Rotor fatigue loss online detection method and device based on full convolutional neural network

Abstract

The embodiment of the invention provides a rotor fatigue loss online detection method and device based on a full convolutional neural network. The method comprises the following steps: constructing a turbine rotor geometric model; determining a temperature monitoring point in the rotor geometric model through a finite element analysis tool; performing stress prediction according to the temperature parameters of the temperature monitoring points through a pre-constructed rotor stress field reconstruction model, and generating a rotor equivalent stress; the method comprises the following steps: generating rotor fatigue loss according to rotor equivalent stress through a pre-constructed online calculation model of the fatigue loss of the steam turbine rotor, and directly predicting stress distribution of the rotor according to temperature results of a small number of measuring points in a peak shaving process of the rotor through a rotor stress field reconstruction model of a deep full convolutional neural network. According to the method, high-precision and high-efficiency stress prediction can be provided for the steam turbine under different peak regulation working conditions, so that the fatigue loss of the rotor is accurately detected in real time, and safe operation of the steam turbine is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of energy and power technology, and in particular to a method and apparatus for online detection of rotor fatigue loss based on a fully convolutional neural network. Background Technology

[0002] The high- and intermediate-pressure rotor of a steam turbine is a critical high-temperature rotating component in a generator set. During operation, it not only bears its own weight and the centrifugal force generated by rotation, but also the tensile force from the centrifugal force of the blades, as well as the thermal stress caused by the temperature difference between the inside and outside due to its large volume. Compared with the long-term stable full-load operation mode, units participating in peak shaving must be able to frequently start and stop and handle large load changes during flexible operation. However, for the unit, shorter start-up times often mean more drastic load changes. The huge temperature difference between the inside and outside of the rotor will cause the thermal stress level to rise sharply, and the rotor will have to withstand more severe alternating mechanical and thermal loads, which will seriously impair the safety and stability of the unit's operation. Therefore, it is very important to conduct research on real-time detection of steam turbine rotor fatigue loss. In related technologies, steam turbine rotor analysis is usually performed through numerical calculation methods. However, this analysis method involves a large amount of calculation and is difficult to achieve real-time detection, thus failing to guarantee the safe operation of the steam turbine. Summary of the Invention

[0003] One objective of this invention is to provide an online rotor fatigue loss detection method based on a fully convolutional neural network. This method directly predicts the rotor stress distribution based on temperature results from a small number of measuring points during rotor peak shaving using a rotor stress field reconstruction model constructed from a deep fully convolutional neural network. This provides high-precision and high-efficiency stress prediction for steam turbines under different peak shaving conditions, thereby enabling real-time and accurate detection of rotor fatigue loss and ensuring the safe operation of the steam turbine. Another objective of this invention is to provide an online rotor fatigue loss detection device based on a fully convolutional neural network. A further objective of this invention is to provide a computer-readable medium. A final objective of this invention is to provide a computer device.

[0004] To achieve the above objectives, this invention discloses, in one aspect, an online rotor fatigue loss detection method based on a fully convolutional neural network, comprising:

[0005] Construct a geometric model of the steam turbine rotor;

[0006] The temperature monitoring points in the rotor geometric model were determined using finite element analysis tools.

[0007] By using a pre-built rotor stress field reconstruction model, stress prediction is performed based on the temperature parameters of the temperature monitoring points to generate rotor equivalent stress. The stress field reconstruction model is obtained by training a fully convolutional neural network.

[0008] By using a pre-built online calculation model for turbine rotor fatigue loss, rotor fatigue loss is generated based on the rotor equivalent stress.

[0009] Preferably, the construction of the turbine rotor geometric model includes:

[0010] Obtain the turbine rotor assembly drawing;

[0011] Using finite element analysis tools, an initial rotor geometric model was constructed based on the turbine rotor assembly drawing;

[0012] The initial rotor geometry model is used to refine the mesh, thereby generating the turbine rotor geometry model.

[0013] Preferably, the temperature monitoring points in the rotor geometric model are determined using finite element analysis tools, including:

[0014] Using finite element analysis tools, the rotor stress field results are generated based on preset boundary conditions;

[0015] Based on the rotor stress field results, the temperature monitoring points were determined.

[0016] Preferably, the method further includes:

[0017] A model dataset was constructed using finite element analysis tools based on temperature monitoring points.

[0018] A fully convolutional neural network was trained using the model dataset to construct a rotor stress field reconstruction model.

[0019] Preferably, a model dataset is constructed using finite element analysis tools based on temperature monitoring points, including:

[0020] Under different peak-shaving conditions, temperature data is collected from temperature monitoring points according to a preset time series to generate corresponding temperature time series data.

[0021] By using a pre-built main steam temperature time model, the temperature time series data is parameterized to generate the main steam temperature and the temperature change rate for each time period.

[0022] Using finite element analysis tools, the corresponding actual stress values ​​are generated based on the main steam temperature for each time period.

[0023] A model dataset is constructed based on the main steam temperature, the rate of temperature change, and the actual stress value for each time period.

[0024] Preferably, the model dataset includes a training dataset and a test dataset;

[0025] Using the model dataset, a fully convolutional neural network is trained to construct a rotor stress field reconstruction model, including:

[0026] Based on the training dataset, the fully convolutional neural network is trained to generate an initial reconstructed model;

[0027] By using a preset loss function, the initial reconstruction model is optimized based on the test dataset to generate a rotor stress field reconstruction model.

