defined by 5 key control points, where the initial temperature and load increase time are dynamically adjusted according to the startup type.

[0010] Optionally, the training of T-GCN and S-GCN includes: Divide the data sample set in a ratio of 7:3, and use 70% of the sample quantity of the data sample set as the training set; Train the T-GCN and S-GCN respectively using the Adam and Adamax optimization algorithms; Set initial learning rates for the CNN and GCN respectively, and dynamically adjust the initial learning rates based on the mean squared error (MSE) as the loss function; the expression of the loss function is: ; ; where is the turbine rotor temperature distribution data output by the T-GCN, , and are the stress, strain, and displacement distribution data of the turbine rotor output by the S-GCN respectively, , , and are the temperature, stress, strain, and displacement distribution data of the real turbine rotor respectively; N is the number of samples in the data sample set, and c is the c-th sample data in the data sample set; If the value of the loss function exceeds the preset threshold, optimize the network parameters of the cascaded graph combination convolution network through the Adam optimization algorithm and the Adamax optimization algorithm until the value of the loss function is less than or equal to the preset threshold.

[0011] Optionally, obtaining the temperature, stress, strain, and displacement distribution data of the turbine rotor through the reconstruction model prediction includes: Taking the preset node temperature and position of the turbine rotor as inputs, dimensionality-expand the real-time temperature and position of the preset nodes through the fully connected layer of the CNN in the first reconstruction model T-GCN, reconstruct the spatial features through the upsampling layer, extract the correlation features between nodes of the spatial features through the multi-layer graph convolution kernel of the GCN and the Gaussian deviation linear unit (GELU) activation function, and predict and output the temperature distribution data through the correlation features; taking the temperature distribution data as the input of the second reconstruction model S-GCN to obtain the output stress, strain, and displacement distribution data.

[0012] A physical field reconstruction device for a gas turbine turbine rotor, the device includes: An acquisition module, configured to divide the turbine rotor into multiple nodes based on a grid of a preset size, and acquire the temperature data and node positions of the preset nodes of the turbine rotor; A prediction module, configured to predict the temperature distribution data of all nodes of the turbine rotor through the first reconstruction model based on the temperature data and node positions of the preset nodes; and further predict the stress, strain, and displacement distribution data of the turbine rotor through the second reconstruction model based on the temperature distribution data of all nodes of the turbine rotor; A reconstruction module is used to perform linear fitting on the temperature, stress, strain, and displacement distribution data respectively, to obtain the reconstructed temperature field, stress field, strain field, and displacement field of the turbine rotor.

[0013] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the above-mentioned method for reconstructing the physical fields of a gas turbine turbine rotor.

[0014] A computer device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the above-mentioned method for reconstructing the physical fields of a gas turbine turbine rotor.

[0015] The method for reconstructing the physical fields of a gas turbine turbine rotor provided by the present invention has the following beneficial effects: This method significantly improves the real-time performance and accuracy of the state monitoring of the gas turbine turbine rotor through multi-node data acquisition and dual model reconstruction technology. First, by dividing the rotor into multiple nodes and obtaining the temperature data and position information of the preset nodes, a monitoring foundation from local to global is established; secondly, the first reconstruction model predicts the temperature distribution of all nodes based on limited node data, and the second reconstruction model maps the temperature field to the stress, strain, and displacement fields, realizing the synchronous inversion of thermo-mechanical coupling parameters; finally, continuous field distribution is generated through linear fitting, greatly improving the visualization degree of the rotor state. This hierarchical and progressive processing method compensates for the physical monitoring blind area through a mathematical model, avoids the lag caused by the complicated process of numerical calculation, can predict the global situation in real time based on local conditions, and provides high-precision data support for real-time diagnosis of potential risks such as rotor thermal deformation and creep fatigue. Description of the Drawings

[0016] In order to more clearly illustrate the embodiments of the present invention and their design schemes, the accompanying drawings required for this embodiment will be briefly introduced below. The accompanying drawings in the following description are only partial embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 It is a schematic flow chart of a method for reconstructing the physical fields of a gas turbine turbine rotor provided by the present invention according to an exemplary embodiment.

[0018] Figure 2 It is a schematic flow chart of another method for reconstructing the physical fields of a gas turbine turbine rotor provided by the present invention according to an exemplary embodiment.

