A multi-channel temperature parameter pre-cooling process liquid rocket engine temperature field display method

Abstract

The application provides a multi-channel temperature parameter precooling process liquid rocket engine temperature field display method, solves the technical problem that in the prior art, based on temperature sensors, numerical measurement is performed on key temperature measuring points, and due to temperature parameter measurement through multiple measuring points by a data acquisition system, long-time numerical display is insufficient to intuitively show the change of a structure system when a power system is preparing; the application is based on simulation calculation with high time complexity, cannot achieve real-time calculation, and uses a recurrent neural network to perform feature analysis, completes model training based on gradient descent, finally realizes interpolation calculation of a temperature field in a channel dimension, can effectively improve the resolution of channel dimension data, and further improves the resolution of a temperature field image, and is convenient for subsequent modeling and analysis of a thrust chamber temperature field based on a precooling whole process.

Description

Technical Field

[0001] This invention relates to a method for establishing a temperature field, specifically to a method for displaying the temperature field of a liquid rocket engine during the precooling process using multi-channel temperature parameters. Background Technology

[0002] Liquid oxygen-kerosene rocket engines are rocket propulsion systems based on cryogenic propellants. As a cryogenic propellant, liquid oxygen requires pre-cooling of the engine's oxygen system before startup to ensure the smoothness of the engine startup process. Engine testing and actual flight are characterized by high risk, high cost, high precision, and high value. By utilizing the pre-cooling process before engine testing, the operating status of key sensors and the temperature field change trend of the structural system can be monitored. This enables the monitoring of the engine structural system's status before testing / flight, ensuring the reliability of key sensor measurement parameters.

[0003] As a power system subjected to high temperature, high pressure, strong vibration, and high speed, liquid rocket engines rely heavily on temperature measurement to ensure proper engine operation. Current technologies primarily rely on temperature sensors to measure key temperature points. By monitoring the temperature parameters at these points in real time, the system monitors both sensor status and engine system status. Data acquisition systems measure temperature parameters from multiple points and display the results visually. However, during the preparation phase of the power system, prolonged numerical displays are insufficient to clearly demonstrate changes in the structural system. Summary of the Invention

[0004] The purpose of this invention is to solve the technical problem that existing technologies rely on temperature sensors to perform numerical measurements on key temperature points. Because the data acquisition system measures temperature parameters at multiple points, the long-term numerical display is insufficient to intuitively show the changes in the structural system during the preparation of the power system. Therefore, this invention provides a method for displaying the temperature field of a liquid rocket engine during the precooling process using multi-channel temperature parameters.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for displaying the temperature field of a liquid rocket engine during the precooling process using multi-channel temperature parameters, characterized by the following steps:

[0007] Step 1: Set up d temperature sensors at d measuring points on the rocket engine near the oxygen system to collect temperature data at each measuring point.

[0008] Step 2: Collect temperature data at each measuring point by moving a sliding window to construct a dataset. The sliding window has n sampling times, and the temperature data collected each time is used as input data. At the same time, simulate the temperature field of the rocket engine to obtain two-dimensional temperature field simulation data corresponding to each sampling time. The simulation data includes the temperature values ​​corresponding to each position of the rocket engine and each pre-cooling time.

[0009] Step 3: Construct the recurrent neural network model and loss function;

[0010] Step 4: Input the dataset into the recurrent neural network model, calculate the training error by comparing the output results with the corresponding two-dimensional temperature field simulation data using the loss function, and train the weights in the recurrent neural network model based on the training error to obtain an optimized recurrent neural network model.

[0011] Step 5: Input the temperature data collected from the d temperature sensors into the optimized recurrent neural network model to obtain the channel-dimensional enhanced temperature field numerical matrix;

[0012] Step 6: Draw a temperature field image based on the channel-dimensional enhanced temperature field numerical matrix to complete the display of the temperature field image of the channel-dimensional enhanced precooling process.

[0013] Furthermore, in step 1, the value of d is 3;

[0014] In step 2, the value of n is 10;

[0015] In step 3, the recurrent neural network model includes a 3×10 input module, a 30×10 state layer, and a 30×10 output module;

[0016] The input module, state layer, and output module are connected in sequence.

[0017] Furthermore, the input module is represented as:

[0018]

[0019] in, Indicates input data, This represents the i-th temperature measured by the first temperature sensor. Temperature data at i+9 sampling times, where i represents the starting sampling time; This represents the i-th temperature measured by the second temperature sensor. Temperature data at i+9 sampling times; This represents the i-th temperature measured by the third temperature sensor. Temperature data at sampling time i+9, where i≥1.

