Aero-engine exhaust temperature trend prediction method inspired by climbing plants

Abstract

The invention provides an aero-engine exhaust temperature trend prediction method inspired by climbing plants. The problem that a neural network cannot serve a specific research object due to the fact that the neural network is hard to carry when the aero-engine exhaust temperature is predicted based on the neural network is solved. Analyzing the characteristics of the data of the engine gas path parameter sensor; the sensor data is reasonably classified based on the thought of division and treatment; improving time sequence features of learning data of the long-short-term memory neural network; inspired by a growth mode that the climbing plant climbs upwards under the constraint and guidance of a support, the exhaust temperature is regarded as the climbing plant, and the numerical value of the climbing plant in a period of time in the future is predicted based on the support such as a throttle lever instruction. The method can accurately predict the exhaust temperature trend of the aero-engine.

Description

Technical Field

[0001] The present invention proposes a method for predicting the exhaust gas temperature trend of an aero-engine inspired by climbing plants, belonging to the field of intelligent aero-engine health management. Background Technique

[0002] The exhaust gas temperature is an important measure of the health status of an aero-gas turbine engine. Monitoring and predicting the exhaust gas temperature is an important part of aero-engine prognostics and health management technology, and plays a very important role in the safety and reliability of the entire aircraft.

[0003] Predicting the exhaust gas temperature trend of an aero-engine is a complex problem. An aero-engine is a highly complex and precise thermal machine, and there are coupling effects between various components. Since an aero-engine contains multiple types of components, there are a wide variety of state parameters. However, only some of the parameters can be measured, so the behavior of the system can only be observed from a limited perspective. It is quite difficult to use the limited available data to characterize the complex operating state of the engine. At the same time, as the service time increases, the engine performance deteriorates, and the collected state parameters will change to a certain extent. Different working environments of the aero-engine will also result in different parameters. In addition, the engine sensor data is easily contaminated by noise, affecting the assessment of the engine working state.

[0004] The current mainstream algorithms for predicting the exhaust gas temperature trend of aero-engines can be roughly divided into three categories: traditional exhaust gas temperature trend prediction methods based on historical information, exhaust gas temperature trend prediction methods based on physical models, and exhaust gas temperature trend prediction methods based on data-driven. The data-driven method can avoid establishing complex mathematical models, which has attracted wide attention. The patent with the publication number CN113297680A discloses a method for analyzing the performance trend of a small bypass ratio aero-gas turbine engine, obtaining the engine exhaust gas temperature data within a predetermined operating time, and constructing an autoregressive moving average combined model based on the processed data. This method evaluates the exhaust gas temperature change in combination with the operating data and predicts future performance changes. However, this method assumes that the state of the engine remains unchanged or has a certain attenuation law in a future period of time. In fact, when the engine is performing a flight mission, the flight state is not always the same or always follows a certain pattern. The patent with the publication number CN117168649A discloses a method for predicting the exhaust gas temperature of an aero-engine based on a graph neural network, dividing the graph nodes and edges based on the prior knowledge of the aero-engine, constructing the node feature vectors of the sensor monitoring data, and establishing a graph neural network model. This method uses a relational graph attention network to update the node features, fuses the component space information, and improves the accuracy of exhaust gas temperature prediction. However, this method uses a neural network to learn the relationship between the exhaust gas temperature and other gas path parameters and cannot predict the trend of the exhaust gas temperature in the future for a period of time. In addition, when using a neural network to solve related problems, either mechanically apply it without making full use of the capabilities of the neural network and being unable to serve the specific research object, or overshadow the main issue and make the research object data become a verification tool for the neural network instead. The aero-engine exhaust gas temperature trend prediction method inspired by climbing plants proposed in the present invention can effectively solve these problems. Summary of the Invention

[0005] Aiming at the problems existing in the existing data-driven methods, the present invention proposes an aero-engine exhaust gas temperature trend prediction method inspired by climbing plants to realize the prediction of the exhaust gas temperature trend of the aero-engine.