[0028] This invention also discloses an online rotor fatigue loss detection device based on a fully convolutional neural network, comprising:

[0029] Geometric model building unit, used to build the geometric model of the steam turbine rotor;

[0030] The temperature monitoring point determination unit is used to determine the temperature monitoring points in the rotor geometric model using finite element analysis tools.

[0031] The equivalent stress prediction unit is used to predict the stress based on the temperature parameters of the temperature monitoring point through a pre-built rotor stress field reconstruction model, and generate the rotor equivalent stress. The stress field reconstruction model is obtained by training a fully convolutional neural network.

[0032] The rotor fatigue loss detection unit is used to generate rotor fatigue loss based on the rotor equivalent stress by using a pre-built online calculation model of turbine rotor fatigue loss.

[0033] Preferably, the geometric model building unit is specifically used to obtain the turbine rotor assembly drawing; to construct an initial rotor geometric model based on the turbine rotor assembly drawing using a finite element analysis tool; and to refine the mesh based on the initial rotor geometric model to generate the turbine rotor geometric model.

[0034] Preferably, the temperature monitoring point determination unit is specifically used to generate rotor stress field results based on preset boundary conditions using finite element analysis tools; and to determine the temperature monitoring point based on the rotor stress field results.

[0035] Preferably, the device further includes:

[0036] The dataset building unit is used to build a model dataset based on temperature monitoring points using finite element analysis tools.

[0037] The reconstructed model training unit is used to train a fully convolutional neural network using a model dataset to construct a rotor stress field reconstruction model.

[0038] Preferably, the dataset construction unit is specifically used to collect temperature data from temperature monitoring points according to a preset time series under different peak-shaving conditions, generating corresponding temperature time-series data; parameterizing the temperature time-series data using a pre-built main steam temperature time model, generating the main steam temperature and temperature change rate for each time period; generating the corresponding actual stress value based on the main steam temperature for each time period using a finite element analysis tool; and constructing a model dataset based on the main steam temperature, temperature change rate, and actual stress value for each time period.

[0039] Preferably, the model dataset includes a training dataset and a test dataset;

[0040] The reconstructed model training unit is specifically used to train a fully convolutional neural network based on the training dataset to generate an initial reconstructed model; and to optimize the initial reconstructed model based on the test dataset using a preset loss function to generate a rotor stress field reconstructed model.

[0041] The present invention also discloses a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the method described above.

[0042] The present invention also discloses a computer device, including a memory and a processor, wherein the memory is used to store information including program instructions, and the processor is used to control the execution of the program instructions, wherein the processor executes the program to implement the method described above.

[0043] The present invention also discloses a computer program product, including a computer program / instruction, which, when executed by a processor, implements the method described above.

[0044] This invention constructs a turbine rotor geometric model; uses finite element analysis tools to determine the temperature monitoring points in the rotor geometric model; uses a pre-constructed rotor stress field reconstruction model to predict stress based on the temperature parameters of the monitoring points, generating equivalent rotor stress; uses a pre-constructed online turbine rotor fatigue loss calculation model to generate rotor fatigue loss based on the equivalent rotor stress; and uses a deep fully convolutional neural network-based rotor stress field reconstruction model to directly predict the rotor stress distribution based on the temperature results of a small number of measuring points during rotor peak shaving. This provides high-precision and high-efficiency stress prediction for turbines under different peak shaving conditions, thereby enabling real-time and accurate detection of rotor fatigue loss and ensuring the safe operation of the turbine. Attached Figure Description

[0045] 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.

[0046] Figure 1 A flowchart of an online rotor fatigue loss detection method based on a fully convolutional neural network provided in this embodiment of the invention;

[0047] Figure 2 A flowchart illustrating another online rotor fatigue loss detection method based on a fully convolutional neural network provided in this embodiment of the invention;

[0048] Figure 3 This is a schematic diagram showing the distribution of temperature detection points on the outer surface of a rotor, provided in an embodiment of the present invention.

[0049] Figure 4 A schematic diagram of the structure of an online rotor fatigue loss detection device based on a fully convolutional neural network provided in an embodiment of the present invention;

[0050] Figure 5 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0051] 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.

[0052] It should be noted that the rotor fatigue loss online detection method and device based on fully convolutional neural network disclosed in this application can be used in the field of artificial intelligence technology, or in any field other than artificial intelligence technology. The application field of the rotor fatigue loss online detection method and device based on fully convolutional neural network disclosed in this application is not limited.

[0053] To facilitate understanding of the technical solution provided in this application, the relevant content of the technical solution will be explained below. Deep learning algorithms are fast and accurate, providing new ideas for solving many engineering problems. A Fully Convolutional Network (FCN) is a deep learning model for image segmentation tasks. It retains the feature maps of all convolutional layers in a convolutional neural network and performs pixel-by-pixel classification on these feature maps, resulting in fewer parameters and a stronger ability to acquire key physical field information.

[0054] This invention calculates the rotor temperature and stress fields using finite element analysis under different peak-shaving conditions. Relevant temperature and stress field calculation data are extracted to establish a dataset, and a rotor stress field reconstruction model based on a deep fully convolutional neural network (FCN) is established. This model can directly predict the rotor stress distribution based on temperature results from a small number of measurement points during peak-shaving. The data-driven deep learning model established in this invention has advantages such as high accuracy, fast response, and strong stability. It can provide high-precision and high-efficiency stress prediction for steam turbines under different peak-shaving conditions, thereby enabling online detection of steam turbine rotor fatigue loss.