[0019] Figure 3 It is a schematic diagram of parameterization of a gas turbine startup process curve provided by the present invention according to an exemplary embodiment.

[0020] Figure 4 This is a schematic diagram of the positions of limited measurement points provided by the present invention according to an exemplary embodiment.

[0021] Figure 5 This is a flowchart for constructing a training data set of a gas turbine turbine rotor provided by the present invention according to an exemplary embodiment.

[0022] Figure 6 This is a schematic diagram of the data processing flow of a series of graph combined convolutional networks provided by the present invention according to an exemplary embodiment.

[0023] Figure 7 This is a block diagram of a physical field reconstruction device for a gas turbine turbine rotor provided by the present invention according to an exemplary embodiment. Detailed implementation manners

[0024] In order to enable those skilled in the art to better understand the technical solutions of the present invention and be able to implement them, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention, and cannot be used to limit the protection scope of the present invention.

[0025] A method for reconstructing the physical field of a gas turbine turbine rotor based on limited measurement points provided by the present invention, as Figure 1 shown, first sequentially perform the following steps: selection of monitoring point positions and sensor arrangement, parameterization of the start-up process curve of the gas turbine, sampling method is used to construct a start-up curve sample set, finite element calculation and correction of the temperature field of the turbine rotor, finite element calculation and analysis of the stress, strain, and displacement fields of the turbine rotor, data structure processing and data set construction, division of the training set and the test set. Secondly, construct and train the T-GCN temperature field reconstruction network and the S-GCN stress, strain, and displacement field reconstruction network respectively. Then, based on the measured point data, reconstruct the temperature field of the turbine rotor, and based on the reconstructed temperature field, reconstruct the stress, strain, and displacement fields. Finally, perform algorithm maintenance regularly.

[0026] The following will detail the technical solutions provided by each embodiment of the present invention with reference to the accompanying drawings.

[0027] First, the present invention provides a method for reconstructing the physical field of a gas turbine turbine rotor, specifically as Figure 2 shown, including the following steps: S101. Divide the turbine rotor into multiple nodes based on a grid of a preset size, and obtain the temperature data and node positions of the preset nodes of the turbine rotor.

[0028] In this step, first, the turbine rotor is divided into multiple nodes based on a grid of a preset size, which can be adjusted according to the actual size or shape of the turbine rotor. Then, temperature sensors can be arranged in the regions with significant surface temperature gradients of the turbine rotor to obtain real-time temperature data.

[0029] For example, first, select the monitoring point positions of the turbine rotor and arrange temperature sensors at the monitoring points. According to the actual structural shape of the turbine rotor, select the positions on the rotor surface where the temperature changes violently and where it is easy to arrange temperature sensors as temperature monitoring points, and arrange temperature sensors at the monitoring point positions to monitor the temperature information in real time during the actual operation process. In this embodiment, the turbine rotor is a four-stage disk model. Since the temperature difference between the inner and outer walls of the rotor is relatively large during the load increase stage, resulting in a relatively high thermal stress level, therefore, for the four-stage disk model in this embodiment, a total of 8 positions on the inner and outer walls of each stage of the disk are used as temperature monitoring points.

[0030] Secondly, parameterize the start-up process curve. The start-up process of a gas turbine usually includes typical processes such as barring gear turning, purging, ignition, warm-up, speed increase, and load increase. For the load increase stage, the rotational speed of the turbine rotor reaches the rated value, and at the same time, the temperature of the gas in the turbine flow passage continuously rises, which makes the radial temperature gradient of the rotor continuously increase and the thermal stress level increase. This is a typical high-stress level stage in the gas turbine start-up process and is the first stage that needs to be considered for shortening during the rapid start-up process under the flexible operation conditions of the unit. Parameterize it according to the average exhaust gas temperature curve during the load increase stage in the gas turbine start-up process. As Figure 3 shown, the average exhaust gas temperature curve during the load increase stage in the start-up process is controlled by 5 key points, where t is time, T is the average exhaust gas temperature, the subscript 0 represents the initial moment, and the subscript f represents the end moment. For cold start and warm start processes, the temperature at the initial moment T 0 and the time at the end moment t f have certain differences and should be selected according to the actual situation; t i are different moments during the load increase stage, i = 1 - 3, T j are the average exhaust gas temperatures at different moments during the load increase stage, j = 1 - 3.