[0006] To achieve the above object, the concept and technical solution of the present invention are realized as follows:

[0007] The basic concept of the present invention is to analyze the characteristics of the data of the aero-engine gas path parameter sensors; based on the idea of divide and conquer, the sensor data is reasonably classified to more effectively process and analyze different types of data; an improved long short-term memory neural network is proposed to learn the time series characteristics of the data, so as to realize the prediction of the exhaust gas temperature; the exhaust gas temperature trend prediction is compared to the growth of climbing plants. Climbing plants continuously grow upward by attaching to a support. Similarly, key parameters such as the throttle lever command are used as supports to achieve the short-term trend prediction of the exhaust gas temperature. This method can not only effectively extract the time series characteristics in the data, but also improve the accuracy and reliability of the prediction.

[0008] Based on the above basic concept, the technical solution proposed by the present invention is a method for predicting the exhaust gas temperature trend of an aero-engine inspired by climbing plants, including the following steps:

[0009] Step 1: Analyze the characteristics of the data of the aero-engine gas path parameter sensors;

[0010] Step 2: Based on the idea of divide and conquer, reasonably classify the sensor data;

[0011] Step 3: Propose an improved long short-term memory neural network to learn the time series characteristics of the data;

[0012] Step 4: Inspired by the growth mode of climbing plants, regard the exhaust gas temperature as a climbing plant. Climbing plants grow upward by attaching to a support object. Similarly, the neural network realizes the short-term trend prediction of the exhaust gas temperature based on the planned values of parameters such as altitude and throttle lever command.

[0013] Further, in the process of analyzing the characteristics of the data of the aero-engine gas path parameter sensors in Step 1, it includes 4 state parameters and 14 sensor parameters, specifically: the state parameters are altitude alt, flight Mach number Mach, throttle valve angle TRA, and inlet total temperature T2; the sensor parameters are fuel flow Wf, physical fan speed Nf, physical core speed Nc, low-pressure compressor outlet total temperature T24, high-pressure compressor outlet total temperature T30, high-pressure turbine outlet total temperature T48, low-pressure turbine outlet total temperature T50, bypass duct total pressure P15, fan inlet total pressure P2, fan outlet total pressure P21, low-pressure compressor outlet total pressure P24, high-pressure compressor outlet static pressure Ps30, combustor outlet total pressure P40, and low-pressure turbine outlet total pressure P50.

[0014] The engine performs flight missions including climbing, cruising, and descending within the allowable flight envelope, etc. Different flight missions, such as low-altitude low-speed short-range flights and high-altitude long-range flights, correspond to different state values.

[0015] Further, in step 2, based on the divide-and-conquer idea, the sensor data is reasonably classified. An aeroengine has multiple flight conditions, which increases the difficulty of predicting the exhaust gas temperature. Therefore, the exhaust gas temperature prediction problem is divided into multiple sub-problems according to the working conditions. The execution of flight missions by an aeroengine can be divided into three states: climb, cruise, and descent, and the parameter data of altitude can be used for division.

[0016] Further, in step 3, an improved long short-term memory neural network is proposed to learn the time series features of the data. The long short-term memory neural network deletes or adds information through a structure called a gate. The long short-term memory neural network has three gates to protect and control information, namely the forget gate, the input gate, and the output gate. The forget gate is responsible for determining how much of the previous cell state is retained to the current cell state, the input gate is responsible for determining how much of the current input is retained to the current cell state, and the output gate is responsible for determining how much of the current cell state is output. The formulas are as follows:

[0017] f t = σ(W f h t-1 + U f x t + b f )

[0018] i t = σ(W i h t-1 + U i x t + b i )

[0019] a t = tanh(W a h t-1 + U a x t + b a )

[0020] o t = σ(W o h t-1 + U o x t + b o )

[0021] In the formula, x t represents the input at time t, and h t represents the hidden layer state at time t. f, i, o, and a represent the forget gate, the input gate, the output gate, and the feature extraction process respectively, tanh represents the hyperbolic tangent function, and σ represents the activation function Sigmoid. W represents the weight of h, U represents the weight of x, and b represents the bias.

