A method for on-line monitoring of steam turbine expansion difference based on strain measurement

Abstract

The application provides a steam turbine expansion difference online monitoring method based on strain measurement, and aims to solve the defects that the direct measurement method in the prior art has higher requirements for the space size of the measured position, the area size of the measured surface and the actual structure of the unit, and the sensor range cannot meet the requirements for large-capacity steam turbine generator units and low-pressure cylinders of steam turbines with obvious Poisson effect. The application relates to the field of steam turbine expansion difference monitoring, and the method comprises a steam cylinder support structure strain measurement scheme, a steam turbine expansion difference prediction model and a steam turbine key position expansion difference monitoring and overrun early warning method. The strain gauges are arranged on the steam cylinder support structure, the strain gauge measurement results are combined with the axial gap prediction model based on the deep neural network, and the expansion difference monitoring and overrun early warning of the steam turbine key position are realized.

Description

Technical Field

[0001] This invention belongs to the field of turbine expansion difference monitoring, specifically relating to an online monitoring method for turbine expansion difference based on strain measurement. Background Technology

[0002] During turbine startup, shutdown, or deep peak-shaving operation, the inconsistent temperature changes of the cylinder and rotor lead to a difference in axial deformation between them, known as differential expansion. A positive differential expansion occurs when the rotor's axial expansion deformation is greater than that of the cylinder, and vice versa. Excessive differential expansion can cause rubbing between moving and stationary turbine components, increased unit vibration, and even serious accidents such as blade loss and shaft bending, severely impacting the safe operation of the turbine generator unit. Therefore, turbine differential expansion monitoring is a crucial aspect of turbine operation regulation and protection, and accurate and reliable assessment methods are key to effective differential expansion monitoring.

[0003] Currently, the measurement of turbine expansion difference mainly relies on arranging large-diameter eddy current displacement sensors at the turbine thrust bearing position. This direct measurement method has high requirements for the size of the space at the measured position, the area of ​​the measured surface, and the actual structure of the unit. Furthermore, for large-capacity turbine generator sets and turbine low-pressure cylinders with significant Poisson effects, the sensor range may not meet the requirements. Summary of the Invention

[0004] To address the shortcomings of existing methods for measuring turbine expansion differential, this invention provides an online monitoring method for turbine expansion differential based on strain measurement. This method achieves expansion differential monitoring and over-limit early warning at key locations of the turbine by arranging strain gauges on the cylinder support structure and combining the strain gauge measurement results with an axial clearance prediction model based on a deep neural network.

[0005] To achieve the objective of this invention, the technical solution adopted is as follows:

[0006] A method for online monitoring of turbine expansion difference based on strain measurement includes a strain measurement scheme for cylinder support structure, a turbine expansion difference prediction model, and a method for monitoring and early warning of expansion difference at key locations of the turbine.

[0007] The cylinder support structure strain measurement scheme utilizes strain measurement equipment to monitor the strain in key areas of the cylinder's front and rear support structures in real time; the strain measurement equipment includes strain gauges, strain meters, and computers.

[0008] Preferably, the strain gauge in the above-mentioned strain measurement device is a triaxial 45° strain rosette.

[0009] The turbine expansion difference prediction model includes the following steps:

[0010] (1) Determine the temperature and force boundary conditions of the turbine cylinder and rotor under various operating conditions, and establish a three-dimensional finite element thermo-mechanical coupling analysis model of the cylinder and rotor; the boundary conditions need to be determined based on common operating conditions such as cold start, warm start, hot start, 15~100% MCR condition, and normal shutdown, and determine the temperature boundary value space of each surface of the turbine cylinder and rotor under all operating conditions. Rotor speed range ; The subscript 2 represents the working fluid temperature and surface heat transfer coefficient at the temperature boundary. m This represents the number of corresponding areas inside the cylinder and rotor where friction may occur.

[0011] (2) Using the Latin hypercube sampling method, samples are taken on the operating boundaries of the turbine cylinder and rotor to establish a sample consisting of boundary conditions under different operating conditions; the sample size of the boundary condition sample set needs to be determined. Each sample is composed of It consists of elements, among which One element represents the temperature boundary, and one element represents the force boundary. The range of values ​​for each element is evenly divided into... First, draw a number from each of the previously undrawn equal-probability regions. Then, randomly draw one number from each of the undrawn equal-probability regions for different elements, and repeat the process. This forms the boundary condition sample set. .