[0031] S102. Based on the temperature data and node positions of the preset nodes, predict the temperature distribution data of all nodes of the turbine rotor through the first reconstruction model; and then, based on the temperature distribution data of all nodes of the turbine rotor, predict the stress, strain, and displacement distribution data of the turbine rotor through the second reconstruction model.

[0032] In this step, the temperature, stress, strain, and displacement distribution data of the turbine rotor are predicted through the reconstruction model. The reconstruction model includes a first reconstruction model T-GCN and a second reconstruction model S-GCN. Specifically, a convolutional neural network CNN and a graph convolutional neural network GCN are cascaded to construct a temperature reconstruction network T-GCN, and T-GCN is trained; the same network structure as the GCN in T-GCN is used as the stress-strain-displacement reconstruction network S-GCN, and S-GCN is trained; the trained T-GCN and S-GCN are cascaded to construct a cascaded graph combined convolutional network as the reconstruction model.

[0033] In one embodiment, the first reconstruction model T-GCN for turbine rotor temperature prediction includes a convolutional neural network CNN and a graph convolutional neural network GCN. The convolutional neural network CNN consists of an input layer, an upsampling layer, and an output layer. The input layer is composed of multiple fully connected layers, which play a role in dimensionality increase of the input design variables. After reshaping their shapes, they are input into the upsampling layer; each upsampling layer consists of an interpolation layer, a batch normalization layer, and a convolutional kernel, and the rectified linear unit is used as the activation function; the output layer is a convolutional layer, which is used to output the result of the convolutional neural network CNN. The graph convolutional neural network GCN consists of an input layer, an intermediate layer, and an output layer. The input layer consists of a graph convolutional kernel and an activation function, where the activation function is a Gaussian bias linear unit. The input layer is used to receive the result of the convolutional neural network CNN for subsequent feature extraction; the intermediate layer has the same structure as the input layer, which is used to extract features from the data, and its parameters have an important impact on the prediction effect of the cascaded graph combined convolutional network and should be reasonably selected according to the actual situation; the output layer is used to convert the reconstruction result into the corresponding output format for output. Through the data processing and feature extraction of the convolutional neural network CNN and the graph convolutional neural network GCN, the turbine rotor temperature reconstruction network T-GCN can realize the prediction from the temperature at limited measurement points and node coordinates to the temperature distribution data of the turbine rotor.

[0034] In this way, taking the preset node temperature and position of the turbine rotor as the input, the real-time temperature and position of the preset node are dimensionally increased through the fully connected layer of CNN in the first reconstruction model T-GCN, the spatial features are reconstructed through the upsampling layer, the correlation features between the nodes of the spatial features are extracted through the multi-layer graph convolutional kernel of GCN and the Gaussian bias linear unit GELU activation function, and the temperature distribution data is predicted and output through the correlation features; taking this temperature distribution data as the input of the second reconstruction model S-GCN, the output stress, strain, and displacement distribution data are obtained.

[0035] To train the T-GCN, based on the real-time temperature data, parametric modeling can be performed on the average exhaust temperature curve in the load increase stage of the gas turbine startup process to generate a sample set including cold startup and warm startup; using the transient thermo-solid coupling finite element method, combined with the sample set generated by this parametric modeling, calculate the temperature, stress, strain, and displacement distribution data of the turbine rotor, and correct the numerical model through the measured data of the corresponding physical parameter sensors; based on the corrected numerical model, extract the three-dimensional coordinates, temperature, stress, strain, displacement distribution data of the turbine rotor nodes, and the node temperature data corresponding to the position of the temperature sensor to construct a training data set.

[0036] Among them, the parametric modeling uses the Latin hypercube sampling method to sample the average exhaust temperature curves in the load increase stages of cold startup and warm startup, generating a sample set of startup curves defined by 5 key control points, where the initial temperature and load increase time are dynamically adjusted according to the startup type.

[0037] Then, an appropriate sampling method is used to construct the sample set of startup curves. In this embodiment, the Latin hypercube sampling method is selected to sample the startup curves in the cold startup and warm startup states, and the sample size for each startup state is N. The value ranges of the sampling parameters are as follows: .

