Real-time online monitoring method for weld stress of high-temperature molten salt tank

Abstract

The invention discloses a real-time online monitoring method for weld stress of a high-temperature molten salt tank. The real-time online monitoring method comprises the following steps: monitoring a temperature value of a measuring point around a weld at the bottom of the molten salt tank, an environment temperature, a molten salt liquid level, a tank body material parameter and a tank body geometric parameter in a running process of the molten salt tank in real time; inputting monitoring data into the trained BP neural network model, and outputting a stress value of the key position of the welding seam of the molten salt tank in real time; the training process of the BP neural network model is as follows: judging whether a stress value is greater than a design value, if so, starting a molten salt pump, and adjusting the liquid level or temperature of a molten salt tank until the stress value is less than the design value; and if not, the operation is safe, and real-time monitoring is continued. According to the method disclosed by the invention, the stress change rule of the molten salt tank is learned through the BP neural network, and accurate prediction of the stress of the molten salt tank is realized. Meanwhile, when the data set is constructed, the situation that a large number of experiments which are difficult to carry out are adopted to measure stress data is avoided, the experiment cost is reduced, and meanwhile the accuracy is guaranteed.

Description

Technical Field

[0001] The invention relates to the field of molten salt tank stress monitoring, and in particular to a real-time online monitoring method for weld stress of a high-temperature molten salt tank. Background Art

[0002] Molten salt tanks are important equipment in the fields of energy conversion and energy storage. Molten salt tanks play a key role in solar thermal power generation systems and have many advantages such as efficient energy storage, good power generation stability, environmental protection and energy saving. However, at present, high-temperature molten salt tanks frequently leak molten salt, resulting in system shutdown, complex and time-consuming maintenance, huge losses, and affecting the economic benefits of solar thermal power generation. Molten salt tank leakage is generally prone to occur at the welds between the tank wall and the bottom of the tank. When designing the molten salt tank, the tank stress, especially the stress at the weld, is required to be less than the design stress value. However, the actual operating conditions are complex and it is difficult to fully consider all operating conditions. At the same time, it is impossible to monitor the stress value of the tank body in real time. The existing strategy can only indirectly regulate the molten salt pump through the temperature difference between the tank wall and the bottom of the tank to reduce the temperature gradient. The impact and influence of the molten salt liquid level on the tank body stress are not taken into account. In the molten salt tank system currently in operation, there is no system for real-time monitoring of the molten salt tank stress.

[0003] During the operation of the molten salt tank, changes in temperature and liquid level will have a significant impact on its stress distribution, thereby affecting the safety and service life of the molten salt tank. Figure 1 It is a model diagram of the molten salt tank and its foundation, including tank top 1, insulation layer 2, tank wall 3, tank bottom 4, fine sand cushion layer 5, sand and gravel cushion layer 6, refractory brick ring wall 7, steel plate ring 8, and expanded clay layer 9; 10 is a C20 concrete layer, which needs to ensure that the maximum temperature is less than 70°C; 11 is soil; 12 is a ventilation pipe to prevent the temperature of C20 concrete from being too high; 13 is a temperature measuring point 1 in the C20 concrete, located below the ventilation pipe 12; 14 is a number of temperature measuring points 2 arranged between the molten salt tank and the insulation layer.

[0004] The welds at the bottom of the molten salt tank, especially the large-angle welds, have a series of remarkable characteristics, which are mainly derived from their unique structural position and functional requirements. The large-angle weld is the connection between the tank wall and the bottom plate. The structural discontinuity at this position is significant, so it is a key area of ​​stress concentration. Because it is located at the bottom of the tank, the large-angle weld is directly affected by the static pressure of the liquid column, which is the main source of its force characteristics. Under the action of the static pressure of the liquid column, seismic load and mechanical load, high stress areas are prone to appear in the large-angle weld area. The welds at the bottom of the molten salt tank, especially the large-angle welds, have special structural positions and complex stress distribution. Generally, strain gauges can be pasted at the welds, and stress and strain can be tested by experimental methods, but the strain gauges may fall off and fail when the tank expands. This method is only suitable for short-term testing and not for long-term operation monitoring.

[0005] Therefore, developing a new technology that can accurately predict the stress of molten salt tanks in real time has important practical value. Summary of the invention

[0006] In order to solve the above technical problems, the present invention provides a real-time online monitoring method for the stress of the weld of a high-temperature molten salt tank, so as to achieve the purpose of accurately predicting the stress of the molten salt tank.