[0012] (3) Complete the finite element analysis of the cylinder and rotor under all boundary conditions in the sample, extract the Mises equivalent stress of the cylinder support arm as the input and the expansion difference of each interface end face of the cylinder and rotor as the output, and establish a dataset with input and output mapping relationship; the position of the support structure stress extracted in the finite element analysis results must correspond to the position of the strain gauge in the strain measurement, and the interface end face of the cylinder and rotor should include the axial end face that is prone to collision and wear, such as the stationary blades and moving blades of each stage, the front and rear steam seal teeth and the shaft shoulder.

[0013] (4) A prediction model for the differential expansion of the turbine cylinder rotor is established based on a deep neural network, and the model is trained and tested using a dataset; the deep neural network is a multi-layer fully connected neural network, which consists of... The input layer of each neuron Hidden layers of each neuron and The output layer consists of 10 neurons, in which This represents the number of strain gauges attached to the support arm. This refers to the number of end faces at the junction of the cylinder and the rotor. The number of hidden layers and neurons per layer is determined by the network's structure, but is typically no less than 64. The dataset is divided into training and testing sets in a 7:3 ratio. During network training, a total of 300 iterations are set. The optimizer is initially set to Adam with an initial learning rate of 0.001. After training reaches 100, 180, 230, and 260 steps, the learning rate is reduced to 1 / 5 of its original value. After training, if the model's error prediction accuracy on the test set is less than 98%, the number of neurons in the hidden layers is adjusted. The model is retrained until the accuracy meets the requirements, including the total number of iterations, the initial learning rate, and other hyperparameters.

[0014] Among them, the method for monitoring and warning of expansion difference at key locations of the steam turbine is to convert the strain measured by the cylinder support structure into stress and input it into the steam turbine expansion difference prediction model to obtain the expansion difference monitoring results at key locations; and to determine different expansion difference limits according to ambient temperature and operating conditions to establish a flexible expansion difference over-limit warning scheme.

[0015] The above technical solution has the following beneficial effects:

[0016] This invention proposes an online monitoring method for turbine expansion differential based on strain measurement, used for monitoring expansion differential at critical locations in the turbine and providing early warning of over-limit conditions. Unlike the traditional direct measurement method at the thrust bearing, this invention uses strain gauges placed at the cylinder support location for indirect measurement. It is not limited by the size of the space at the measured location, the area of ​​the measured surface, or the actual structure of the unit. It can monitor expansion differential at multiple axial end faces prone to rubbing, such as the stationary and moving blades of the cylinder and rotor stages, and the front and rear steam seal teeth and shaft shoulders. It has advantages such as convenient installation and abundant monitoring data, effectively ensuring the safety and stability of the unit during variable operating conditions. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the strain gauge arrangement for a low-pressure inner cylinder floor-mounted support.

[0018] Figure 2 This is a flowchart of the online monitoring method for turbine differential expansion based on strain measurement provided by the present invention;

[0019] Figure 3 This is a schematic diagram of the strain gauge measurement data processing method of the present invention;

[0020] Figure 4 This is a strain gauge measurement signal diagram from Embodiment 1 of the present invention;

[0021] Figure 5 This is a diagram of the expansion difference output signal in Embodiment 1 of the present invention. Detailed Implementation

[0022] The present invention will be further described in detail below with reference to specific examples. The following description is one application of the present invention, but it is not limited thereto; those skilled in the art can modify the parameters according to specific circumstances.

[0023] Example 1

[0024] like Figure 1 The diagram shows a strain gauge arrangement for a low-pressure inner cylinder floor-mounted support. A triaxial 45° strain rosette is arranged on the end face and belly of each support arm, and connected to a computer via a strain gauge. This embodiment combines... Figure 1 and Figure 2 The specific implementation steps are as follows:

[0025] (1) Considering common operating conditions such as cold start, warm start, hot start, 15%~100% MCR, and normal shutdown, determine the temperature boundary value space of each surface of the low-pressure inner cylinder and rotor under these operating conditions. Rotor speed range The temperature and force boundary conditions of the cylinder and rotor under various operating conditions are determined, and a three-dimensional finite element thermo-mechanical coupling analysis model of the cylinder and rotor is established. In this embodiment, m =12, , .