[0038] After that, finite element calculation and correction of the turbine rotor temperature distribution are carried out. Using the transient thermo-solid coupling calculation method, calculate the temperature distribution data of the turbine rotor at different times in the cold startup and warm startup states. The temperature load of the turbine rotor will have an important impact on its stress distribution. In particular, in the subsequent steps, the actual temperature sensor data will be used to predict the temperature distribution data of the turbine rotor. Therefore, the numerical calculation results are compared with the real monitoring results of the temperature sensor under actual operation to correct the numerical calculation method so that the numerical calculation results are close to the real monitoring results.

[0039] Then, finite element calculation and analysis of the turbine rotor strength are carried out. Based on the calculated temperature distribution data, it is applied to the turbine rotor as a temperature load, and the stress, strain, and displacement distribution data of the turbine rotor at different times in the cold startup and warm startup states are obtained by finite element calculation.

[0040] Finally, data structure processing and data set construction are carried out. As Figure 4 and Figure 5As shown, the temperature, stress, strain, and displacement distribution data of the turbine rotor at different times in the cold start and warm start states are extracted for each sample. At the same time, the node Mm corresponding to the actual sensor position in the numerical model used for the finite element calculation is selected, and the temperature value data of the corresponding node Mm in the turbine rotor numerical model for each sample under different start states and different times are extracted, and processed into the input format of the series graph combined convolutional network to form the gas turbine turbine rotor training dataset D. In this embodiment, the gas turbine turbine rotor training dataset D is: , where n is the sample number, n = 1~N; are the x, y, and z coordinates of all nodes under each sample, is the node temperature data, is the node stress data, is the node strain data, is the node displacement data, and the subscript a is the node number under each sample, a = 1 - A, where A is the total number of nodes; is the temperature data of the node corresponding to the actual temperature sensor layout position in the numerical model, where m is the corresponding node number. In this embodiment, a total of 8 temperature sensors are arranged, so the temperature data of 8 corresponding nodes Mm need to be extracted, m = 1 - 8.

[0041] After the dataset is constructed, the training set and the test set need to be divided. The gas turbine turbine rotor training dataset D is divided into the training set TD and the test set VD according to a certain proportion. The training set TD is used to train the series graph combined convolutional network, and the test set VD is used to verify the training results of the series graph combined convolutional network. In this embodiment, the training set TD and the test set VD are divided in a ratio of 7:3. Therefore, the number of samples in the training set TD is 70%×N, and the number of samples in the test set VD is 30%×N, that is, the data sample set is divided in a ratio of 7:3, and 70% of the data sample set is used as the training set, and 30% of the data sample set is used as the test set.

[0042] Secondly, the Adam and Adamax optimization algorithms are used to train the T-GCN and S-GCN respectively, and the initial learning rates are set for the CNN and GCN respectively, and the initial learning rate is dynamically adjusted based on the mean square error MSE as the loss function.

[0043] For example, according to the training set TD, two reconstruction networks are trained respectively. Among them, the data used for training the turbine rotor temperature reconstruction network T-GCN is , and its output is the turbine rotor temperature distribution data ; the data used for training the turbine rotor stress, strain, and displacement reconstruction network S-GCN is , and its output is the turbine rotor stress distribution data , the turbine rotor strain distribution data , and the turbine rotor displacement distribution data . During the training process, relevant optimization algorithms, loss functions, and relevant hyperparameters of the convolutional neural network CNN and the graph convolutional neural network GCN should be reasonably selected according to the actual situation. In this embodiment, the convolutional neural network CNN is trained using the Adam optimization algorithm with an initial learning rate of 0.001, the graph convolutional neural network GCN is trained using the Adamax optimization algorithm with an initial learning rate of 0.005, the batch sample size is 64, and the adaptive learning rate descent algorithm based on the loss function is used. The loss function selects the MSE loss function, and its expression is as follows: ; ; where, is the turbine rotor temperature distribution data output by T-GCN, , and are the stress, strain, and displacement distribution data of the turbine rotor output by S-GCN respectively, , , and are the temperature, stress, strain, and displacement distribution data of the real turbine rotor respectively; N is the number of samples in the data sample set, and c is the label of the sample data in the data sample set, that is, the cth sample data in the data sample set.

[0044] If the value of the loss function exceeds the preset threshold, the network parameters of the series graph combined convolutional network are correspondingly optimized through the Adam optimization algorithm and the Adamax optimization algorithm until the value of the loss function is less than or equal to the preset threshold.

