A method for failure trend prediction and health assessment of liquid rocket engines

Abstract

This invention discloses a method for predicting the failure trend and assessing the health of a liquid rocket engine. The method includes: defining a health index and health evaluation level for the liquid rocket engine based on a fuzzy membership method; predicting the engine's thrust and health using an LSTM network; and comprehensively assessing the engine's failure development trend. This invention utilizes fuzzy evaluation to design and calculate health indices for engine thrust and state parameters reflecting the engine's overall performance, and classifies health degradation levels. Based on the ReliefF algorithm, it selects input feature parameters for failure trend prediction, and uses an LSTM network to predict engine thrust and state health indices. It comprehensively assesses the engine's propulsion performance and safety status under failure development. This method has advantages in long-term series prediction, solves the problem of quantitatively assessing the comprehensive operating status of the engine under multiple factors affecting failure development, and can obtain timely information on engine failure development to avoid non-fatal failures affecting rocket flight.

Description

Technical Field

[0001] This invention belongs to the field of liquid rocket engine fault simulation, specifically relating to a method for predicting the fault trend and assessing the health of liquid rocket engines. Background Technology

[0002] Liquid rocket engines operate under extreme conditions of high temperature, high pressure, strong oxidation, and high-density energy release, making them particularly prone to failure in launch vehicles. Therefore, research on engine failure trend prediction and health assessment is urgently needed to obtain timely information on failure development and prevent non-catastrophic engine failures from causing rocket flight problems.

[0003] The transition of a liquid rocket engine from a normal state to a state of complete failure is rarely instantaneous. There is a transitional process of fault development and propagation as the system evolves from normal to complete failure. Therefore, the purpose of fault trend prediction is to predict the severity of the fault over a subsequent period, in the early stages of engine failure. Studying the evolution trend of engine failures and predicting the operating state after a failure is a crucial approach to identifying the severity of faults early on, detecting serious faults, and taking appropriate measures. This improves the safety and reliability of the rocket propulsion system, ensuring the success of rocket launch missions.

[0004] To address the need for predicting and assessing the operational status of liquid rocket engines during launch vehicle flight, current research on liquid rocket engine fault prediction focuses primarily on predicting fault occurrence through engine parameter characterization. However, it lacks systematic research on predicting fault evolution trends and comprehensively assessing engine operational status after a fault occurs. Therefore, it is essential to design an effective and reliable system for predicting liquid rocket engine fault trends and conducting comprehensive health assessments. This system should enable comprehensive prediction of propulsion performance and safety status after a liquid rocket engine fault, and timely acquisition of the engine's fault development status. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention proposes a method for predicting the failure trend and assessing the health of liquid rocket engines. Based on the characteristics of engine thrust decay and the characteristics of state parameters reflecting the overall performance of the engine, a health index is designed and calculated for the engine's thrust and state parameters using the fuzzy membership method. An LSTM network is then used to predict the thrust and health index, and a comprehensive assessment of the engine's health degradation level is performed. Simulation results show that the proposed scheme can effectively predict and evaluate the engine's failure development trend.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0007] A method for predicting failure trends and assessing the health of a liquid rocket engine includes the following steps:

[0008] Step A): Based on the characteristics of engine thrust decay, design a single-parameter thrust health index based on the fuzzy membership method and calculate the thrust health; determine the thrust health level based on the correspondence between thrust health and thrust health level.

[0009] Step B): Based on the characteristics of the state parameters reflecting the overall performance of the engine, design a state health index based on the fuzzy membership method, calculate the state health, and thus assess the state degradation level.

[0010] Step C) uses the ReliefF algorithm to select fault trend prediction input feature parameters from the monitoring parameters of the liquid rocket engine. The dataset constructed using the fault trend prediction input feature parameters is used to train an LSTM network to predict engine thrust and condition health indicators. A health assessment logic that integrates thrust and condition degradation is designed to comprehensively assess the engine fault development trend.

[0011] As a further optimization of the liquid rocket engine failure trend prediction and health assessment method of the present invention, the specific steps of step A) are as follows:

[0012] Step A1): Based on the characteristics of engine thrust decline, define four evaluation levels: healthy, sub-healthy, critical fault, and fault. Establish a normal distribution probability density function as the membership function for the three fuzzy sets of healthy, sub-healthy, and critical fault, and establish a semi-trapezoidal function as the membership function for the fault fuzzy set. Substitute the engine thrust value into these functions to obtain the membership degree of the thrust to the four fuzzy sets, and calculate the fuzzy relation matrix at time point i.

[0013]

[0014] in, Let be the membership degree of the thrust at time point i with respect to the fuzzy set j, where j = 1, 2, 3, 4 represent the healthy, sub-healthy, critical fault, and fault fuzzy sets, respectively.

[0015] Step A2): Calculate the membership degree for each fuzzy set at each time point i = 1, 2, ..., n. Where n is the total number of time series points involved in the evaluation, the fuzzy relation matrix R of thrust at multiple time series points with respect to each fuzzy set is obtained. Thrust :

[0016]

[0017] Then, a membership degree fusion was performed at multiple time-series points using a two-level indicator fusion method; in the first-level fusion, membership degrees were fused based on three primary evaluation indicators. Obtain the evaluation matrix

[0018]

[0019] In the second-level fusion, the three primary evaluation indicators are weighted and calculated based on the secondary importance coefficients. The membership degree of the thrust to each fuzzy set after multi-time-series fusion is obtained by the following formula:

[0020]

[0021] Among them, b 1Thrust ,b 2Thrust ,b 3Thrust ,b 4Thrust , respectively represent the membership degree of the thrust obtained after multi-time-series point fusion to each fuzzy set, and A is the second-order importance coefficient;

[0022] Step A3), calculate thrust health HD from membership degree. Thrust :

[0023]

[0024] Where, k 1Thrust ,k 2Thrust ,…,k 7Thrust Undetermined coefficients for thrust health were calculated and obtained through curve fitting.