[0055] The following example uses a rotor fatigue loss online detection device based on a fully convolutional neural network as the execution subject to illustrate the implementation process of the rotor fatigue loss online detection method based on a fully convolutional neural network provided in this embodiment of the invention. It is understood that the execution subject of the rotor fatigue loss online detection method based on a fully convolutional neural network provided in this embodiment of the invention includes, but is not limited to, a rotor fatigue loss online detection device based on a fully convolutional neural network.

[0056] Figure 1 A flowchart of an online rotor fatigue loss detection method based on a fully convolutional neural network, provided as an embodiment of the present invention, is shown below. Figure 1 As shown, the method includes:

[0057] Step 101: Construct the geometric model of the steam turbine rotor.

[0058] In this embodiment of the invention, a geometric model is generated from the turbine rotor configuration diagram using a finite element analysis tool.

[0059] Step 102: Use finite element analysis tools to determine the temperature monitoring points in the rotor geometric model.

[0060] Step 103: Using the pre-constructed rotor stress field reconstruction model, stress prediction is performed based on the temperature parameters of the temperature monitoring points to generate rotor equivalent stress.

[0061] In this embodiment of the invention, the stress field reconstruction model is obtained by training a fully convolutional neural network (FCN).

[0062] Step 104: Using a pre-built online calculation model for turbine rotor fatigue loss, generate rotor fatigue loss based on rotor equivalent stress.

[0063] In this embodiment of the invention, the online calculation model for turbine rotor fatigue loss is a real-time calculation model based on Miner linear cumulative fatigue loss calculation, which calculates the fatigue loss generated by the rotor during each start-up or shutdown process. It includes the relationship between the number of cycles and strain, and the loss-stress fitting relationship.

[0064] In the technical solution provided by this invention, a turbine rotor geometric model is constructed; temperature monitoring points in the rotor geometric model are determined using finite element analysis tools; stress prediction is performed based on the temperature parameters of the temperature monitoring points using a pre-constructed rotor stress field reconstruction model to generate rotor equivalent stress; rotor fatigue loss is generated based on the rotor equivalent stress using a pre-constructed online calculation model for turbine rotor fatigue loss; and the rotor stress distribution is directly predicted based on the temperature results of a small number of measuring points during rotor peak shaving using a deep fully convolutional neural network rotor stress field reconstruction model. This provides high-precision and high-efficiency stress prediction for the turbine under different peak shaving conditions, thereby enabling real-time and accurate detection of rotor fatigue loss and ensuring the safe operation of the turbine.

[0065] Figure 2 A flowchart of another online rotor fatigue loss detection method based on a fully convolutional neural network provided in this embodiment of the invention is shown below. Figure 2 As shown, the method includes:

[0066] Step 201: Obtain the turbine rotor assembly drawing.

[0067] In this embodiment of the invention, each step is performed directly by a rotor fatigue loss online detection device based on a fully convolutional neural network.

[0068] In this embodiment of the invention, the turbine rotor is a high-pressure and intermediate-pressure rotor. The high-pressure and intermediate-pressure rotor assembly drawing is usually used to show the composition of the high-pressure and intermediate-pressure rotor components in mechanical equipment such as steam turbines. It can show in detail the relative positions, connection methods and assembly sequence of each part.

[0069] Step 202: Using finite element analysis tools, construct an initial rotor geometric model based on the turbine rotor assembly drawing.

[0070] In this embodiment of the invention, the finite element analysis tool is a numerical method used to predict the performance of products in the real world. It is widely used in engineering design to solve complex mechanical, thermal, and electromagnetic problems. Finite element analysis solves a large problem by decomposing it into many small, simple parts, which are called "elements". The solutions of each element are combined to obtain an approximate solution for the entire system.

[0071] Specifically, the turbine rotor assembly drawing is input into a finite element analysis tool for geometric modeling. The blades and tooth grooves at each stage are appropriately simplified to construct an initial rotor geometric model.

[0072] It is worth noting that the selection of a suitable finite element analysis tool depends on the specific needs of the project, including factors such as budget, required analysis type, and user skill level. This embodiment of the invention does not impose any limitations on these factors.

[0073] Step 203: Refine the mesh based on the initial rotor geometry model to generate the turbine rotor geometry model.

[0074] In this embodiment of the invention, during mesh generation, there is a large stress concentration phenomenon at the fillets and abrupt shape changes between each stage of the rotor, and the heat transfer on the rotor surface is complex. During the variable operating conditions, the thermal stress in these areas may be very large, and the maximum equivalent stress may exceed the yield limit of the material at the corresponding temperature, resulting in a large amount of plastic deformation. This causes the rotor to produce a large amount of low-cycle fatigue loss. Therefore, it is necessary to refine the mesh separately for these areas to improve the accuracy of the calculation.

[0075] Specifically, stress concentration areas are identified; the mesh of the stress concentration areas is refined to generate a turbine rotor geometric model. The stress concentration areas include, but are not limited to, rounded transition areas, abrupt shape changes (such as blade roots and disk-shaft connections), and areas with significant temperature differences between the surface and the interior.

[0076] Furthermore, based on the steam parameters, empirical formulas are used to calculate the heat transfer coefficients of various parts of the rotor. These steam parameters include, but are not limited to, temperature, pressure, and flow rate; the empirical formulas include, but are not limited to, the Dittus-Boelter formula and the Gnielinski formula; and the heat transfer coefficients include, but are not limited to, the convective heat transfer coefficient.