[0007] To achieve the above object, the technical solution of the present invention is as follows:

[0008] A real-time online monitoring method for high-temperature molten salt tank weld stress comprises the following steps:

[0009] Step 1: Real-time monitoring of the temperature values ​​of the measuring points around the weld at the bottom of the molten salt tank, the ambient temperature, the molten salt level, the tank material parameters and the tank geometric parameters during the operation of the molten salt tank;

[0010] Step 2: Input the monitoring data into the trained BP neural network model, and output the stress value of the key position of the weld of the molten salt tank in real time; the training process of the BP neural network model is as follows:

[0011] (1) Collect the temperature values ​​of the measuring points around the weld of the molten salt tank bottom, the ambient temperature, the molten salt level, the tank material parameters and the tank geometric parameters during normal operation;

[0012] (2) Establish a temperature field calculation model for the molten salt tank using finite element software, input the collected data into the temperature field calculation model, and output the overall temperature distribution of the molten salt tank;

[0013] (3) Establish a stress field calculation model for the molten salt tank using finite element software, input the overall temperature distribution of the molten salt tank, the tank material parameters, and the tank geometry parameters into the stress field calculation model, obtain the stress distribution of the molten salt tank, and extract the stress value at the weld;

[0014] (4) Correcting the stress field calculation model according to the stress value at the weld tested by the strain gauge experiment to obtain a corrected stress field calculation model;

[0015] (5) Repeat steps (1) and (2), input the obtained overall temperature distribution of the molten salt tank, the tank material parameters, and the tank geometric parameters into the corrected stress field calculation model, and obtain the corrected stress value at the weld;

[0016] (6) Establishing a one-to-one correspondence between the data collected in step (1) and the corrected stress value at the weld, constructing a data set, performing minimum-maximum normalization processing on the data set, scaling the data to the [0,1] interval, and then inputting it into the BP neural network model for training;

[0017] Step 3, determine whether the stress value is greater than the design value. If so, start the molten salt pump and adjust the liquid level or temperature of the molten salt tank until the stress value is less than the design value; if not, it indicates safe operation and continues real-time monitoring.

[0018] In the above scheme, the tank material parameters include thermal conductivity, density, elastic modulus, Poisson's ratio, yield strength, and thermal expansion coefficient.

[0019] In the above scheme, the tank body geometric parameters include tank diameter, tank bottom wall thickness, tank wall thickness, and tank cross-sectional area.

[0020] In the above scheme, there are two methods for correcting the stress field calculation model. One method is to modify the friction coefficient, contact stiffness factor, and mesh accuracy of the bottom and foundation of the molten salt tank, recalculate, and compare the experimental value and the calculated value again until the allowable error is met; the other method is to perform polynomial fitting based on the experimental value and the calculated value, add, subtract, multiply, and divide the coefficients according to the mapping relationship, map the calculated value to the experimental value, and finally make the calculation result of the calculation model equivalent to the experimental value and within the allowable error.

[0021] In the above scheme, the ReLU activation function is used in the BP neural network to increase the nonlinearity of the network. The formula is as follows:

[0022] f(x)=max(0,x)

[0023] Among them, x is the output of the BP layer, and f(x) is the output of the activation function.

[0024] In the above scheme, when training the BP neural network, the back propagation algorithm is used to update the weights and biases in the network. The back propagation algorithm backpropagates the gradient of the loss function to each layer of the network through the chain rule, and updates the weights and biases accordingly; the mean square error loss function is used, and its formula is:

[0025]

[0026] Among them, y i is the true value, is the predicted value and m is the number of samples.

[0027] Through the above technical solution, the real-time online monitoring method of the stress of the high-temperature molten salt tank weld provided by the present invention has the following characteristics:

[0028] Beneficial effects:

[0029] The present invention adopts a BP neural network model, which can utilize the temperature values ​​of the measuring points around the bottom weld of the molten salt tank, the ambient temperature, the molten salt liquid level, the tank material parameters and the tank geometric parameter data to learn the stress change law of the molten salt tank through the BP neural network, thereby realizing accurate prediction of the stress of the molten salt tank.

[0030] In the process of constructing the data set of the present invention, a small amount of experimental values ​​are combined to correct the finite element calculation model, and then a large amount of stress data at the weld is obtained through finite element calculation, thereby avoiding the use of a large number of difficult experiments to measure stress data, reducing the experimental cost while ensuring accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below.

[0032] Figure 1 A simplified model diagram of a molten salt tank and a tank foundation disclosed in an embodiment of the present invention;

[0033] Figure 2 This is a schematic diagram of the BP neural network calculation process for the molten salt tank stress;

[0034] Figure 3 It is a schematic diagram of the training process of the BP neural network model;

[0035] Figure 4 Schematic diagram of BP neural network model.