[0025] Step A4): Based on the correspondence between thrust health and thrust health level, the thrust health level is classified according to the calculated thrust health.

[0026] As a further optimization of the liquid rocket engine failure trend prediction and health assessment method of the present invention, the specific steps of step B) are as follows:

[0027] Step B1), select the main turbine speed n t Oxygen main pump post-pump pressure P po The pressure P after the first stage fuel pump pf1 Liquid oxygen main valve flow rate Q ovalve Fuel main valve flow rate Q fvalve This serves as a set of evaluation factors for the Condition Health (HD) index, which reflects the overall performance of an engine.

[0028] Step B2): Based on the characteristics of the state parameters reflecting the overall performance of the engine, four evaluation levels are defined: healthy, sub-healthy, critical fault, and fault. A normal distribution probability density function is established as the membership function for the three fuzzy sets of healthy, sub-healthy, and critical fault, and a semi-trapezoidal function is established as the membership function for the fault fuzzy set. The values ​​of each state parameter reflecting the overall performance of the engine are substituted into the membership function to obtain the membership degree of each state parameter to the four fuzzy sets. The fuzzy relation matrix R is calculated at time point i. i :

[0029]

[0030] in, Let be the membership degree of the state parameter at time point i with respect to the fuzzy set j, where j = 1, 2, 3, 4 represent the healthy, sub-healthy, critical fault and fault fuzzy sets respectively, and m represents the number of state parameters for evaluating the comprehensive performance of the engine.

[0031] Step B3), calculate the scale value d of the j-th parameter time point i. ij =c ij / s ij , where c ij Let s be the absolute value of the j-th parameter at time point i and the best estimate of that parameter. ij Let S be the arithmetic mean of the absolute values ​​of the differences between the value of the j-th parameter at time point i when the membership degree of each fuzzy set is 1 and the best estimate; construct the judgment matrix S. i :

[0032]

[0033] Find the judgment matrix S i The largest eigenvalue and its corresponding eigenvector are normalized to obtain the weight coefficients of each parameter.

[0034] Step B4) involves weighting the membership degree of each parameter to the fuzzy set with the weight coefficients of each parameter to obtain the fusion membership degree of each parameter to time point i, i.e.

[0035]

[0036] Among them, b ij Let a be the membership degree of time series point i for the j-th fuzzy set after multi-parameter fusion. ip Let be the weight coefficient of the p-th parameter at time point i. Let be the membership degree of the p-th parameter at time point i with respect to the j-th fuzzy set;

[0037] Step B5): Calculate the membership degree B for each fuzzy set after multi-parameter fusion at each time point i = 1, 2, ..., n. i =(bi1 b i2 b i3 b i4 ), where n is the total number of time series points participating in the evaluation, and the fuzzy relation matrix R of the state parameters with respect to each fuzzy set at multiple time series points is obtained:

[0038]

[0039] Then, membership fusion is performed at multiple time-series points using a two-level indicator fusion method; in the first-level fusion, the evaluation matrix B is obtained based on three primary evaluation indicators B1, B2, and B3. time1 :

[0040]

[0041] In the second-level fusion, the three primary evaluation indicators are weighted and calculated based on the secondary importance coefficients. The membership degree of the state parameters to each fuzzy set after multi-time-series point fusion is obtained by the following formula:

[0042] B time2 =(b1 b2 b3 b4) = A·B time1

[0043] Where b1, b2, b3, and b4 represent the membership degrees of the state parameters obtained after multi-time-series point fusion to each fuzzy set, and A is the second-order importance coefficient;

[0044] Step B6), calculate the engine health status HD based on membership degree:

[0045]

[0046] Where k1, k2, ..., k7 are the undetermined coefficients for calculating the health status, obtained by curve fitting;

[0047] Step B7): Based on the correspondence between state health and state degradation level, classify the state degradation level using the calculated state health.

[0048] As a further optimization of the liquid rocket engine failure trend prediction and health assessment method of the present invention, step C) is as follows:

[0049] Step C1) simulates the development and evolution of typical failures of liquid rocket engines by adding the time-varying failure factor parameter F of the components to the liquid rocket engine model.

[0050] Step C2) uses the ReliefF algorithm to select input feature parameters for fault trend prediction, taking a time window length of l, and determines the thrust LSTM prediction model expressions for oxygen preload turbine blade ablation, oxygen main pump cavitation, and combustion chamber gas leakage faults as follows:

[0051] Thrust(t+l)=lstm(n tppo (t),P ppo (t),Q ovalve (t),Q fvalve (t))

[0052] Thrust(t+l)=lstm(P po (t),Q ovalve (t),P ppo (t))

[0053] Thrust(t+l)=lstm(Q fvalve (t),Q ovalve (t))

[0054] The expression for the LSTM prediction model of health status is:

[0055] HD(t+l)=lstm(HD(t),Q fvalve (t),Q ovalve (t),P ppo (t),n tppo (t),P po (t))

[0056] Where t represents the current time, t+l represents the length of a future time window from the current time, lstm(·) represents the prediction model established by the LSTM network, and n tppo P is the engine oxygen preload turbine speed. ppo Q is the pressure after the oxygen pre-compression pump. ovalve Q is the flow rate of the liquid oxygen main valve. fvalve P is the flow rate of the main fuel valve. po This refers to the pressure after the main oxygen pump.