[0026] (2) Initialize the sample size of the boundary condition sample set In the temperature boundary value space Rotor speed range Latin hypercube sampling was performed to establish a boundary condition sample set. .

[0027] (3) Complete Finite element analysis of the cylinder and rotor under all boundary conditions is performed. The Mises equivalent stress at the location corresponding to the strain rosette position on the low-pressure cylinder support arm in the example is extracted to form the input tensor. The expansion difference between the stationary and moving blades, and between the front and rear steam seal teeth and the shaft shoulder, is extracted to form the output tensor. Establish a dataset with input-output mapping relationships. Table 1 shows the data for a sample in the mapping dataset.

[0028] Table 1 shows the data of a sample in the input-output mapping dataset of this embodiment.

[0029]

[0030] (4) A prediction model for the differential expansion of the turbine cylinder rotor is established based on a multi-layer fully connected neural network, and the model structure parameters are set as shown in Table 2. The dataset is... The network was divided into training and test sets in a 7:3 ratio. During network training, the total number of iterations was set to 300. The optimizer was initially set to Adam with an initial learning rate of 0.001. After training reached 100, 180, 230, and 260 steps, the learning rate was reduced to 1 / 5 of its original value. After training, if the model's error prediction accuracy on the test set was less than 98%, the model's structural parameters, total number of iterations, initial learning rate, and other hyperparameters were modified, and the model was retrained until the required accuracy was achieved. For example: if the model's accuracy on both the training and test sets is below 90%, you can try increasing the number of hidden layers, increasing the number of neurons in the hidden layers, reducing the Dropout in the hidden layers, increasing the total number of iterations, or increasing the initial learning rate. If the model's accuracy on the training set is above 98%, but its accuracy on the test set is below 95%, you can try reducing the number of hidden layers and neurons, or increasing the Dropout in the hidden layers. If the model's accuracy on the training set is below 60% compared to its accuracy on the test set, you need to check whether the correspondence between the input tensors and output tensors in the dataset is correct.

[0031] Table 2 shows the model structure parameters of the turbine cylinder rotor expansion difference prediction model in this embodiment.

[0032]

[0033] (5) Deploy the turbine cylinder rotor expansion difference prediction model on the computer; compile the strain data measured by the cylinder support arm for each sampling period of the strain gauge according to... Figure 3 The process is performed as shown to obtain the equivalent stress tensor corresponding to the finite element analysis. The data is then input into the turbine expansion difference prediction model to obtain the expansion difference monitoring results at key locations. .

[0034] (6) Determine different upper limits for expansion difference based on ambient temperature and operating conditions. and lower limit value When the expansion difference monitoring results of the end faces of the stationary blade and the moving blade, the front and rear steam seal teeth and the shaft shoulder, etc. If the differential expansion is outside the specified limits, an alarm will be triggered, and the turbine's operating parameters will be adjusted. Specifically, if the differential expansion monitoring result exceeds the upper alarm limit or falls below the lower alarm limit, on-site technicians must closely monitor the trend of the differential expansion monitoring results and data from other sensors, such as the shaft vibration sensor, during turbine operation. If the monitoring data remains stable, technicians can adjust the start-up or load change rate by changing the opening of the main steam regulating valve; if the monitoring data continues to increase and quickly exceeds the shutdown limit, and the rotor shaft vibration becomes increasingly severe, technicians need to consider an emergency shutdown of the turbine. Figure 4 As shown in the figure, this embodiment presents the measurement results of the strain gauge over a period of time. The measurement results are input into the turbine cylinder rotor expansion difference prediction model to obtain... Figure 5 The results of the expansion difference monitoring are shown. Figure 5 For example, the results of the expansion difference monitoring show that the changes are stable and all are within the expansion difference limit range, indicating that the turbine rotor and stator expansion coordination is good during operation and the turbine operation is stable.