[0022] The results of the forget gate and the input gate calculations act on c t-1 , which constitutes the cell state c at time step t . It is expressed by the formula:

[0023] c t = c t-1 ⊙ f t + i t ⊙ a t

[0024] where ⊙ is the Hadamard product.

[0025] Finally, the state h of the hidden layer at time step t t is obtained from the output gate o t and the cell state c at the current time step t as follows:

[0026] h t = o t ⊙ tanh(c t )

[0027] The bidirectional long short-term memory neural network consists of two long short-term memory neural networks. The input sequence is input into these two long short-term memory neural networks in forward and reverse order respectively for feature extraction, and the two output vectors are concatenated to form the final feature. The bidirectional long short-term memory neural network can better capture the feature change relationship of the sequence and more accurately predict the next state of the sequence. The long short-term memory neural network with a convolutional kernel extracts the features of the data in the spatio-temporal vicinity by using a convolutional structure in each cell.

[0028] Based on the bidirectional long short-term memory neural network and the long short-term memory neural network with a convolutional kernel, a long short-term memory neural network with a convolutional kernel is proposed to better capture the features of time series data. This network combines the advantages of both, enabling the cell to obtain both past and future information and utilize the convolutional structure to extract the features of the data in the spatio-temporal vicinity.

[0029] Furthermore, inspired by the growth pattern of climbing plants in step 4, the exhaust temperature is regarded as a climbing plant. Climbing plants grow upward by attaching to a supporting object. Similarly, the neural network realizes the short-term trend prediction of the exhaust temperature based on the planned values of parameters such as altitude and throttle lever command. The specific steps are as follows:

[0030] Step 4.1: After obtaining the sensor data, preprocess the data;

[0031] Step 4.2: When the engine is performing a flight mission, the flight instructions for a short period in the future should be known numbers because, under normal circumstances, the operation instructions of the operator are planned and clear. For example, if the engine is about to perform a descent mission next, then the altitude of the engine drops, and the throttle lever instruction changes according to the fuel plan. The throttle lever instruction and altitude are like supports, restricting and guiding the future trend of the exhaust gas temperature. The planned instructions and the sensor data obtained currently are used as the input of the neural network together;

[0032] Step 4.3: Utilize the improved long short-term memory neural network to extract the time dependence of features, and then use the fully connected layer and activation function to achieve the prediction of the exhaust gas temperature trend.

[0033] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0034] In the present invention, considering the working environment and mission execution of the engine, the characteristics of the engine's gas path parameters are analyzed. The prediction method proposed based on these characteristics can effectively serve the specific problems of the engine and realize the local application of methods such as neural networks. Based on the divide-and-conquer idea and the improved long short-term memory neural network, the exhaust gas temperature is predicted. Divide-and-conquer can simplify complex problems and reduce the burden on the model; the units of the improved long short-term memory neural network can obtain both past and future information and can also utilize the convolutional structure to extract the features of data in the vicinity in space-time. Inspired by the growth mode of climbing plants, climbing plants rely on support objects to grow upward, and the neural network realizes the short-term trend prediction of the exhaust gas temperature based on the planned values of parameters such as altitude and throttle lever instruction. Description of the Drawings

[0035] Figure 1 is the flowchart of the method for predicting the exhaust gas temperature trend of an aeroengine inspired by climbing plants provided by the present invention;

[0036] Figure 2 is the schematic structural diagram of an aero-turbofan engine provided by the present invention;

[0037] Figure 3 is the schematic diagram of the trajectory of the state parameters of a single flight cycle of the engine provided by the present invention;

[0038] Figure 4 is the schematic diagram of the exhaust gas temperature trend prediction based on divide-and-conquer provided by the present invention;

[0039] Figure 5 is the schematic diagram of the altitude trajectory of a single flight cycle for verifying the engine provided by the present invention;

[0040] Figure 6 is the schematic diagram of the exhaust gas temperature trend prediction in the climbing state provided by the present invention;

[0041] Figure 7 It is a schematic diagram of the exhaust gas temperature trend prediction under the cruise state provided by the present invention;