[0057] Step C3): Based on the dataset constructed using the fault trend prediction input feature parameters, train an LSTM network to predict engine thrust and condition health indicators.

[0058] Step C4): Combining the predicted and evaluated results of engine thrust and condition degradation, a comprehensive health assessment is conducted on the engine failure development trend. The logic of the comprehensive health assessment is as follows:

[0059] ①When the thrust health level is the same as the condition degradation level, the engine's overall health level is that health level;

[0060] ② When the thrust health level and the condition degradation level differ by one level, the engine's overall health level shall be the one that is relatively closer to the fault level.

[0061] ③ When the thrust health level differs from the condition degradation level by two or more levels, the engine health assessment fails.

[0062] As a further optimization of the liquid rocket engine failure trend prediction and health assessment method of the present invention, step C3) includes:

[0063] Step C3.1) predicts thrust by using the current data of the feature selection parameters as input to the algorithm and thrust as output. An LSTM thrust prediction model is trained to predict subsequent thrust trends. A fuzzy evaluation of the thrust is then performed to obtain the thrust health status (HD). Thrust ;

[0064] Step C3.2) performs a fuzzy evaluation on the state parameters reflecting the overall performance of the engine to obtain the state health degree HD; predicts the state health degree by taking the current data of the feature selection parameters as the input of the algorithm and the state health degree as the output of the algorithm, training an LSTM state health degree prediction model, and predicting the subsequent trend of state health degree changes.

[0065] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0066] (1) The present invention proposes a method for predicting the failure trend and assessing the health of a liquid rocket engine, based on thrust health HD. Thrust The prediction of engine condition health (HD) assesses the thrust and condition degradation under engine failure development, and provides a comprehensive evaluation of the engine failure development trend. It can obtain the engine failure development situation in a timely manner to avoid non-catastrophic engine failures affecting rocket flight.

[0067] (2) The present invention proposes a liquid rocket engine failure trend prediction and health assessment method, which adopts a combination of fuzzy logic evaluation and LSTM prediction. The health index is designed based on the fuzzy membership method, which is conducive to quantitatively evaluating the comprehensive working status of the engine affected by multiple factors. The LSTM network is established for parameter prediction, which has advantages in long-term series prediction of engine failure development. Attached Figure Description

[0068] Figure 1 This is a structural diagram of a liquid rocket engine failure trend prediction and health assessment method proposed in this invention;

[0069] Figure 2It is a membership function curve of health, sub-health, critical fault, and fault fuzzy set;

[0070] Figure 3 This is a diagram showing the main structure and flow path distribution of a certain type of liquid oxygen / kerosene staged combustion cycle liquid rocket engine.

[0071] Figure 4 These are curves showing the development and changes of fault factor parameters for each typical fault mode.

[0072] Figure 5 The diagram shows the characteristic weights of each parameter with respect to thrust, including (a) the characteristic weights of each parameter under oxygen preload turbine blade ablation failure, (b) the characteristic weights of each parameter under oxygen main pump cavitation failure, and (c) the characteristic weights of each parameter under combustion chamber gas leakage failure.

[0073] Figure 6 It is a feature weight diagram of each parameter with respect to the state of health;

[0074] Figure 7 The figure shows the predicted thrust trend of the engine under the oxygen preload turbine blade ablation failure development under the rated conditions of the main stage, where (a) engine thrust prediction and (b) absolute value of the relative error of thrust prediction.

[0075] Figure 8 The figure shows the trend prediction and evaluation results of engine thrust health index under the development of oxygen preloaded turbine blade ablation failure under the rated conditions of the main stage, where (a) engine thrust health prediction and (b) engine thrust health level evaluation.

[0076] Figure 9 The figure shows the predicted thrust trend of the engine under the development of oxygen main pump cavitation failure under the rated conditions of the main stage, where (a) engine thrust prediction and (b) absolute value of the relative error of thrust prediction.

[0077] Figure 10 The figure shows the trend prediction and evaluation results of engine thrust health index under the development of oxygen main pump cavitation failure under the rated conditions of the main stage, where (a) engine thrust health prediction and (b) engine thrust health level evaluation.

[0078] Figure 11 The figure shows the predicted thrust trend of the engine under the rated conditions of the main stage under the combustion chamber gas leakage fault development. (a) Engine thrust prediction, (b) Absolute value of the relative error of thrust prediction.

[0079] Figure 12 The figure shows the trend prediction and evaluation results of engine thrust health index under the development of combustion chamber gas leakage fault under the rated conditions of the main stage, where (a) engine thrust health prediction and (b) engine thrust health level evaluation.

[0080] Figure 13 The graph shows the trend prediction and evaluation results of engine health indicators under the development of oxygen preload turbine blade ablation failure under the rated conditions of the main stage, where (a) engine health prediction and (b) engine degradation level evaluation.

[0081] Figure 14 The graph shows the trend prediction and evaluation results of engine health indicators under the development of oxygen main pump cavitation failure under the rated conditions of the main stage, including (a) engine health prediction and (b) engine condition degradation level evaluation.

[0082] Figure 15 This is a graph showing the trend prediction and evaluation results of engine health indicators under the development of combustion chamber gas leakage faults under the main stage operating conditions and rated conditions. (a) Engine health prediction, (b) Engine degradation level evaluation. Detailed Implementation

[0083] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings.