[0077] It is worth noting that the empirical formula can also be selected according to the heat transfer coefficient to be calculated and the actual situation, and the embodiments of the present invention do not limit this.

[0078] Step 204: Using the finite element analysis tool, generate the rotor stress field results based on the preset boundary conditions.

[0079] In this embodiment of the invention, the boundary conditions include thermal boundary conditions and transient thermal analysis boundary conditions. It is worth noting that the thermal boundary conditions and transient thermal analysis boundary conditions can be set according to actual needs and experience, and this embodiment of the invention does not impose any limitations on this.

[0080] In this embodiment of the invention, the initial thermal boundary conditions of the rotor are determined based on steam parameters and heat transfer coefficients. These initial thermal boundary conditions are then applied to the turbine rotor geometric model, and the temperature distribution when the rotor reaches a stable operating state, i.e., the steady-state temperature field, is calculated using finite element analysis tools. The steady-state temperature field is used as the initial condition for transient analysis. Transient temperature field analysis is performed using finite element analysis tools based on the transient thermal analysis boundary conditions to obtain the temperature field results. The temperature field results obtained from the transient thermal analysis are imported into the structural analysis module, and a suitable element type is selected (converting from thermal elements to structural elements). Next, the physical properties of the rotor material (such as elastic modulus, Poisson's ratio, coefficient of thermal expansion, etc.) and mechanical boundary conditions (including constraints and external loads) are input to calculate the stress distribution inside the rotor under different operating conditions, generating the rotor stress field results.

[0081] Step 205: Determine the temperature monitoring point based on the rotor stress field results.

[0082] In this embodiment of the invention, given that a high level of thermal stress may occur on the outer surface of the rotor when the main steam temperature rises, especially stress concentration may occur on the steam inlet side, steam outlet side, each stage of the impeller and the steam seal position, multiple key temperature monitoring points can be selected for temperature detection. Figure 3 This is a schematic diagram showing the distribution of temperature detection points on the outer surface of a rotor, as provided in an embodiment of the present invention. Figure 3 As shown, the rotor geometric model is a two-dimensional finite element mesh model. Fourteen key temperature monitoring points were selected on the outer surface of the rotor, numbered 1 to 14 from left to right. The mesh section represents the internal region of the rotor geometric model. In-depth analysis of the stress values ​​at these temperature monitoring points allows for more accurate detection of the rotor's safety performance under actual operating conditions and provides a basis for optimized design. In summary, this method, which comprehensively considers heat conduction and mechanical response, can comprehensively and accurately evaluate the behavior characteristics of the turbine rotor under dynamic operating conditions, which is of great significance for preventing failures and extending equipment service life.

[0083] Step 206: Using finite element analysis tools, construct a model dataset based on temperature monitoring points.

[0084] In this embodiment of the invention, step 206 specifically includes:

[0085] Step 2061: Under different peak-shaving conditions, collect temperature data from temperature monitoring points according to a preset time series to generate corresponding temperature time series data.

[0086] In this embodiment of the invention, during the peak shaving process of the steam turbine, the temperature of each temperature monitoring point is collected according to a preset time sequence for different peak shaving conditions to generate temperature time sequence data.

[0087] Step 2062: Parameterize the temperature time series data using a pre-built main steam temperature time model to generate the main steam temperature and the temperature change rate for each time period.

[0088] In this embodiment of the invention, based on temperature time-series data, the main steam temperature during the turbine peak-shaving process is parametrically modeled to construct a main steam temperature time model, specifically as follows:

[0089] T = α i (tt i )+T i,b ,t i ≤t<t i+1 i = 1, 2, K, N s

[0090] Where T is the main steam temperature; t is the current time; t i For the i-th time period; t i+1 For the (i+1)th time period; α i The rate of temperature change over the i-th time interval can be calculated from temperature time series data based on the time span of the i-th time interval; T i,b The initial temperature of the stage within the i-th time period can be obtained by querying the temperature time series data; N s This represents the number of segments in the time period.

[0091] It is worth noting that the time span can be set according to actual needs, and this embodiment of the invention does not limit it.

[0092] In this embodiment of the invention, the main steam temperature T is a piecewise function of the current time t, and the main steam temperature has a linear relationship with the current time in each time segment.

[0093] Furthermore, the Latin hypercube sampling method was used to sample different peak-shaving conditions to obtain sample peak-shaving conditions. For the sample peak-shaving conditions, temperature data was collected from temperature monitoring points according to a preset time series to generate corresponding temperature time-series data. The temperature time-series data was parameterized using a pre-constructed main steam temperature time model to generate the main steam temperature and temperature change rate for each time period. Based on the main steam temperature and temperature change rate corresponding to the sample peak-shaving conditions, subsequent model datasets were constructed.

[0094] Latin Hypercube Sampling (LHS) is a statistical sampling method commonly used to draw samples from multidimensional distributions. Compared to simple random sampling, LHS can more uniformly cover the entire input space, thereby improving the accuracy and stability of model predictions.

[0095] Step 2063: Using finite element analysis tools, generate the corresponding actual stress values ​​based on the main steam temperature for each time period.

[0096] In this embodiment of the invention, the main steam temperature corresponding to each time period is input into the finite element analysis tool to calculate the stress value, and the corresponding actual stress value is output.

[0097] Step 2064: Construct a model dataset based on the main steam temperature, the rate of temperature change, and the actual stress value for each time period.