[0020] Furthermore, the state layer is represented as follows:

[0021]

[0022] In the formula, This represents the output of the 30×10 state layer. This represents the output feature result of each node in the state layer, where m represents the feature index of the state layer in the measurement point dimension, with a value range of 1, ..., 30; j represents the feature index of the state layer in the sampling time dimension, with a value range of i, ..., i+9; the feature of the column vector at the i-th sampling time of the state layer of the recurrent neural network model is... It can be expressed by the following formula:

[0023] ;

[0024] in, This represents the output of the column vector features of the state layer of the previous recurrent neural network model. When i=1, ; This represents the weight matrix of the input data to the output of the state layer. The weight matrix representing the output of the state layer from the previous state is as follows:

[0025]

[0026] =

[0027] in, … Weight matrix Element; … Weight matrix Element.

[0028] Furthermore, the expression for the output module is:

[0029] ;

[0030] in, This indicates the output result of the output module; This represents the temperature data measured at the p-th sampling time of the o-th channel dimension after channel dimension enhancement of the recurrent neural network model, where the value of o ranges from 1 to 30; and the value of p ranges from i to i+9.

[0031] The output module of the recurrent neural network model outputs the column-oriented output result at the i-th sampling time. This can be expressed by the following formula:

[0032] = ;

[0033] ;

[0034] in, This represents the weight matrix from the state layer to the output module. … Weight matrix Element.

[0035] Further, in step 3, the expression for the loss function is:

[0036] Loss=MSE( ;

[0037] Where Loss represents the training error. This represents the output of the recurrent neural network model. This represents the corresponding two-dimensional temperature field simulation data selected based on the measurement point location and pre-cooling time.

[0038] Further, in step 4, training the weights in the recurrent neural network model based on the training error specifically involves:

[0039] Determine if the following conditions are met: training error < 0.001 or the number of iterations reaches 100,000. If so, the recurrent neural network model corresponding to the current weights is an optimized recurrent neural network model; otherwise, use gradient descent to optimize the weight matrix. Weight matrix Weight matrix Update.

[0040] The beneficial effects of this invention are:

[0041] (1) This invention provides a method for displaying the temperature field of a liquid rocket engine in the precooling process of multi-channel temperature parameters. The method obtains the temperature field numerical matrix with enhanced channel dimension through an optimized recurrent neural network model, thereby obtaining the temperature field image. Existing simulation-based calculations have high time complexity and cannot achieve real-time calculation. However, this invention uses a recurrent neural network model for feature analysis and implements neural network model training based on gradient descent. Finally, it realizes the interpolation calculation of the temperature field in the channel dimension, which can effectively improve the resolution of the channel dimension data, thereby improving the resolution of the temperature field image, which is convenient for subsequent modeling and analysis of the thrust chamber temperature field based on the precooling process.

[0042] (2) This invention provides a method for displaying the temperature field of a liquid rocket engine during the precooling process using multi-channel temperature parameters. By analyzing the temperature parameters based on the temperature measurement results, the precooling process with temperature change trends can be monitored, enabling digital monitoring of the engine test process and digital-based test status monitoring. Attached Figure Description

[0043] Figure 1 This is a temperature data change trend diagram measured by three temperature sensors in an embodiment of a method for displaying the temperature field of a liquid rocket engine during the precooling process of multi-channel temperature parameters according to the present invention.

[0044] Figure 2 This is a schematic diagram of the recurrent neural network model in an embodiment of the method for displaying the temperature field of a liquid rocket engine during the precooling process of multi-channel temperature parameters according to the present invention.

[0045] Figure 3 This is the result of displaying the temperature field based on temperature data measured by three temperature sensors in an embodiment of a method for displaying the temperature field of a liquid rocket engine during the precooling process of a multi-channel temperature parameter according to the present invention.

[0046] Figure 4 This is an image of the temperature field display method for liquid rocket engine temperature field during precooling process of multi-channel temperature parameters according to the present invention. It uses an optimized recurrent neural network model to display the temperature field of the data after interpolation of the measurement point dimensions. Detailed Implementation

[0047] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings and embodiments. 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.

[0048] This invention provides a method for displaying the temperature field of a liquid rocket engine during the precooling process using multi-channel temperature parameters, comprising the following steps:

[0049] Step 1: Set up three temperature sensors at three measuring points on the rocket engine near the oxygen system to collect temperature data at each measuring point.

[0050] Step 2: Collect temperature data at each measuring point by moving a sliding window to construct a dataset. The sliding window has 10 sampling times. Use the temperature data collected each time as input data. At the same time, simulate the temperature field of the rocket engine to obtain the two-dimensional temperature field simulation data corresponding to each sampling time. The simulation data includes the temperature values ​​corresponding to each position of the rocket engine and each pre-cooling time.