[0084] The specific implementation of this invention takes the design of a fault trend prediction and health assessment method for a certain type of liquid oxygen / kerosene staged combustion cycle liquid rocket engine of the Long March series carrier rockets as an example. The structure of the method is as follows: Figure 1 As shown, the specific steps include:

[0085] Step A): Based on the characteristics of engine thrust decay, design a single-parameter thrust health index based on the fuzzy membership method and calculate the thrust health; determine the thrust health level based on the correspondence between thrust health and thrust health level.

[0086] Step B): Based on the characteristics of the state parameters reflecting the overall performance of the engine, design a state health index based on the fuzzy membership method, calculate the state health, and thus assess the state degradation level.

[0087] Step C) uses the ReliefF algorithm to select fault trend prediction input feature parameters from the monitoring parameters of the liquid rocket engine. The dataset constructed using the fault trend prediction input feature parameters is used to train an LSTM network to predict engine thrust and condition health indicators. A health assessment logic that integrates thrust and condition degradation is designed to comprehensively assess the engine fault development trend.

[0088] The detailed steps of step A) are as follows:

[0089] Step A1): Based on the characteristics of engine thrust decline, define four evaluation levels: healthy, sub-healthy, critical fault, and fault. Establish a normal distribution probability density function as the membership function for the three fuzzy sets of healthy, sub-healthy, and critical fault, and establish a semi-trapezoidal function as the membership function for the fault fuzzy set. The function graphs are shown below. Figure 2 As shown, the function expression is as follows:

[0090]

[0091]

[0092]

[0093]

[0094] in, Let be the membership degree of the thrust at time point i with respect to the fuzzy set j, where j = 1, 2, 3, 4 represent the healthy, sub-healthy, critical fault, and fault fuzzy sets, respectively. Let μ be the thrust value at time point i. i The average value of the parameters at time point i is used as the best estimate of the expected state, δ i Let be the standard deviation at time point i. Substituting the engine thrust value into the membership function above yields the membership degree of thrust to each fuzzy set. The fuzzy relation matrix is ​​then calculated at time point i.

[0095]

[0096] Step A2): Calculate the membership degree for each fuzzy set at each time point i = 1, 2, ..., n. Where n is the total number of time series points involved in the evaluation, the fuzzy relation matrix R of thrust at multiple time series points with respect to each fuzzy set is obtained. Thrust :

[0097]

[0098] Then, membership fusion was performed at the multi-time-series level using a two-level indicator fusion method. In the first-level fusion, the importance of data at each time-series point was equal, and three fuzzy computation models were used for fusion, namely:

[0099] The first type is M(∧,∨) fuzzy operation. Take a i =1, i=1,2,…,n, the fusion result is expressed as

[0100]

[0101] The second type is M(∨,∧) fuzzy operation. Take ai =1, i=1,2,…,n, the fusion result is expressed as

[0102]

[0103] The third type is M(·,+) fuzzy arithmetic. Take a i =1 / n, i = 1, 2, ..., n, the fusion result is expressed as

[0104]

[0105] Therefore, based on the above three primary evaluation indicators... Obtain the evaluation matrix

[0106]

[0107] The results of the first-level fusion are further processed in the second-level fusion. Based on the principle that the closer the data at each time point is to the fault, the greater the impact on the engine, the secondary importance coefficient vector A is determined, as shown in Table 1. Then, the membership degree of thrust to each fuzzy set after multi-time-point fusion is calculated by the following formula:

[0108]

[0109] Among them, b 1Thrust ,b 2Thrust ,b 3Thrust ,b 4Thrust These represent the membership degrees of the thrust obtained after multi-time-series point fusion to each fuzzy set;

[0110] Table 1. Importance Coefficient of Second-Level Engine Second-Level Indicator Integration Evaluation

[0111]

[0112] according to Determine the fuzzy set to which it belongs, and select the corresponding importance coefficient from the table above;

[0113] Step A3), calculate the thrust health (HD) based on the membership degree of the thrust to each fuzzy set. Thrust :

[0114]

[0115] Where, k 1Thrust ,k 2Thrust ,…,k 7Thrust Undetermined coefficients for thrust health were calculated and obtained through curve fitting.

[0116] Step A4), the calculated health score ranges from [0,1]. Within this range, a higher health score indicates a closer relationship to a healthy state and a higher level of health. Based on the correspondence between thrust health score and thrust health level shown in Table 2, the thrust health level is classified according to the calculated thrust health score.

[0117] Table 2 Correspondence between Health Status and Health Deterioration Level

[0118]

[0119] The detailed steps of step B) are as follows:

[0120] Step B1), select the main turbine speed n according to the flow path structure of the liquid rocket engine. t Oxygen main pump post-pump pressure P po The pressure P after the first stage fuel pump pf1 Liquid oxygen main valve flow rate Q ovalve Fuel main valve flow rate Q fvalve This serves as a set of evaluation factors for the Condition Health (HD) index, which reflects the overall performance of an engine.

[0121] Step B2): Based on the characteristics of the state parameters reflecting the overall performance of the engine, four evaluation levels are defined: healthy, sub-healthy, critical fault, and fault. A normal distribution probability density function is established as the membership function for the three fuzzy sets of healthy, sub-healthy, and critical fault, and a semi-trapezoidal function is established as the membership function for the fault fuzzy set. The function graphs are shown below. Figure 2 As shown, the function expression is as follows:

[0122]

[0123]

[0124]

[0125]

[0126] Where, r ij Let x be the membership degree of the state parameter at time point i with respect to the fuzzy set j, where j = 1, 2, 3, 4 represent the healthy, sub-healthy, critical fault, and fault fuzzy sets, respectively. i Let μ be the state parameter value at time point i. i The average value of the parameters at time point i is used as the best estimate of the expected state, δ i Let be the standard deviation at time point i. Substituting the values ​​of each state parameter reflecting the engine's overall performance into the membership function yields the membership degree of each state parameter to the four fuzzy sets. The fuzzy relation matrix R is then calculated at time point i. i :

[0127]

[0128] in, Let be the membership degree of the state parameter at time point i with respect to the fuzzy set j, where j = 1, 2, 3, 4 represent the healthy, sub-healthy, critical fault and fault fuzzy sets respectively, and m represents the number of state parameters for evaluating the comprehensive performance of the engine.