[0098] In this embodiment of the invention, the main steam temperature and the temperature change rate corresponding to each time period are used as input data; the corresponding actual stress value is used as output data; and the combination of the input data and the corresponding output data is determined as the model dataset.

[0099] In this embodiment of the invention, for a single sample, the input data consists of the main steam temperature and temperature change rate at a certain moment on the rotor surface monitoring point, while the output data consists of the stress values ​​of the grid nodes. As an optional scheme, the input data size is (2, 15), the output data size is (6, 607), and the sample size of the dataset is 15442.

[0100] Furthermore, since the input and output data have different dimensions and significant numerical differences, a min-max normalization method is used to map the input and output data to the interval [0,1] in order to remove the dimensionality and accelerate the convergence speed of the model during training.

[0101]

[0102] Where x is the original data (input data / output data) that needs to be normalized; The data is normalized; min(x) is the minimum value in the original data that needs to be normalized; max(x) is the maximum value in the original data that needs to be normalized.

[0103] Step 207: Train the fully convolutional neural network using the model dataset to construct a rotor stress field reconstruction model.

[0104] In this embodiment of the invention, the model dataset includes a training dataset and a test dataset. Specifically, the normalized model dataset is divided into a training dataset and a test dataset according to a preset ratio.

[0105] It is worth noting that the preset ratio can be set according to actual needs, and this embodiment of the invention does not limit this. As an optional solution, the ratio between the training dataset and the test dataset is 7:3.

[0106] In this embodiment of the invention, step 207 specifically includes:

[0107] Step 2071: Train the fully convolutional neural network (FCN) based on the training dataset to generate the initial reconstruction model.

[0108] In this embodiment of the invention, considering that the number of rows of the input and output tensors in a single sample is small, a rotor stress field reconstruction model based on finite measurement points is established using FCN. The feature width is reduced and the feature depth is increased by using a downsampling layer to transform the low-dimensional features of the input data into high-dimensional features. Then, the high-dimensional features of the extracted input data are transformed into low-dimensional features of the output data by using an upsampling layer, thereby realizing the stress field reconstruction of the turbine rotor.

[0109] This invention employs an open-source deep learning framework to establish and train an FCN stress field reconstruction model. The deep neural network structure mainly consists of multiple downsampling layers, upsampling layers, and an output layer. The downsampling layers consist of multiple convolutional layers and max-pooling layers, the upsampling layers consist of multiple interpolation layers and convolutional layers, and the output layer is a single convolutional layer. As an optional approach, the kernel size of the downsampling layers is always 3, and the stride is always 1. The kernel parameters of the upsampling layers are the same as those of the downsampling layers.

[0110] It is worth noting that the convolution kernel and stride of the upsampling layer and the convolution kernel and stride of the downsampling layer can be set according to actual needs, and the embodiments of the present invention do not limit this.

[0111] Specifically, the training dataset is input into the FCN for iterative training to generate the initial reconstructed model.

[0112] Step 2072: Optimize the initial reconstruction model using a preset loss function based on the test dataset to generate a rotor stress field reconstruction model.

[0113] Specifically, the input data from the test dataset is input into the initial reconstruction model, and the output is the predicted data; the loss function is used to calculate the loss between the predicted data and the output data from the test dataset to generate the loss result; the model parameters of the initial reconstruction model are adjusted according to the loss result to achieve model optimization and generate the rotor stress field reconstruction model.

[0114] In this embodiment of the invention: FCN uses a pixel-level loss function (e.g., cross-entropy loss) to detect the classification accuracy of each pixel. This allows the network to update each weight through the backpropagation algorithm, thereby improving segmentation accuracy. During the training of the neural network, the model parameters are updated based on the gradient of the error, which is calculated using the loss function. In regression tasks, the mean squared error loss function is typically chosen, and its expression is as follows:

[0115]

[0116] Where, λ mse For the loss result; N Ω The quantity in the test dataset; y real For output data; y pred For prediction data.

[0117] Step 208: Using the rotor stress field reconstruction model, stress prediction is performed based on the temperature parameters of the temperature monitoring points to generate rotor equivalent stress.

[0118] In this embodiment of the invention, the temperature parameters include the main steam temperature and the corresponding temperature change rate.

[0119] Specifically, the main steam temperature and the corresponding temperature change rate at the temperature monitoring point are input into the rotor stress field reconstruction model to predict stress and generate rotor equivalent stress.

[0120] Step 209: Using a pre-built online calculation model for turbine rotor fatigue loss, generate rotor fatigue loss based on rotor equivalent stress.

[0121] In this embodiment of the invention, based on the reconstructed rotor stress field and using low-cycle fatigue theory and life loss research methods, an online calculation model for turbine rotor fatigue loss is established, combined with the rotor stress-strain characteristics during peak shaving. This invention uses Miner's linear cumulative fatigue loss calculation to determine the fatigue loss generated by the rotor during each complete start-up or shutdown process:

[0122]

[0123] The relationship between the number of cycles and strain is:

[0124]

[0125] Where, φ f For rotor fatigue loss; Δε f It is the total strain of the material; N i The strain is Δε f The number of cycles in which the first macroscopic crack appears in the material, i.e., the number of cycles that cause cracking; n i The strain is Δεf The number of cycles the material undergoes; K′ is the cyclic strength coefficient; n′ is the cyclic strain hardening exponent; σ eq It is the equivalent stress; σ i ε′ represents fatigue strength; f α represents fatigue plasticity; β and β represent fatigue strength and plasticity index, respectively; E represents the elastic modulus of the material, which is an inherent property of the material.