[0045] Reconstruct the temperature field of the turbine rotor based on the measured point data and node grid information. As a high-temperature rotating component, the surface temperature of the gas turbine turbine rotor is relatively high, and relatively expensive high-temperature sensors need to be used, and the cost is usually high. Moreover, to reduce the temperature of the turbine rotor, cooling structures are usually arranged inside and on the surface, and it is difficult to arrange sensors. It is almost impossible to obtain the complete temperature field of the turbine rotor based on sensors. The present invention combines the temperature data of the temperature sensors at limited measured points at different times with the node grid information of the turbine rotor numerical model, and inputs it into the trained turbine rotor temperature reconstruction network T-GCN to predict the turbine rotor temperature distribution data at different times in real time Compared with numerical calculation methods, the present invention uses actual temperature sensor data for prediction, which not only increases data reliability but also enables real-time prediction and requires less computing resources. Compared with experimental measurement methods, the present invention can obtain global temperature field information with limited measurement points, which is conducive to grasping the complete state of the turbine rotor. It solves the problem that traditional measurement methods require the deployment of multiple sensors and cannot obtain internal data of the turbine rotor, greatly reducing measurement costs.

[0046] In another embodiment, the second reconstruction model S-GCN for turbine rotor stress, strain, and displacement prediction is structurally identical to the graph convolutional neural network GCN in the first reconstruction model T-GCN for turbine rotor temperature prediction. It receives temperature distribution data from the turbine rotor temperature field reconstruction network T-GCN to realize the prediction from the turbine rotor temperature field to the turbine rotor stress, strain, and displacement distribution data.

[0047] For the training process of S-GCN, you can refer to the training process of T-GCN mentioned above. Construct a second training set based on the temperature field and the corresponding stress field, strain field, and displacement field to train S-GCN.

[0048] A cascade graph combination convolutional network is constructed based on the temperature field reconstruction network T-GCN and the stress-strain-displacement field reconstruction network S-GCN. The cascade graph combination convolutional network consists of two layers. The first layer is the turbine rotor temperature field reconstruction network T-GCN, and the second layer is the turbine rotor stress, strain, and displacement field reconstruction network S-GCN.

[0049] S103. Perform linear fitting on the temperature, stress, strain and displacement distribution data to obtain the temperature field, stress field, strain field and displacement field of the turbine rotor.

[0050] When the gas turbine is in operation, the rotor temperature fluctuates dramatically and is subjected to centrifugal loads caused by high rotational speeds. The rotor needs to withstand high stress for a long time. The conventional measurement method is to use strain gauges. Similarly, the cost of high-temperature strain gauges is high, and it is impossible to obtain complete stress and strain field data of the turbine rotor. In addition, compared with temperature sensors, strain gauges need to be tightly attached to the surface of the turbine rotor. The accuracy of strain gauge measurement is closely related to the attachment form and state. The reliability of the data cannot be guaranteed during actual operation. Figure 6 As shown, the temperature distribution data of the turbine rotor obtained by the present invention based on T-GCN prediction , input it into the reconstruction network S-GCN of the trained turbine rotor stress, strain, and displacement distribution data, and predict the stress distribution data of the turbine rotor at different times in real time , strain distribution data , displacement distribution data , finally, perform linear fitting on the distributed data respectively to obtain the corresponding reconstructed fields. The present invention uses the temperature distribution data of the turbine rotor as the input to predict the stress distribution data of the turbine rotor , strain distribution data and displacement distribution data . Therefore, the actual operation data of the temperature sensors is also considered. The data results are authentic and do not rely on traditional strain gauge measurement methods, greatly improving the continuity and reliability of data acquisition. There are also significant advantages in the acquisition speed of the stress field, strain field, and displacement field compared with existing numerical calculation methods.

[0051] Finally, algorithm maintenance can also be carried out regularly according to the actual situation or according to the actual situation.