[0035] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for online monitoring of turbine differential expansion based on strain measurement, characterized in that, include: Cylinder support structure strain measurement scheme: Use strain measurement equipment to monitor the strain in key areas of the front and rear support structures of the cylinder in real time; The establishment of a turbine expansion difference prediction model includes: (1) Determine the temperature and force boundary conditions of the turbine cylinder and rotor under various operating conditions, and establish a three-dimensional finite element thermo-mechanical coupling analysis model of the cylinder and rotor; (2) The Latin hypercube sampling method is used to sample the operating boundaries of the turbine cylinder and rotor to establish a sample composed of boundary conditions under different operating conditions. (3) Complete the finite element analysis of the cylinder and rotor under all boundary conditions in the sample, extract the Mises equivalent stress of the cylinder support arm as input, and the expansion difference of each interface end face of the cylinder and rotor as output, and establish a dataset with input and output mapping relationship; (4) Establish a prediction model for the differential expansion of the turbine cylinder rotor based on a deep neural network, and use the dataset to complete the training and testing of the model; Methods for monitoring and warning of expansion difference at critical locations: The strain measured in the cylinder support structure is converted into stress and input into the turbine expansion difference prediction model to obtain the expansion difference monitoring results at critical locations; different expansion difference limits are determined according to ambient temperature and operating conditions to establish a flexible expansion difference over-limit warning scheme; The strain measurement equipment includes strain gauges, a strain meter, and a computer; the strain gauges are triaxial 45° strain rosettes. The determination of the boundary conditions in step (1) is based on cold start, warm start, hot start, 15~100% MCR operating conditions, and normal shutdown operating conditions, to determine the temperature boundary value space of each surface of the turbine cylinder and rotor under all operating conditions. Rotor speed range The The subscript 2 represents the working fluid temperature and surface heat transfer coefficient at the temperature boundary; the subscript... m This represents the number of corresponding areas where friction occurs inside the cylinder and rotor; The sample consisting of boundary conditions described in step (2) determines the sample size of the boundary condition sample set. Each sample is composed of It consists of elements, among which One element represents the temperature boundary, and one element represents the force boundary; the value range of each element is evenly divided into... First, draw a number from each of the previously undrawn equal-probability regions. Then, randomly draw one number from each of the undrawn equal-probability regions for different elements, and repeat the process. This forms the boundary condition sample set. .

2. The online monitoring method for turbine differential expansion based on strain measurement according to claim 1, characterized in that: The position of the supporting structure stress extracted from the finite element analysis results in step (3) must correspond to the position of the strain gauge in the strain measurement. The interface between the cylinder and the rotor should include the axial end faces of the stationary blades and moving blades at each stage, the front and rear steam seal teeth and the shaft shoulder that are prone to rubbing.

3. The online monitoring method for turbine differential expansion based on strain measurement according to claim 1, characterized in that: The deep neural network described in step (4) is a multi-layer fully connected neural network.

4. The online monitoring method for turbine differential expansion based on strain measurement according to claim 3, characterized in that: The multilayer fully connected neural network consists of... The input layer of each neuron Hidden layers of each neuron and The output layer consists of 10 neurons.

5. The online monitoring method for turbine differential expansion based on strain measurement according to claim 4, characterized in that: The The number of strain gauges attached to the support arm; The number of end faces at the junction of the cylinder and the rotor; The number of hidden layers and the number of neurons in each layer are determined by the number of layers, and should not be less than 64.

6. The online monitoring method for turbine differential expansion based on strain measurement according to claim 1, characterized in that: Step (4) involves dividing the dataset into a training set and a test set in a 7:3 ratio. During network training, the total number of iterations is set to 300 rounds. The optimizer is set to Adam, and the initial learning rate is set to 0.

001. Then, when the number of training steps reaches 100, 180, 230, and 260, the learning rate is reduced to 1 / 5 of the original rate. After the model training is completed, if the model's error prediction accuracy on the test set is less than 98%, the number of hidden layer neurons is modified. The model is retrained until the accuracy meets the requirements, including the total number of iterations, the initial learning rate, and other hyperparameters.

Images