[0126] However, during the low-cycle fatigue loss of the rotor, the number of cyclic cracking cycles N i Since the above formula cannot be calculated in real time, this invention uses polynomial fitting to obtain the relationship between the number of cyclic fracturing cycles and strain, and establishes a loss-stress fitting calculation model:

[0127]

[0128] Where, N i The strain is Δε f The number of cycles at which the first macroscopic crack appears in the material, i.e., the number of cycles leading to crack initiation; Δε f P represents the total strain of the material. i The fitting coefficients of the model are calculated to fit the model.

[0129] During the training process of the online fatigue loss calculation model for steam turbine rotors, the relationship between cycle number and strain, and the loss-stress fitting calculation model are iteratively fitted using the model dataset to generate fitting coefficients for the fitted calculation model. Subsequently, when performing online fatigue loss detection on the same material rotor, it is not necessary to recalculate the fitting coefficients; the online fatigue loss calculation model for steam turbine rotors can be directly applied.

[0130] In the application of the online calculation model for turbine rotor fatigue loss, step 209 specifically includes:

[0131] Step 2091: Generate the total strain of the material based on the rotor equivalent stress.

[0132] Specifically, through The equivalent stress of the rotor, cyclic strength coefficient, cyclic strain hardening exponent, and elastic modulus of the material are calculated to generate the total strain of the material. Wherein, Δε f It is the total strain of the material; σ eq is the rotor equivalent stress; K′ is the cyclic strength coefficient; n′ is the cyclic strain hardening index; E is the elastic modulus of the material, which is an inherent property of the material.

[0133] Step 2092: Calculate the number of cyclic fracturing cycles based on the total material strain.

[0134] Specifically, through the following formula:

[0135]

[0136] The total material strain and the pre-trained fitting coefficients are used to calculate the number of cyclic fracture initiation cycles. Wherein, N... i The strain is Δε f The number of cycles at which the first macroscopic crack appears in the material, i.e., the number of cycles leading to crack initiation; Δε f P represents the total strain of the material. i The fitting coefficients of the model are calculated to fit the model.

[0137] Step 2093: Generate rotor fatigue loss based on the number of cyclic fracturing cycles and the number of cycles experienced by the obtained material.

[0138] Specifically, through The rotor fatigue loss is generated by calculating the number of cyclic fracturing cycles and the number of cycles experienced by the obtained material. Wherein, φ f For rotor fatigue loss; N i The strain is Δε f The number of cycles in which the first macroscopic crack appears in the material, i.e., the number of cycles that cause cracking; n i The strain is Δε f The number of cycles the material undergoes.

[0139] It is worth noting that the acquisition, storage, use, and processing of data in the technical solution of this application all comply with relevant laws and regulations. The user information in the embodiments of this application was obtained through legal and compliant means, and the acquisition, storage, use, and processing of user information have been authorized and agreed upon by the client.

[0140] It is worth noting that the information collected in this application is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with the relevant laws, regulations and standards of the relevant countries and regions, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation portals are provided for users to choose to authorize or refuse.

[0141] It is worth noting that the technical solution provided in this application provides users with a corresponding operation entry point, allowing users to choose to agree to or reject the automated decision-making result; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0142] This invention integrates the advantages and disadvantages of existing methods for calculating low-cycle fatigue loss of turbine rotors, and improves and innovates methods for predicting low-cycle fatigue loss of high- and medium-pressure rotors in supercritical turbines. It proposes a method for detecting rotor fatigue loss in supercritical turbine units based on FCN (Fatigue-Coefficient of Performance). Based on a rotor stress field reconstruction model and low-cycle fatigue life theory, combined with the rotor stress-strain characteristics during peak shaving, a real-time online calculation and detection method for fatigue loss of high- and medium-pressure rotors in supercritical turbines is established. This method has advantages such as high accuracy, fast response, and strong stability, and can provide high-precision and high-efficiency real-time stress prediction for turbines under different peak shaving conditions. It also effectively improves the problem of the difficulty in reconciling online calculation and high-precision calculation of rotor low-cycle fatigue loss.

[0143] The technical solution of the online rotor fatigue loss detection method based on fully convolutional neural networks provided in this invention involves constructing a turbine rotor geometric model; determining temperature monitoring points in the rotor geometric model using finite element analysis tools; generating rotor equivalent stress by predicting stress based on temperature parameters at the monitoring points using a pre-constructed rotor stress field reconstruction model; generating rotor fatigue loss based on rotor equivalent stress using a pre-constructed online turbine rotor fatigue loss calculation model; and directly predicting the rotor stress distribution based on temperature results from a small number of measuring points during rotor peak shaving using a deep fully convolutional neural network rotor stress field reconstruction model. This provides high-precision and high-efficiency stress prediction for turbines under different peak shaving conditions, thereby enabling real-time and accurate detection of rotor fatigue loss and ensuring the safe operation of the turbine.

[0144] Figure 4 This is a schematic diagram of a rotor fatigue loss online detection device based on a fully convolutional neural network, provided in an embodiment of the present invention. This device is used to execute the aforementioned rotor fatigue loss online detection method based on a fully convolutional neural network. Figure 4 As shown, the device includes: a geometric model construction unit 11, a temperature monitoring point determination unit 12, an equivalent stress prediction unit 13, and a rotor fatigue loss detection unit 14.