[0052] By adopting the above method, through multi-node data acquisition and dual model reconstruction technology, the real-time performance and accuracy of the gas turbine turbine rotor condition monitoring are significantly improved. First, by dividing the rotor into multiple nodes and obtaining the temperature data and position information of the preset nodes, a monitoring foundation from local to global is established; secondly, the first reconstruction model predicts the temperature distribution of all nodes based on the limited node data, and the second reconstruction model maps the temperature field to the stress, strain, and displacement fields, realizing the synchronous inversion of thermo-mechanical coupling parameters; finally, continuous field distributions are generated through linear fitting, greatly improving the visualization degree of the rotor state. This hierarchical and progressive processing method compensates for the physical monitoring blind area through a mathematical model, avoiding the lag caused by the complicated process of numerical calculation, and can predict the global situation in real time based on the local situation, providing high-precision data support for real-time diagnosis of potential risks such as rotor thermal deformation and creep fatigue.

[0053] Secondly, the present invention also provides a physical field reconstruction device for a gas turbine turbine rotor, as Figure 7 shown, including: An acquisition module 201, configured to divide the turbine rotor into multiple nodes based on a grid of a preset size, and acquire the temperature data and node positions of the preset nodes of the turbine rotor.

[0054] A prediction module 202, configured to predict the temperature distribution data of all nodes of the turbine rotor through the first reconstruction model based on the temperature data and node positions of the preset nodes; and further predict the stress, strain, and displacement distribution data of the turbine rotor through the second reconstruction model based on the temperature distribution data of all nodes of the turbine rotor.

[0055] A reconstruction module 203, configured to perform linear fitting on the temperature, stress, strain, and displacement distribution data respectively to obtain the reconstructed temperature field, stress field, strain field, and displacement field of the turbine rotor.

[0056] Using the above device, through multi-node data acquisition and dual model reconstruction technology, the real-time performance and accuracy of gas turbine turbine rotor condition monitoring have been significantly improved. First, by dividing the rotor into multiple nodes and obtaining the temperature data and position information of preset nodes, a monitoring foundation from local to global has been established; second, the first reconstruction model predicts the temperature distribution of all nodes based on limited node data, and through the second reconstruction model, the temperature field is mapped into stress, strain, and displacement fields, realizing the synchronous inversion of thermo-mechanical coupling parameters; finally, a continuous field distribution is generated through linear fitting, greatly improving the visualization degree of the rotor state. This hierarchical and progressive processing method compensates for the physical monitoring blind area through a mathematical model, avoids the lag caused by the complicated process of numerical calculation, can predict the global situation in real time based on local conditions, and provides high-precision data support for real-time diagnosis of potential risks such as rotor thermal deformation and creep fatigue.

[0057] The present invention also provides a computer-readable storage medium, which stores a computer program that can be used to execute the steps of the gas turbine turbine rotor physical field reconstruction method provided above. Figure 2 The steps of the gas turbine turbine rotor physical field reconstruction method provided above.

[0058] The present invention also provides a computer device. At the hardware level, the computer device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory. Of course, it may also include other hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the steps of the gas turbine turbine rotor physical field reconstruction method provided above. Figure 2 The steps of the gas turbine turbine rotor physical field reconstruction method provided above.

[0059] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0060] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams can be implemented by computer program instructions, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate for implementation in the process Figure 1one or more processes and / or blocks Figure 1 means for the functions specified in one or more blocks

[0061] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including an instruction means that implements the processes Figure 1 one or more processes and / or blocks Figure 1 the functions specified in one or more blocks

[0062] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the processes Figure 1 one or more processes and / or blocks Figure 1 the steps of the functions specified in one or more blocks

[0063] It should be noted that the above-described specific embodiments can enable those skilled in the art to more comprehensively understand the present invention, but do not limit the present invention in any way. Therefore, although this specification has described the present invention in detail, those skilled in the art should understand that the present invention can still be modified or equivalently replaced; and all technical solutions and improvements that do not depart from the spirit and scope of the present invention are covered by the protection scope of the patent of the present invention. Any reference signs in the claims should not be construed as limiting the claimed claims.

Claims

1. A method for reconstructing the physical field of a gas turbine turbine rotor, characterized in that The method includes: Dividing the turbine rotor into multiple nodes based on a grid of a preset size, and obtaining the temperature data and node positions of the preset nodes of the turbine rotor; Based on the temperature data and node positions of the preset nodes, predicting the temperature distribution data of all nodes of the turbine rotor through a first reconstruction model; and then predicting the stress, strain, and displacement distribution data of the turbine rotor through a second reconstruction model based on the temperature distribution data of all nodes of the turbine rotor; Performing linear fitting on the temperature, stress, strain, and displacement distribution data respectively to obtain the reconstructed temperature field, stress field, strain field, and displacement field of the turbine rotor.