[0051] Step 3: Construct a recurrent neural network model and loss function.

[0052] The recurrent neural network model includes a 3×10 input module, a 30×10 state layer, and a 30×10 output module. The input module, state layer, and output layer are connected in sequence, and parameters are calculated based on the recurrent neural network.

[0053] The input module is represented as:

[0054]

[0055] in, Indicates input data, This represents the i-th temperature measured by the first temperature sensor. Temperature data at i+9 sampling times, where i represents the starting sampling time; This represents the i-th temperature measured by the second temperature sensor. Temperature data at i+9 sampling times; This represents the i-th temperature measured by the third temperature sensor. Temperature data at sampling time i+9, where i≥1.

[0056] The state layer is represented as follows:

[0057]

[0058] In the formula, This represents the output of the 30×10 state layer. This represents the output feature result of each node in the state layer. m represents the feature index of the state layer in the measurement point dimension, with a value range of 1, ..., 30; j represents the feature index of the state layer in the sampling time dimension, with a value range of i, ..., i+9.

[0059] Pick ;

[0060] Pick Then the column vector feature of the state layer of the recurrent neural network model at the i-th sampling time is: It can be expressed by the following formula:

[0061] = ;

[0062] ;

[0063] in, This represents the column output of the state layer of the previous recurrent neural network model. When i=1, , This represents the weight matrix of the input data to the output of the state layer. This represents the weight matrix of the previous state's output to the state layer. The elements of the matrix are represented as follows:

[0064]

[0065] =

[0066] in, … Weight matrix Element; … Weight matrix Element.

[0067] The expression for the output module is:

[0068] ;

[0069] in, This indicates the output result of the output module; This represents the temperature data measured at the p-th sampling time of the o-th channel dimension after channel dimension enhancement of the recurrent neural network model, where the value of o ranges from 1 to 30; and the value of p ranges from i to i+9.

[0070] The output module of the recurrent neural network model outputs the column-oriented output result at the i-th sampling time. This can be expressed by the following formula:

[0071] ;

[0072] in, ;

[0073] in, This represents the weight matrix from the state layer to the output layer. … Weight matrix Element.

[0074] like Figure 2 As shown, Figure 2 In this context, t-1, t, and t+1 represent the time sequence of the recurrent neural network model.

[0075] The expression for the loss function is:

[0076] Loss=MSE( ;

[0077] Where Loss represents the training error. This represents the output of the recurrent neural network model. This represents the corresponding two-dimensional temperature field simulation data selected based on the measurement point location and pre-cooling time.

[0078] Step 4: Input the dataset into the recurrent neural network model. Compare the output with the high-resolution two-dimensional temperature field simulation data using the loss function to calculate the training error. Determine if the following conditions are met: training error < 0.001 or the number of iterations reaches 100,000. If met, the recurrent neural network model corresponding to the current weights is an optimized recurrent neural network model; otherwise, use gradient descent to optimize the weight matrix. Weight matrix Weight matrix Update.

[0079] In this embodiment, gradient descent is used to apply weights to the matrix. Weight matrix Weight matrix The update enables interpolation calculation of the temperature field in the channel dimension, effectively improves the resolution of channel dimension data and the resolution of temperature field images, which facilitates subsequent modeling and analysis of the thrust chamber temperature field based on the entire precooling process.

[0080] Step 5: Input the temperature data collected from the three temperature sensors into the optimized recurrent neural network model to obtain a channel-dimensional enhanced temperature field numerical matrix.

[0081] Step 6: Draw a temperature field image based on the channel-dimensional enhanced temperature field numerical matrix to complete the display of the temperature field image of the channel-dimensional enhanced precooling process.

[0082] In this embodiment, three temperature sensors are placed on the rocket engine near the oxygen system, with each temperature sensor serving as a temperature measurement channel. Figure 1 This is a waveform diagram of temperature data obtained from measurements by a 3-channel temperature sensor.

[0083] The temperature data measured by three temperature sensors is used as raw data to display the temperature field. The results are as follows: Figure 3 As shown.

[0084] A 3×10 matrix is ​​formed by taking temperature data from 10 sampling times using a sliding window and used as input data for the recurrent neural network model. A 30×10 matrix of simulated temperature field values ​​is used as labels for training the recurrent neural network model. When the training error is less than 0.001 or the number of iterations is greater than 100,000, the recurrent neural network model training is terminated, and the optimal weight matrix is ​​obtained. , and The recurrent neural network model, i.e. the optimized recurrent neural network model.