[0145] The geometric model building unit 11 is used to build the geometric model of the steam turbine rotor.

[0146] The temperature monitoring point determination unit 12 is used to determine the temperature monitoring points in the rotor geometric model using finite element analysis tools.

[0147] The equivalent stress prediction unit 13 is used to predict the stress based on the temperature parameters of the temperature monitoring point by using a pre-built rotor stress field reconstruction model, and generate the rotor equivalent stress.

[0148] The rotor fatigue loss detection unit 14 is used to generate rotor fatigue loss based on the rotor equivalent stress by using a pre-built online calculation model of turbine rotor fatigue loss.

[0149] In this embodiment of the invention, the geometric model construction unit 11 is specifically used to obtain the turbine rotor assembly drawing; construct an initial rotor geometric model based on the turbine rotor assembly drawing using a finite element analysis tool; and refine the mesh based on the initial rotor geometric model to generate a turbine rotor geometric model.

[0150] In this embodiment of the invention, the temperature monitoring point determination unit 12 is specifically used to generate rotor stress field results based on preset boundary conditions using finite element analysis tools; and to determine the temperature monitoring point based on the rotor stress field results.

[0151] In this embodiment of the invention, the device further includes a dataset construction unit 15 and a reconstructed model training unit 16.

[0152] Dataset building unit 15 is used to build a model dataset based on temperature monitoring points using finite element analysis tools.

[0153] The reconstruction model training unit 16 is used to train a fully convolutional neural network using a model dataset to construct a rotor stress field reconstruction model.

[0154] In this embodiment of the invention, the dataset construction unit 15 is specifically used to collect temperature data from temperature monitoring points according to a preset time series under different peak-shaving conditions, and generate corresponding temperature time series data; to parameterize the temperature time series data through a pre-constructed main steam temperature time model, generating the main steam temperature and the temperature change rate corresponding to each time period; to generate the corresponding actual stress value based on the main steam temperature corresponding to each time period using a finite element analysis tool; and to construct a model dataset based on the main steam temperature, the temperature change rate, and the corresponding actual stress value corresponding to each time period.

[0155] In this embodiment of the invention, the model dataset includes a training dataset and a test dataset; the reconstruction model training unit 16 is specifically used to train a fully convolutional neural network based on the training dataset to generate an initial reconstruction model; and to optimize the initial reconstruction model based on the test dataset using a preset loss function to generate a rotor stress field reconstruction model.

[0156] In this embodiment of the invention, a turbine rotor geometric model is constructed; temperature monitoring points in the rotor geometric model are determined using finite element analysis tools; stress prediction is performed based on the temperature parameters of the temperature monitoring points using a pre-constructed rotor stress field reconstruction model to generate equivalent rotor stress; rotor fatigue loss is generated based on the equivalent rotor stress using a pre-constructed online calculation model for turbine rotor fatigue loss; and the rotor stress distribution is directly predicted based on the temperature results of a small number of measuring points during rotor peak shaving using a deep fully convolutional neural network rotor stress field reconstruction model. This provides high-precision and high-efficiency stress prediction for the turbine under different peak shaving conditions, thereby enabling real-time and accurate detection of rotor fatigue loss and ensuring the safe operation of the turbine.

[0157] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer device, specifically, a computer device can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0158] This invention provides a computer device, including a memory and a processor. The memory stores information including program instructions, and the processor controls the execution of the program instructions. When the program instructions are loaded and executed by the processor, they implement the steps of the above-described embodiment of the online rotor fatigue loss detection method based on a fully convolutional neural network. For a detailed description, please refer to the above-described embodiment of the online rotor fatigue loss detection method based on a fully convolutional neural network.

[0159] The following is for reference. Figure 5 It shows a schematic diagram of the structure of a computer device 600 suitable for implementing the embodiments of this application.

[0160] like Figure 5 As shown, the computer device 600 includes a central processing unit (CPU) 601, which can perform various appropriate tasks and processes based on programs stored in read-only memory (ROM) 602 or programs loaded from storage section 608 into random access memory (RAM) 603. The RAM 603 also stores various programs and data required for the operation of the computer device 600. The CPU 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0161] The following components are connected to I / O interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal feedback (LCD), etc., and speakers, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to I / O interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 610 as needed so that computer programs read from it can be installed in storage section 608 as needed.

[0162] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program tangibly embodied on a machine-readable medium, the computer program including program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 609, and / or installed from removable medium 611.

[0163] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0164] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0165] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0166] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0167] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0168] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0169] The acquisition, storage, use, and processing of data in this application all comply with the relevant provisions of national laws and regulations.

[0170] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0171] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0172] This application can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0173] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0174] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. An online detection method for rotor fatigue loss based on a fully convolutional neural network, characterized in that: The method comprises: Construct the turbine rotor geometry model; Determining temperature monitoring points in the rotor geometric model by using a finite element analysis tool; By using a pre-constructed rotor stress field reconstruction model, stress prediction is performed according to the temperature parameters of the temperature monitoring point to generate rotor equivalent stress, wherein the stress field reconstruction model is obtained by training a fully convolutional neural network; The rotor fatigue loss is generated according to the rotor equivalent stress by using a pre-constructed online calculation model for turbine rotor fatigue loss.