2. The method according to claim 1, wherein The first reconstruction model and the second reconstruction model include: Constructing a first reconstruction model T-GCN by cascading a convolutional neural network CNN and a graph convolutional neural network GCN, and training T-GCN; Using the same network structure as GCN in T-GCN as a second reconstruction model S-GCN for stress-strain-displacement prediction, and training S-GCN.

3. The method according to claim 2, wherein Constructing a data sample set for starting a gas turbine to train T-GCN and S-GCN. The construction of the data sample set for starting a gas turbine includes: Arranging temperature sensors in the regions with significant surface temperature gradients of the turbine rotor to obtain real-time temperature data; Based on the real-time temperature data, parametrically modeling the average exhaust temperature curve in the load increase stage during the starting process of the gas turbine to generate a sample set including two starting types: cold start and warm start; Using the transient thermo-solid coupling finite element method, combining with the sample set generated by the parametric modeling, calculating the temperature, stress, strain, and displacement distribution data of the turbine rotor, and correcting the numerical model through the measured data of the corresponding physical parameter sensors; Based on the corrected numerical model, extracting the three-dimensional coordinates, temperature, stress, strain, and displacement distribution data of the preset nodes of the turbine rotor, and the node temperature data corresponding to the positions of the temperature sensors to construct a training data set.

4. The method according to claim 3, wherein The parametric modeling uses the Latin hypercube sampling method to sample the average exhaust temperature curves in the load increase stages of cold start and warm start to generate a sample set of starting curves defined by 5 key control points, where the initial temperature and load increase time are dynamically adjusted according to the starting type.

5. The method according to claim 4, wherein The training of T-GCN and S-GCN includes: Dividing the data sample set in a ratio of 7:3, and taking 70% of the sample quantity of the data sample set as the training set; Training T-GCN and S-GCN using the Adam and Adamax optimization algorithms respectively; Setting initial learning rates for the CNN and GCN respectively, and dynamically adjusting the initial learning rates based on the mean square error MSE as the loss function; the expression of the loss function is: ； ； Among them, is the turbine rotor temperature distribution data output by T-GCN, , and are the stress, strain and displacement distribution data of the turbine rotor output by S-GCN respectively, , , and are the temperature, stress, strain and displacement distribution data of the real turbine rotor respectively; N is the number of samples in the data sample set, and c is the c-th sample data in the data sample set; If the value of the loss function exceeds a preset threshold, optimizing the network parameters of the cascaded graph combined convolutional network through the Adam optimization algorithm and the Adamax optimization algorithm until the value of the loss function is less than or equal to the preset threshold.

6. The method according to claim 2, wherein The temperature, stress, strain, and displacement distribution data of the turbine rotor predicted by the reconstruction model include: Taking the preset node temperature and position of the turbine rotor as inputs, the real-time temperature and position of the preset nodes are dimensionally increased through the fully connected layer of the CNN in the first reconstruction model T-GCN. The spatial features are reconstructed through the upsampling layer, and the correlation features between the nodes of the spatial features are extracted through the multi-layer graph convolutional kernel of the GCN and the Gaussian deviation linear unit GELU activation function. The temperature distribution data is predicted and output through the correlation features. The temperature distribution data is used as the input of the second reconstruction model S-GCN to obtain the output stress, strain, and displacement distribution data.

7. A physical field reconstruction device for a gas turbine turbine rotor, characterized in that, The device includes: An acquisition module, configured to divide the turbine rotor into multiple nodes based on a grid of a preset size, and acquire the temperature data and node positions of the preset nodes of the turbine rotor; A prediction module, configured to predict the temperature distribution data of all nodes of the turbine rotor through the first reconstruction model based on the temperature data and node positions of the preset nodes; and further predict the stress, strain, and displacement distribution data of the turbine rotor through the second reconstruction model based on the temperature distribution data of all nodes of the turbine rotor; A reconstruction module, configured to perform linear fitting on the temperature, stress, strain, and displacement distribution data respectively to obtain the reconstructed temperature field, stress field, strain field, and displacement field of the turbine rotor.

8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, and when the computer program is executed by a processor, the method described in any one of claims 1 to 6 above is implemented.

9. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the method described in any one of claims 1 to 6 above is implemented.

Images