[0036] In the figure, 1. tank top; 2. insulation layer; 3. tank wall; 4. tank bottom; 5. fine sand cushion layer; 6. sand and gravel cushion layer; 7. refractory brick ring wall; 8. steel plate ring; 9. expanded clay layer; 10. C20 concrete layer; 11. soil; 12. ventilation pipe; 13. temperature measuring point one; 14. temperature measuring point two. DETAILED DESCRIPTION

[0037] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0038] The present invention provides a real-time online monitoring method for the stress of the weld of a high-temperature molten salt tank. Figure 2 As shown, the following steps are included:

[0039] Step 1: Real-time monitoring of the temperature values ​​of the measuring points around the weld at the bottom of the molten salt tank, the ambient temperature, the molten salt level, the tank material parameters and the tank geometric parameters during the operation of the molten salt tank.

[0040] The tank material parameters include thermal conductivity, density, elastic modulus, Poisson's ratio, yield strength, and thermal expansion coefficient. The tank geometric parameters include tank diameter, tank bottom wall thickness, tank wall thickness, and tank cross-sectional area.

[0041] Step 2: Input the monitoring data into the trained BP neural network model and output the stress values ​​of key positions of the molten salt tank weld in real time.

[0042] BP neural network is a multi-layer feedforward network trained by error back propagation algorithm, which has powerful nonlinear mapping ability. The core operations of BP neural network include initializing network parameters, forward propagation, calculating errors, back propagation, iterative training and preventing overfitting steps.

[0043] like Figure 3 As shown, the training process of the BP neural network model is as follows:

[0044] First, build the dataset:

[0045] (1) Collect the temperature values ​​of the measuring points around the weld of the molten salt tank bottom, the ambient temperature, the molten salt level, the tank material parameters and the tank geometric parameters during normal operation;

[0046] (2) Establish a temperature field calculation model for the molten salt tank using finite element software, input the collected data into the temperature field calculation model, and output the overall temperature distribution of the molten salt tank;

[0047] (3) Establish a stress field calculation model for the molten salt tank using finite element software, input the overall temperature distribution of the molten salt tank, the tank material parameters, and the tank geometry parameters into the stress field calculation model, obtain the stress distribution of the molten salt tank, and extract the stress value at the weld;

[0048] (4) Correcting the stress field calculation model according to the stress value at the weld tested by the strain gauge experiment to obtain a corrected stress field calculation model;

[0049] There are two methods for correcting the stress field calculation model. One method is to modify the friction coefficient, contact stiffness factor, and mesh accuracy of the bottom and foundation of the molten salt tank, recalculate, and compare the experimental value and the calculated value again until the allowable error is met. The other method is to perform polynomial fitting based on the experimental value and the calculated value, add, subtract, multiply, and divide the coefficients according to the mapping relationship, map the calculated value to the experimental value, and finally make the calculation result of the calculation model equivalent to the experimental value and within the allowable error.

[0050] (5) Repeat steps (1) and (2), input the obtained overall temperature distribution of the molten salt tank, the tank material parameters, and the tank geometric parameters into the corrected stress field calculation model, and obtain the corrected stress value at the weld;

[0051] (6) Establish a one-to-one correspondence between the data collected in step (1) and the corrected stress value at the weld, construct a data set, perform minimum-maximum normalization on the data set, scale the data to the [0,1] interval, and then input it into the BP neural network model for training.

[0052] When training a BP neural network, the back propagation algorithm is used to update the weights and biases in the network. Back propagation propagates the gradient of the loss function back to each layer of the network through the chain rule, and updates the weights and biases accordingly.

[0053] The loss function is used to measure the difference between the model prediction and the true label. The present invention adopts the mean square error loss function. Its formula is:

[0054]

[0055] Among them, y i is the true value, is the predicted value and m is the number of samples.

[0056] The activation function is usually applied after the BP layer to increase the nonlinearity of the network. This patent uses the ReLU activation function, and its formula is:

[0057] f(x)=max(0,x)

[0058] Among them, x is the output of the BP layer, and f(x) is the output of the activation function.

[0059] like Figure 4 As shown, the present invention adopts a BP neural network including an input layer (6 nodes), several hidden layers (several nodes) and an output layer (1 node). The data set comes from the combination of finite element calculation results and experimental values. Finite element calculation can provide a large number of data sets with high data quality and uniform data distribution. The input layer receives temperature values, molten salt liquid level, tank diameter, and wall thickness parameters, the hidden layer extracts and learns features through multiple neurons, and the output layer outputs the predicted molten salt tank stress value.