[0129] Step B3) Determine the weight coefficients of each parameter in the multi-parameter evaluation using the analytic hierarchy process (AHP). Calculate the scale value d of the j-th parameter at time point i. ij =c ij / s ij , where c ij Let s be the absolute value of the j-th parameter at time point i and the best estimate of that parameter. ij Let S be the arithmetic mean of the absolute values ​​of the differences between the value of the j-th parameter at time point i when the membership degree of each fuzzy set is 1 and the best estimate; construct the judgment matrix S. i :

[0130]

[0131] Find the judgment matrix S i The largest eigenvalue and its corresponding eigenvector are normalized to obtain the weight coefficients of each parameter.

[0132] Step B4) involves calculating the weighted average of the fuzzy set membership degree and the weight coefficients of each parameter to obtain the fusion membership degree of each parameter with respect to time series point i, i.e.

[0133]

[0134] Among them, b ij Let a be the membership degree of time series point i for the j-th fuzzy set after multi-parameter fusion. ip Let be the weight coefficient of the p-th parameter at time point i. Let be the membership degree of the p-th parameter at time point i with respect to the j-th fuzzy set;

[0135] Step B5): Calculate the membership degree B for each fuzzy set after multi-parameter fusion at each time point i = 1, 2, ..., n. i =(b i1 b i2 b i3 b i4 ), where n is the total number of time series points participating in the evaluation, and the fuzzy relation matrix R of the state parameters with respect to each fuzzy set at multiple time series points is obtained:

[0136]

[0137] Then, membership fusion was performed at the multi-time-series level using a two-level indicator fusion method. In the first-level fusion, the importance of data at each time-series point was equal, and three fuzzy computation models were used for fusion, namely:

[0138] The first type is M(∧,∨) fuzzy operation. Take a i =1, i=1,2,…,n, the fusion result is expressed as

[0139]

[0140] The second type is M(∨,∧) fuzzy operation. Take a i =1, i=1,2,…,n, the fusion result is expressed as

[0141]

[0142] The third type is M(·,+) fuzzy arithmetic. Take a i =1 / n, i = 1, 2, ..., n, the fusion result is expressed as

[0143]

[0144] Therefore, the evaluation matrix B is obtained based on the three primary evaluation indicators B1, B2, and B3. time1 :

[0145]

[0146] In the second-level fusion, the result R of the first-level fusion is further processed. Based on the principle that the closer the data at each time point is to the fault, the greater the impact on the engine, the second-level importance coefficient vector A is determined as shown in Table 1. Determine the fuzzy set to which the state parameter belongs, and select the corresponding importance coefficient from Table 1. Then, calculate the membership degree of the state parameter to each fuzzy set after multi-time-series point fusion using the following formula:

[0147] B time2 =(b1 b2 b3 b4) = A·B time1

[0148] Where b1, b2, b3, and b4 represent the membership degrees of the state parameters obtained after multi-temporal point fusion to each fuzzy set.

[0149] Step B6), calculate the engine health status HD based on membership degree:

[0150]

[0151] Where k1, k2, ..., k7 are the undetermined coefficients for calculating the health status, obtained by curve fitting;

[0152] Step B7) Based on the correspondence between state health and state degradation level as shown in Table 2, the calculated state health is used to classify the state degradation level.

[0153] The detailed steps of step C) are as follows:

[0154] Step C1) Simulate the development and evolution of typical liquid rocket engine failures using a liquid rocket engine model. The main structure and flow path distribution of the engine are as follows: Figure 3 As shown.

[0155] For oxygen-preloaded turbine blade ablation failure, its mathematical description can be intuitively represented as:

[0156]

[0157] Where, η t (t) represents the normal value of turbine efficiency, η′ t (t) represents the actual turbine efficiency, and t represents the operating time. f F1 represents the time of failure occurrence and is the turbine failure factor. When 0 ≤ F1 < 1, it indicates that the turbine has experienced blade ablation failure; when F1 = 1, the turbine is operating normally.

[0158] For the cavitation failure of the oxygen main pump, its mathematical description can be intuitively represented as:

[0159]

[0160] Where ΔP(t) is the normal value of pump head, ΔP′(t) is the actual value of pump head, and F2 is the pump failure factor. When 0≤F2<1, it indicates that the pump has cavitation failure; when F2=1, the pump is working normally.

[0161] For combustion chamber gas leakage faults, the mathematical description of the fault can be intuitively represented as follows:

[0162]

[0163] Where, q ig (t), q lo (t), q lf (t) represents the normal values ​​of the gas mass flow rate, liquid oxidant mass flow rate, and liquid fuel mass flow rate flowing into the thermal assembly, respectively. eg (t) represents the normal mass flow rate of the gas exiting the thermal assembly. F3 is the actual value of the working fluid mass change rate of the heating module, and F3 is the heating module failure factor. When F3 > 0, it indicates that the heating module has experienced a gas leakage failure; when F3 = 0, the heating module is working normally.