[0085] Temperature data measured by three temperature sensors were used to create a 3×10 temperature array using a sliding window. This dataset was then input into an optimized recurrent neural network model. During model application, based on the actual collected temperature data from the three temperature sensors, a 3×10 matrix was formed with every 10 sampling points. The optimized recurrent neural network model then calculated a 30×10 temperature field value. The results are as follows: Figure 4 As shown.

[0086] Will Figure 3 and Figure 4 The comparison shows that, Figure 4 The resolution of the temperature field image is significantly enhanced. It can be seen that this embodiment enhances the signal in the channel dimension through an optimized recurrent neural network model, and obtains a channel-dimensional enhanced temperature field numerical matrix. By drawing a temperature field image using the channel-dimensional enhanced temperature field numerical matrix, it is possible to analyze the temperature signal in the pre-cooling stage, which facilitates subsequent status monitoring based on the image.

[0087] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions within the technical scope disclosed in the present invention should be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for displaying the temperature field of a liquid rocket engine during the precooling process using multi-channel temperature parameters, characterized in that, Includes the following steps: Step 1: Set d temperature sensors at d measurement points on the rocket engine near the oxygen system to collect temperature data at each measurement point. Each temperature sensor serves as a temperature measurement channel. The value of d is 3; Step 2: Collect temperature data at each measuring point by moving a sliding window to construct a dataset. The sliding window has n sampling times, and the temperature data collected each time is used as input data. At the same time, simulate the temperature field of the rocket engine to obtain two-dimensional temperature field simulation data corresponding to each sampling time. The simulation data includes the temperature values ​​corresponding to each position of the rocket engine and each pre-cooling time. The value of n is 10. Step 3: Construct a recurrent neural network model and loss function; the recurrent neural network model includes a 3×10 input module, a 30×10 state layer, and a 30×10 output module; the input module, state layer, and output module are connected in sequence; The input module is represented as: ； in, Indicates input data, This represents the i-th temperature measured by the first temperature sensor. Temperature data at i+9 sampling times, where i represents the starting sampling time; This represents the i-th temperature measured by the second temperature sensor. Temperature data at i+9 sampling times; This represents the i-th temperature measured by the third temperature sensor. Temperature data at sampling time i+9, where i≥1; The state layer is represented as follows: ； In the formula, This represents the output of the 30×10 state layer. This represents the output feature result of each node in the state layer, where m represents the feature index of the state layer in the measurement point dimension, with a value range of 1, ..., 30; j represents the feature index of the state layer in the sampling time dimension, with a value range of i, ..., i+9; the feature of the column vector at the i-th sampling time of the state layer of the recurrent neural network model is... It can be expressed by the following formula: = ； = ； in, This represents the output of the column vector features of the state layer of the previous recurrent neural network model. When i=1, ; This represents the weight matrix of the input data to the output of the state layer. The weight matrix representing the output of the state layer from the previous state is shown below: ； = ； in, … Weight matrix Element; … Weight matrix Element; The expression for the output module is: ； in, This indicates the output result of the output module; This represents the temperature data measured at the p-th sampling time of the o-th channel dimension after channel dimension enhancement of the recurrent neural network model, where the value of o ranges from 1 to 30; and the value of p ranges from i to i+9. The output module of the recurrent neural network model outputs the column-oriented output result at the i-th sampling time. This can be expressed by the following formula: = ； ； in, This represents the weight matrix from the state layer to the output module. … matrix Element; The expression for the loss function is: Loss=MSE( ; Where Loss represents the training error. This represents the output of the recurrent neural network model. This represents the corresponding two-dimensional temperature field simulation data selected based on the measurement point location and precooling time; Step 4: Input the dataset into the recurrent neural network model, calculate the training error by comparing the output results with the corresponding two-dimensional temperature field simulation data using the loss function, and train the weights in the recurrent neural network model based on the training error to obtain an optimized recurrent neural network model. Step 5: Input the temperature data collected from the d temperature sensors into the optimized recurrent neural network model to obtain the channel-dimensional enhanced temperature field numerical matrix; Step 6: Draw a temperature field image based on the channel-dimensional enhanced temperature field numerical matrix to complete the display of the temperature field image of the channel-dimensional enhanced precooling process.

2. The method for displaying the temperature field of a liquid rocket engine during the precooling process using multi-channel temperature parameters according to claim 1, characterized in that, In step 4, training the weights in the recurrent neural network model based on the training error specifically involves: Determine if the following conditions are met: training error < 0.001 or the number of iterations reaches 100,000. If so, the recurrent neural network model corresponding to the current weights is an optimized recurrent neural network model; otherwise, use gradient descent to optimize the weight matrix. Weight matrix Weight matrix Update.

Images