[0060] The model is trained using a data set, including several sets of molten salt tank stress data at different temperatures, liquid levels, tank diameters, and wall thicknesses. The back propagation algorithm and gradient descent method are used to optimize the parameters of the BP neural network during the training process. By continuously adjusting the parameters of the BP neural network, the model can learn the complex relationship between temperature, liquid level, tank diameter, wall thickness, and stress.

[0061] This BP neural network model takes into account the diameter parameters of the molten salt tank, has a wide range of applications, and is suitable for molten salt tanks of different diameters and wall thicknesses. The tank temperature and its gradient and the molten salt level are comprehensively considered, and the calculation is more accurate. In practical applications, the tank diameter and wall thickness are fixed values ​​and can be input in advance. By collecting the temperature and liquid level data of the molten salt tank in real time and inputting them into the trained BP neural network model, the predicted stress value can be obtained.

[0062] Step 3, determine whether the stress value is greater than the design value. If so, start the molten salt pump and adjust the liquid level or temperature of the molten salt tank until the stress value is less than the design value; if not, it indicates safe operation and continues real-time monitoring.

[0063] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A real-time online monitoring method for high-temperature molten salt tank weld stress, characterized in that: The steps include: Step 1: Real-time monitoring of the temperature values ​​of the measuring points around the weld at the bottom of the molten salt tank, the ambient temperature, the molten salt level, the tank material parameters and the tank geometric parameters during the operation of the molten salt tank; Step 2: Input the monitoring data into the trained BP neural network model, and output the stress value of the key position of the weld of the molten salt tank in real time; the training process of the BP neural network model is as follows: (1) Collect the temperature values ​​of the measuring points around the weld of the molten salt tank bottom, the ambient temperature, the molten salt level, the tank material parameters and the tank geometric parameters during normal operation; (2) Establish a temperature field calculation model for the molten salt tank using finite element software, input the collected data into the temperature field calculation model, and output the overall temperature distribution of the molten salt tank; (3) Establish a stress field calculation model for the molten salt tank using finite element software, input the overall temperature distribution of the molten salt tank, the tank material parameters, and the tank geometry parameters into the stress field calculation model, obtain the stress distribution of the molten salt tank, and extract the stress value at the weld; (4) Correcting the stress field calculation model according to the stress value at the weld tested by the strain gauge experiment to obtain a corrected stress field calculation model; (5) Repeat steps (1) and (2), input the obtained overall temperature distribution of the molten salt tank, the tank material parameters, and the tank geometric parameters into the corrected stress field calculation model, and obtain the corrected stress value at the weld; (6) Establishing a one-to-one correspondence between the data collected in step (1) and the corrected stress value at the weld, constructing a data set, performing minimum-maximum normalization processing on the data set, scaling the data to the [0,1] interval, and then inputting it into the BP neural network model for training; Step 3, determine whether the stress value is greater than the design value. If so, start the molten salt pump and adjust the liquid level or temperature of the molten salt tank until the stress value is less than the design value; if not, it indicates safe operation and continues real-time monitoring.

2. A real-time online monitoring method for high-temperature molten salt tank weld stress according to claim 1, characterized in that: The tank material parameters include thermal conductivity, density, elastic modulus, Poisson's ratio, yield strength, and thermal expansion coefficient.

3. A real-time online monitoring method for high-temperature molten salt tank weld stress according to claim 1, characterized in that: The tank body geometric parameters include tank diameter, tank bottom wall thickness, tank wall thickness, and tank cross-sectional area.

4. A real-time online monitoring method for high-temperature molten salt tank weld stress according to claim 1, characterized in that: There are two methods for correcting the stress field calculation model. One method is to modify the friction coefficient, contact stiffness factor, and mesh accuracy of the bottom and foundation of the molten salt tank, recalculate, and compare the experimental value and the calculated value again until the allowable error is met. The other method is to perform polynomial fitting based on the experimental value and the calculated value, add, subtract, multiply, and divide the coefficients according to the mapping relationship, map the calculated value to the experimental value, and finally make the calculation result of the calculation model equivalent to the experimental value and within the allowable error.

5. The real-time online monitoring method for high-temperature molten salt tank weld stress according to claim 1 is characterized in that: The ReLU activation function is used in the BP neural network to increase the nonlinearity of the network. The formula is as follows: f(x)=max(0,x) Among them, x is the output of the BP layer, and f(x) is the output of the activation function.

6. A real-time online monitoring method for high-temperature molten salt tank weld stress according to claim 1, characterized in that: When training a BP neural network, the back propagation algorithm is used to update the weights and biases in the network. The back propagation algorithm backpropagates the gradient of the loss function to each layer of the network through the chain rule, and updates the weights and biases accordingly; the mean square error loss function is used, and its formula is: Among them, y i is the true value, is the predicted value and m is the number of samples.

Images