[0164] By setting the time-varying fault factor parameter F of engine components, the evolution trend of typical engine faults can be simulated, such as... Figure 4 As shown;

[0165] Step C2), as shown in Table 3, involves selecting input feature parameters for liquid rocket engine monitoring using the ReliefF algorithm for fault trend prediction:

[0166] Table 3 Monitoring Parameters of Liquid Rocket Engines

[0167]

[0168]

[0169] A time window length of l = 3s was used to generate normalized data samples for three fault conditions: oxygen preload turbine blade ablation fault, oxygen main pump cavitation fault, and combustion chamber gas leakage fault. These samples included monitoring parameters and thrust data after one window length for feature selection. The feature weights of each parameter with respect to thrust obtained by the ReliefF algorithm are as follows: Figure 5 As shown, based on the principle that the feature weight of the selected parameter is significantly higher than that of other parameters and the weight is greater than 0, n is selected for oxygen preload turbine blade ablation failure. tppo 、P ppo Q ovalve Q fvalve As the predictor input; for oxygen main pump cavitation failure, P is selected. po Q ovalve 、P ppo As the predictor input; for combustion chamber gas leakage faults, Q is selected. fvalve Q ovalve As the input to the predictor, if t represents the current time, t+l represents the length of a future time window from the current time, and lstm(·) represents the prediction model established by the LSTM network, the predictor can be represented as follows:

[0170] Thrust(t+l)=lstm(n tppo (t),P ppo (t),Q ovalve (t),Q fvalve (t))

[0171] Thrust(t+l)=lstm(P po (t),Q ovalve (t),P ppo(t))

[0172] Thrust(t+l)=lstm(Q fvalve (t),Q ovalve (t))

[0173] For the prediction of health status, a time window length of l = 3s is also used to conduct feature selection research on data samples of three fault conditions. The feature weights of each parameter with respect to health status obtained by the ReliefF algorithm are as follows: Figure 6 As shown, therefore HD and Q are selected. fvalve Q ovalve 、P ppo n tppo 、P po As input to the predictor, the predictor can be represented as:

[0174] HD(t+l)=lstm(HD(t),Q fvalve (t),Q ovalve (t),P ppo (t),n tppo (t),P po (t))

[0175] Step C3): Based on the dataset constructed using the fault trend prediction input feature parameters, an LSTM network is trained to predict engine thrust and condition health indicators. First, a fuzzy evaluation is performed on the condition parameters reflecting the engine's overall performance to obtain the condition health (HD). Then, the current data of the selected parameters is used as the prediction input to establish an LSTM prediction model for thrust and condition health.

[0176] Step C3.1) Train an LSTM network to predict engine thrust health indicators.

[0177] Under rated operating conditions of the engine main stage, for oxygen preloaded turbine blade ablation faults, injection should begin from 50 seconds. Figure 3 A time-varying fault factor F1 is used, with a prediction window length of l = 3s. An LSTM network is trained to predict thrust. The number of hidden layer nodes in the LSTM network is set to 40, the dropout layer to 0.5, the number of training iterations to 120, the mini-batch size to 10, and the initial learning rate to 0.002. Figure 7 The figure shows the thrust prediction results and prediction errors. Using a fuzzy evaluation method to calculate the thrust health, the thrust health prediction results are as follows: Figure 8 As shown.

[0178] For oxygen main pump cavitation failure, injection should begin from 50 seconds. Figure 3A time-varying fault factor F2 was used to train an LSTM network to predict thrust. The LSTM network was configured with 90 hidden layer nodes, a dropout layer size of 0.5, 120 training iterations, a mini-batch size of 10, and an initial learning rate of 0.004. Figure 9 The figure shows the thrust prediction results and prediction errors. Using a fuzzy evaluation method to calculate the thrust health, the thrust health prediction results are as follows: Figure 10 As shown.

[0179] For combustion chamber gas leakage faults, inject gas starting from 50 seconds. Figure 3 A time-varying fault factor F3 was used to train an LSTM network to predict thrust. The LSTM network was configured with 100 hidden layer nodes, a dropout layer size of 0.5, 120 training iterations, a mini-batch size of 10, and an initial learning rate of 0.003. Figure 11 The figure shows the thrust prediction results and prediction errors. Using a fuzzy evaluation method to calculate the thrust health, the thrust health prediction results are as follows: Figure 12 As shown;

[0180] Step C3.2) Train the LSTM network to predict engine health indicators.

[0181] A fuzzy evaluation method was used to calculate the health status of state parameters reflecting the overall engine performance under the development of oxygen preload turbine blade ablation failure, oxygen main pump cavitation failure, and combustion chamber gas leakage failure. Then, an LSTM network was trained on the time series of state health status data for prediction. The state health status prediction results are shown below. Figure 13 , 14 As shown in Figure 15.

[0182] The effectiveness of the forecasting method is evaluated using two indicators: root mean square error (RMSE) and mean absolute error (MAE). The expressions for these indicators are as follows:

[0183]

[0184]

[0185] Among them, y i This represents the expected output value. The value represents the algorithm's prediction, and N is the total number of samples. The prediction results of this invention under various fault modes are shown in Table 4:

[0186] Table 4. Failure Trend Prediction Errors for Various Failure Modes in Liquid Rocket Engines

[0187]

[0188] Step C4): Combining the predicted and evaluated results of engine thrust and condition degradation, a comprehensive health assessment is conducted on the engine failure development trend. The logic of the comprehensive health assessment is as follows:

[0189] ①When the thrust health level is the same as the condition degradation level, the engine's overall health level is that health level;

[0190] ② When the thrust health level and the condition degradation level differ by one level, the engine's overall health level shall be the one that is relatively closer to the fault level.

[0191] ③ When the thrust health level differs from the condition degradation level by two or more levels, the engine health assessment fails.