[0042] Figure 8 It is a schematic diagram of the exhaust gas temperature trend prediction under the descending state provided by the present invention. Specific embodiments

[0043] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0044] Please refer to Figure 1 , which shows a flowchart of a method for predicting the exhaust gas temperature trend of an aero-engine inspired by climbing plants provided by the present invention, specifically including the following steps:

[0045] Step 1: The process of analyzing the data characteristics of the aero-engine gas path parameters sensor, including 4 state parameters and 14 sensor parameters, specifically: the state parameters are altitude alt, flight Mach number Mach, throttle valve angle TRA, and fan inlet total temperature T2; the sensor parameters are fuel flow Wf, physical fan speed Nf, physical core speed Nc, low-pressure compressor outlet total temperature T24, high-pressure compressor outlet total temperature T30, high-pressure turbine outlet total temperature T48, low-pressure turbine outlet total temperature T50, outer duct total pressure P15, fan inlet total pressure P2, fan outlet total pressure P21, low-pressure compressor outlet total pressure P24, high-pressure compressor outlet static pressure Ps30, combustor outlet total pressure P40, and low-pressure turbine outlet total pressure P50.

[0046] Please refer to Figure 2 , which shows the structure of a typical turbofan engine. The main components of the engine are marked in the figure, which are fan (Fan), low-pressure compressor (LPC), high-pressure compressor (HPC), combustor (Combustor), high-pressure turbine (HPT), low-pressure turbine (HPC), and nozzle (Nozzle). In addition, Wf is the fuel flow, Nf is the fan speed, and Nc is the core speed.

[0047] Please refer to Figure 3 , which shows the state parameter trajectory of a single flight cycle of the engine. The engine's flight mission includes climbing, cruising, and descending within the allowable flight envelope, etc. Different flight missions, such as low-altitude low-speed short-range flights and high-altitude long-range flights, correspond to different state values.

[0048] Step 2: Based on the divide-and-conquer idea, the sensor data is reasonably classified, so that the exhaust gas temperature prediction problem is divided into multiple sub-problems according to different working conditions. Please refer to Figure 4 , the flight missions of aero-engines can be divided into three states: climb, cruise, and descent, and can be divided using the parameter of altitude.

[0049] Step 3: Use the improved long short-term memory neural network to learn the time series features of the data. The improved long short-term memory neural network uses 128 units. Set the learning rate to 0.001, and use the Adam optimizer. Use the Xavier initializer to initialize the weights of all networks. The training period is 500 epochs. The number of samples used in each iteration during the training process is set to 64. To prevent the model from overfitting, use the L2 regularization and DropOut strategies, set the regularization factor to 0.0001, and the packet loss rate to 0.2.

[0050] Step 4: Compare the exhaust gas temperature trend prediction to the growth of a climbing plant. A climbing plant grows upwards by clinging to a support object. Similarly, use key parameters such as the throttle lever command as the support to achieve the short-term trend prediction of the exhaust gas temperature. The evaluation index of this example is the root mean square error (RMSE), and the calculation formula is:

[0051]

[0052] In the formula, n represents the number of samples, Y i pre represents the predicted value of the i-th sample, and Y i represents the true value of the i-th sample.

[0053] The specific steps of the exhaust gas temperature trend prediction of an aero-engine inspired by climbing plants are as follows:

[0054] Step 4.1: Preprocess the sensor data. Please refer to Figure 5 , which shows the flight altitude trajectory of a flight cycle. The total duration of this flight cycle is 7.06 hours, of which the climb state accounts for 2.42 hours, the cruise state accounts for 2.47 hours, and the descent state accounts for 2.17 hours. Select one section of data for each of the three states of engine climb, cruise, and descent. Set the time length for predicting the exhaust gas temperature trend to 1 hour.