2. The online detection method for rotor fatigue loss based on a fully convolutional neural network according to claim 1 is characterized in that: The step of constructing a turbine rotor geometric model comprises: Obtain the steam turbine rotor assembly drawing; Using the finite element analysis tool, constructing an initial rotor geometry model according to the steam turbine rotor assembly drawing; The grid is encrypted according to the initial rotor geometric model to generate the turbine rotor geometric model.

3. The online detection method for rotor fatigue loss based on a fully convolutional neural network according to claim 1 is characterized in that: Determining the temperature monitoring points in the rotor geometric model by using a finite element analysis tool includes: Generate rotor stress field results through the finite element analysis tool according to preset boundary conditions; The temperature monitoring point is determined according to the rotor stress field result.

4. The online detection method for rotor fatigue loss based on a fully convolutional neural network according to claim 1 is characterized in that: The method further comprises: Using a finite element analysis tool, constructing a model data set according to the temperature monitoring points; The fully convolutional neural network is trained using the model data set to construct a rotor stress field reconstruction model.

5. The online detection method for rotor fatigue loss based on a fully convolutional neural network according to claim 4 is characterized in that: The method of constructing a model data set according to the temperature monitoring points by using a finite element analysis tool includes: Under different peak load conditions, the temperature of the temperature monitoring points is collected according to a preset time series to generate corresponding temperature time series data; The temperature time series data is parameterized by using a pre-built main steam temperature time model to generate the main steam temperature corresponding to each time period and the temperature change rate corresponding to each time period; Generate a corresponding actual stress value according to the main steam temperature corresponding to each time period by using the finite element analysis tool; The model data set is constructed according to the main steam temperature corresponding to each time period, the temperature change rate corresponding to each time period, and the corresponding actual stress value.

6. The online detection method for rotor fatigue loss based on a fully convolutional neural network according to claim 4 is characterized in that: The model data set includes a training data set and a test data set; The method of training a fully convolutional neural network through the model data set to construct a rotor stress field reconstruction model includes: According to the training data set, the fully convolutional neural network is trained to generate an initial reconstruction model; The initial reconstruction model is optimized according to the test data set through a preset loss function to generate a rotor stress field reconstruction model.

7. An online detection device for rotor fatigue loss based on a fully convolutional neural network, characterized in that: The device comprises: A geometric model building unit, used for building a geometric model of a steam turbine rotor; A temperature monitoring point determination unit, used to determine the temperature monitoring points in the rotor geometric model by using a finite element analysis tool; An equivalent stress prediction unit, used to generate rotor equivalent stress by performing stress prediction according to the temperature parameters of the temperature monitoring points through a pre-constructed rotor stress field reconstruction model, wherein the stress field reconstruction model is obtained by training a fully convolutional neural network; The rotor fatigue loss detection unit is used to generate rotor fatigue loss according to the rotor equivalent stress through a pre-built online calculation model of turbine rotor fatigue loss.

8. The online detection device for rotor fatigue loss based on a fully convolutional neural network according to claim 7 is characterized in that: The geometric model building unit is specifically used to obtain a steam turbine rotor assembly drawing; construct an initial rotor geometric model based on the steam turbine rotor assembly drawing using the finite element analysis tool; and perform mesh encryption based on the initial rotor geometric model to generate the steam turbine rotor geometric model.

9. The online detection device for rotor fatigue loss based on a fully convolutional neural network according to claim 7 is characterized in that: The temperature monitoring point determination unit is specifically used to generate rotor stress field results according to preset boundary conditions through the finite element analysis tool; and determine the temperature monitoring points according to the rotor stress field results.

10. The online detection device for rotor fatigue loss based on a fully convolutional neural network according to claim 7, characterized in that: The device also includes: A data set construction unit, used to construct a model data set according to the temperature monitoring points by using a finite element analysis tool; The reconstruction model training unit is used to train the full convolutional neural network through the model data set to construct a rotor stress field reconstruction model.

11. The online detection device for rotor fatigue loss based on a fully convolutional neural network according to claim 10 is characterized in that: The data set construction unit is specifically used to collect the temperature of the temperature monitoring points according to a preset time series under different peak load conditions to generate corresponding temperature time series data; parameterize the temperature time series data through a pre-constructed main steam temperature time model to generate the main steam temperature corresponding to each time period and the temperature change rate corresponding to each time period; and generate the corresponding actual stress value according to the main steam temperature corresponding to each time period through the finite element analysis tool; The model data set is constructed according to the main steam temperature corresponding to each time period, the temperature change rate corresponding to each time period, and the corresponding actual stress value.

12. The online detection device for rotor fatigue loss based on a fully convolutional neural network according to claim 10 is characterized in that: The model data set includes a training data set and a test data set; The reconstruction model training unit is specifically used to train the fully convolutional neural network according to the training data set to generate an initial reconstruction model; through a preset loss function, the initial reconstruction model is optimized according to the test data set to generate a rotor stress field reconstruction model.

13. A computer readable medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the online detection method for rotor fatigue loss based on a fully convolutional neural network as described in any one of claims 1 to 6 is implemented.

14. A computer device comprising a memory and a processor, wherein the memory is used to store information including program instructions, and the processor is used to control the execution of the program instructions, characterized in that: When the program instructions are loaded and executed by the processor, the online detection method for rotor fatigue loss based on a fully convolutional neural network as described in any one of claims 1 to 6 is implemented.

15. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by the processor, the online detection method for rotor fatigue loss based on a fully convolutional neural network as described in any one of claims 1 to 6 is implemented.

Images