[0192] The engine thrust health level assessment results are as follows: Figure 8 , 10 As shown in Figures 1 and 12, the assessment results of the state degradation level are as follows: Figure 13 , 14 As shown in Figure 15.

[0193] As shown in the figure, for the oxygen preload turbine blade ablation fault, the thrust and state parameters were only slightly affected for a considerable period after the fault occurred, eventually reaching a sub-healthy level. Therefore, the overall assessment result is a sub-healthy level, indicating that the fault has a relatively limited impact on the safe and reliable operation of the engine, but the risk over time still needs to be considered. For the oxygen main pump cavitation fault, both the thrust and state parameters were significantly affected immediately after the fault occurred, directly reaching the critical fault level. Subsequently, the thrust continued to decline, eventually reaching the fault level. Therefore, the overall assessment result is a fault level, indicating that the fault has a significant impact on the safe and reliable operation of the engine. For the combustion chamber gas leakage fault, both the thrust and state parameters were significantly affected immediately after the combustion chamber gas leakage fault occurred, with the thrust directly reaching the fault level and the state parameters directly reaching the critical fault level. Therefore, the overall assessment result is a fault level, indicating that the fault has a significant and rapid impact on the safe and reliable operation of the engine.

[0194] Depend on Figures 7-15 As shown in Table 4, the RMSE and MAE errors for each failure mode are relatively small, and the thrust health level and condition degradation level under the same operating condition differ by at most one level. This indicates that the liquid rocket engine failure trend prediction and health assessment method proposed in this invention can effectively meet the requirements for predicting the failure development trend of each typical failure mode under the rated operating condition of the engine main stage, and can pass the thrust health HD assessment. ThrustThe prediction of engine condition health (HD) assesses the thrust and condition degradation under engine failure development, thereby providing a comprehensive health evaluation of the engine failure development trend. This invention combines fuzzy logic evaluation with LSTM prediction, designs a health index based on the fuzzy membership method, which is beneficial for quantitatively evaluating the comprehensive operating status of the engine affected by multiple factors; and establishes an LSTM network for parameter prediction, demonstrating the advantages of LSTM in long-term series prediction of engine failure development.

[0195] It should be noted that 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 variations and substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included 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 predicting the failure trend and assessing the health of a liquid rocket engine, characterized in that, Includes the following steps: Step A): Based on the characteristics of engine thrust decay, design a single-parameter thrust health index based on the fuzzy membership method and calculate the thrust health; determine the thrust health level based on the correspondence between thrust health and thrust health level. Step B): Based on the characteristics of the state parameters reflecting the overall performance of the engine, design a state health index based on the fuzzy membership method, calculate the state health, and thus assess the state degradation level. The specific steps are as follows: Step B1), select the main turbine speed n t Oxygen main pump post-pump pressure P po The pressure P after the first stage fuel pump pf1 Liquid oxygen main valve flow rate Q ovalve Fuel main valve flow rate Q fvalve This serves as a set of evaluation factors for the Condition Health (HD) index, which reflects the overall performance of an engine. Step B2): Based on the characteristics of the state parameters reflecting the overall performance of the engine, four evaluation levels are defined: healthy, sub-healthy, critical fault, and fault. A normal distribution probability density function is established as the membership function for the three fuzzy sets of healthy, sub-healthy, and critical fault, and a semi-trapezoidal function is established as the membership function for the fault fuzzy set. The values ​​of each state parameter reflecting the overall performance of the engine are substituted into the membership function to obtain the membership degree of each state parameter to the four fuzzy sets. The fuzzy relation matrix R is calculated at time point i. i : in, Let be the membership degree of the state parameter at time point i with respect to the fuzzy set j, where j = 1, 2, 3, 4 represent the healthy, sub-healthy, critical fault and fault fuzzy sets respectively, and m represents the number of state parameters for evaluating the comprehensive performance of the engine. Step B3), calculate the scale value d of the j-th parameter time point i. ij =c ij / s ij , where c ij Let s be the absolute value of the j-th parameter at time point i and the best estimate of that parameter. ij Let S be the arithmetic mean of the absolute values ​​of the differences between the value of the j-th parameter at time point i when the membership degree of each fuzzy set is 1 and the best estimate; construct the judgment matrix S. i : Find the judgment matrix S i The largest eigenvalue and its corresponding eigenvector are normalized to obtain the weight coefficients of each parameter. Step B4) involves weighting the membership degree of each parameter to the fuzzy set with the weight coefficients of each parameter to obtain the fusion membership degree of each parameter to time series point i, i.e. Among them, b ij Let a be the membership degree of time series point i for the j-th fuzzy set after multi-parameter fusion. ip Let be the weight coefficient of the p-th parameter at time point i. Let be the membership degree of the p-th parameter at time point i with respect to the j-th fuzzy set; Step B5): At each time point i = 1, 2, ..., n, calculate the membership degree B after multi-parameter fusion for each fuzzy set. i =(b i1 b i2 b i3 b i4 ), where n is the total number of time series points participating in the evaluation, and the fuzzy relation matrix R of the state parameters with respect to each fuzzy set at multiple time series points is obtained: Then, membership fusion is performed at multiple time-series points using a two-level indicator fusion method; in the first-level fusion, the evaluation matrix B is obtained based on three primary evaluation indicators B1, B2, and B3. time1 : In the second-level fusion, the three primary evaluation indicators are weighted and calculated based on the secondary importance coefficients. The membership degree of the state parameters to each fuzzy set after multi-time-series point fusion is obtained by the following formula: B time2 =(b1 b2 b3 b4)=AgB time1 Where b1, b2, b3, and b4 represent the membership degrees of the state parameters obtained after multi-time-series point fusion to each fuzzy set, and A is the second-order importance coefficient; Step B6), calculate the engine health status HD based on membership degree: Where k1, k2, ..., k7 are the undetermined coefficients for calculating the health status, obtained by curve fitting; Step B7): Based on the correspondence between state health and state degradation level, classify the state degradation level using the calculated state health. Step C) uses the ReliefF algorithm to select fault trend prediction input feature parameters from the monitoring parameters of the liquid rocket engine. The dataset constructed using the fault trend prediction input feature parameters is used to train an LSTM network to predict engine thrust and condition health indicators. A health assessment logic that integrates thrust and condition degradation is designed to comprehensively assess the engine fault development trend.