[0055] To eliminate the influence of the dimension between sensor data features, select the maximum-minimum normalization processing method, and the formula is:

[0056]

[0057] In the formula, X represents the sensor data, X min and X maxFor the minimum and maximum values of the corresponding sensor data;

[0058] Step 4.2: When the engine is performing a flight mission, the flight instructions in the short-term future are usually known. This is because under normal circumstances, the operation instructions of the operator are planned and clearly formulated. For example, when the engine is preparing to perform a descent mission, it is expected that the altitude will decrease, and at the same time, the throttle lever instruction will be adjusted accordingly according to the fuel plan. Combine these planned instructions with the latest acquired sensor data as the input to the neural network;

[0059] Step 4.3: Import the processed data into the neural network, use the improved long short-term memory neural network to extract the time dependence of features, and then use the fully connected layer and activation function to achieve prediction.

[0060] Please refer to Figures 6 - 8 , which respectively show the predicted results of the exhaust gas temperature trend of the engine in the climb, cruise and descent states. The RMSE of the predicted results of the exhaust gas temperature trend in each state is shown in Table 1. The predicted data curve can better follow the real data curve, indicating that the proposed method can effectively predict the trend of the exhaust gas temperature within the next 1 hour.

[0061] Table 1 Predicted results of the exhaust gas temperature trend in each state

[0062] Status Ascending Cruising Descending RMSE 0.61 0.58 0.79

[0063] The present invention is not limited to the above embodiments. Based on the technical solutions disclosed in the present invention, those skilled in the art can make some simple modifications, equivalent changes and modifications to some of the technical features without creative labor, which all fall within the scope of the technical solutions of the present invention.

Claims

1. A method for predicting the exhaust temperature trend of an aero-engine inspired by climbing plants, characterized in that: The specific steps include: Step 1: Analyze the characteristics of aircraft engine gas path parameter sensor data; Step 2: Classify sensor data reasonably based on the idea of ​​divide and conquer; Step 3: Propose time series features to improve the LSTM neural network learning data; Step 4: Inspired by the way climbing plants grow, treat the exhaust temperature as a climbing plant and predict its future trend based on "supports" such as throttle lever commands.

2. The method for predicting the exhaust temperature trend of an aircraft engine inspired by climbing plants according to claim 1, characterized in that: The step 1 analyzes the characteristics of the air path parameter sensor data of the aircraft engine. The flight mission of the engine includes climbing, cruising and descending within the allowable flight envelope. There are four state description variables: altitude, flight Mach number, throttle lever command and inlet total temperature. Different flight missions (such as low-altitude, low-speed, short-range or high-altitude, long-range flight) correspond to different state values.

3. The method for predicting the exhaust temperature trend of an aircraft engine inspired by climbing plants according to claim 1, characterized in that: The step 2 classifies the sensor data reasonably based on the divide and conquer idea. Aircraft engines have multiple flight conditions, which increases the difficulty of exhaust temperature trend prediction. Therefore, the exhaust temperature trend prediction problem is divided into multiple sub-problems according to the operating conditions. The flight mission can be divided into three states: climbing, cruising and descending, and divided by altitude parameters.

4. The method for predicting the exhaust temperature trend of an aircraft engine inspired by climbing plants according to claim 1, characterized in that: The step 3 proposes to improve the time series characteristics of the long short-term memory neural network learning data. Based on the long short-term memory neural network with convolution kernel and the bidirectional long short-term memory neural network, a bidirectional long short-term memory neural network with convolution kernel is proposed to better capture the characteristics of time series data. This network combines the advantages of both, so that the unit can obtain information about the past and the future, and can use the convolution structure to extract the characteristics of the data in the space-time vicinity.

5. The method for predicting the trend of aircraft engine exhaust temperature inspired by climbing plants according to claim 1, characterized in that: Inspired by the growth pattern of climbing plants, step 4 regards the exhaust temperature as a climbing plant and predicts its future trend based on supports such as throttle lever instructions. Climbing plants can grow better with the support, which constrains and guides their direction. When the engine is performing a flight mission, the future short-term flight instructions are usually known. The state parameters such as throttle lever instructions and altitude are like supports, which constrain and guide the neural network to achieve accurate prediction of the exhaust temperature trend.

Images