2. The method for predicting the failure trend and assessing the health of a liquid rocket engine as described in claim 1, characterized in that, The specific steps of step A) are as follows: Step A1): Based on the characteristics of engine thrust decline, define four evaluation levels: healthy, sub-healthy, critical fault, and fault. Establish a normal distribution probability density function as the membership function for the three fuzzy sets of healthy, sub-healthy, and critical fault, and establish a semi-trapezoidal function as the membership function for the fault fuzzy set. Substitute the engine thrust value into these functions to obtain the membership degree of the thrust to the four fuzzy sets, and calculate the fuzzy relation matrix at time point i. in, Let be the membership degree of the thrust at time point i with respect to the fuzzy set j, where j = 1, 2, 3, 4 represent the healthy, sub-healthy, critical fault, and fault fuzzy sets, respectively. Step A2): Calculate the membership degree for each fuzzy set at each time point i = 1, 2, ..., n. Where n is the total number of time series points involved in the evaluation, the fuzzy relation matrix R of thrust at multiple time series points with respect to each fuzzy set is obtained. Thrust : Then, a membership degree fusion was performed at multiple time-series points using a two-level indicator fusion method; in the first-level fusion, membership degrees were fused based on three primary evaluation indicators. Obtain the evaluation matrix In the second-level fusion, the three first-level evaluation indicators are weighted and calculated based on the second-level importance coefficient. The membership degree of the thrust to each fuzzy set after multi-time-series point fusion is obtained by the following formula: Among them, b 1Thrust ,b 2Thrust ,b 3Thrust ,b 4Thrust , respectively represent the membership degree of the thrust obtained after multi-time-series point fusion to each fuzzy set, and A is the second-order importance coefficient; Step A3), calculate thrust health HD from membership degree. Thrust : Among them, k 1Thrust ,k 2Thrust ,…,k 7Thrust Undetermined coefficients for thrust health were calculated and obtained through curve fitting. Step A4): Based on the correspondence between thrust health and thrust health level, the thrust health level is classified according to the calculated thrust health.

3. The method for predicting the failure trend and assessing the health of a liquid rocket engine as described in claim 1, characterized in that, The specific steps of step C) are as follows: Step C1) simulates the development and evolution of typical failures of liquid rocket engines by adding the time-varying failure factor parameter F of the components to the liquid rocket engine model. Step C2) uses the ReliefF algorithm to select input feature parameters for fault trend prediction, taking a time window length of l, and determines the thrust LSTM prediction model expressions for oxygen preload turbine blade ablation, oxygen main pump cavitation, and combustion chamber gas leakage faults as follows: Thrust(t+l)=lstm(n tppo (t),P ppo (t),Q ovalve (t),Q fvalve (t)) Thrust(t+l)=lstm(P po (t),Q ovalve (t),P ppo (t)) Thrust(t+l)=lstm(Q fvalve (t),Q ovalve (t)) The expression for the LSTM prediction model of health status is: HD(t+l)=lstm(HD(t),Q fvalve (t),Q ovalve (t),P ppo (t),n tppo (t),P po (t)) Where t represents the current time, t+l represents the length of a future time window from the current time, lstm(g) represents the prediction model established by the LSTM network, and n tppo P is the engine oxygen preload turbine speed. ppo Q is the pressure after the oxygen pre-compression pump. ovalve Q is the flow rate of the liquid oxygen main valve. fvalve P is the flow rate of the main fuel valve. po This refers to the pressure after the main oxygen pump. Step C3): Based on the dataset constructed using the fault trend prediction input feature parameters, train an LSTM network to predict engine thrust and condition health indicators. Step C4): Combining the predicted and evaluated results of engine thrust and condition degradation, a comprehensive health assessment is conducted on the engine failure development trend. The logic of the comprehensive health assessment is as follows: ①When the thrust health level is the same as the condition degradation level, the engine's overall health level is that health level; ② When the thrust health level and the condition degradation level differ by one level, the engine's overall health level shall be the one that is relatively closer to the fault level. ③ When the thrust health level differs from the condition degradation level by two or more levels, the engine health assessment fails.

4. The method for predicting the failure trend and assessing the health of a liquid rocket engine as described in claim 3, characterized in that, The specific steps of step C3) are as follows: Step C3.1) predicts thrust by using the current data of the feature selection parameters as input to the algorithm and thrust as output. An LSTM thrust prediction model is trained to predict subsequent thrust trends. Then, a fuzzy evaluation of the thrust is performed to obtain the thrust health status (HD). Thrust ; Step C3.2) involves performing a fuzzy evaluation on the state parameters reflecting the overall performance of the engine to obtain the state health status HD; To predict state health, the current data of the feature selection parameters are used as the input of the algorithm, and the state health is used as the output of the algorithm. The LSTM state health prediction model is trained to predict the subsequent trend of state health changes.

Images