A multivariable fault-tolerant control method for liquid rocket engine thrust holding

Abstract

The application discloses a kind of liquid rocket engine thrust keeping multivariable fault-tolerant control methods, comprising: analyzing liquid rocket engine failure mechanism and influence characteristics, establish typical fault model;With mixing ratio as constraint quantity, construct multivariable predictive optimal controller for thrust keeping, and design fault-tolerant control architecture verifies the performance of control logic under typical fault injection.The application takes the combustion chamber pressure reflecting thrust as the controlled variable, the mixing ratio as the constraint quantity, the flow regulator valve core and the fuel main valve opening as the control variable, based on the neural network prediction model offline training, and according to the error containing penalty function, the control variable quadratic form index is optimized on-line, the control variable for thrust keeping is calculated, and the maximum change of mixing ratio is small, and the specific impulse is higher.The application reduces the transient loss and steady loss of engine after failure, and meets the demand of thrust keeping and mixing ratio constraint in main stage working condition, and has strong fault-tolerant capability.

Description

Technical Field

[0001] This invention belongs to the field of liquid rocket engine control and simulation, and specifically relates to a multivariable fault-tolerant control method for maintaining thrust in a liquid rocket engine. Background Technology

[0002] Liquid-propellant rocket engines (LREs) are the heart of space launch vehicles. They convert the chemical energy of the carried liquid propellant into thermal energy in thermodynamic components, and then into the kinetic energy of the ejected gas in the nozzle, thereby generating thrust. Due to the harsh operating environment of the engine, various failures can easily occur, affecting rocket flight missions. Different failure modes result in different patterns of change in the engine's internal parameters. Therefore, research on the failure characteristics of liquid rocket engines is a hot topic both domestically and internationally. To analyze the failure characteristics of liquid rocket engines, given the high cost of testing and the scarcity of engine test failure data, conducting research on rocket engine failure modeling techniques is essential to improving the reliability and safety of propulsion systems.

[0003] Thrust, as a core characteristic parameter of rocket engine status, directly affects whether the rocket can fly safely and reliably, and is therefore one of the most important performance indicators of the engine. When an engine fails, thrust will decline. In this case, engine thrust maintenance fault-tolerant control technology will minimize the loss of thrust. The mixture ratio is defined as the ratio of the mass flow rate of oxidizer and fuel consumed by the engine. Since the mixture ratio should not change too much during thrust control to facilitate the control of the overall propellant loading of the two rocket tanks and to ensure the stability of the engine's own performance, the constraint of the mixture ratio needs to be taken into consideration to prevent the mixture ratio from deviating too much from the design value.

[0004] Because liquid rocket engines require high system efficiency during thrust control and a stable mixture ratio throughout operation, and because they must maintain thrust and limit mixture ratio deviations from design values ​​after engine failure, the design of the engine control system faces significant challenges. Conventional loop-switching control methods, employing distributed loops to form a multivariable control system, fail to consider coupling effects, thus impacting thrust and mixture ratio maintenance. Therefore, given the frequent failures in liquid rocket engines, it is essential to research a reliable and safe multivariable fault-tolerant control method for maintaining engine thrust, based on typical fault modeling. This method aims to eliminate the effects of coupling, balance thrust and mixture ratio control requirements, and minimize the reliability and safety impacts of engine failures. Summary of the Invention

[0005] This invention addresses the aforementioned technical problems by proposing a multivariable fault-tolerant control method for thrust maintenance in liquid rocket engines. Based on the analysis of the effects and characteristics of typical engine faults, a typical fault model was established. Through offline training of a neural network prediction model, with the objective of minimizing the error with penalty functions and the quadratic performance index of the control quantity, a neural network optimization equation was established, and a multivariable predictive optimization controller for thrust maintenance was designed. Finally, combined with the typical fault model, a thrust maintenance fault-tolerant control architecture based on the multivariable predictive optimization controller was constructed. The results verify that this control method has superior thrust maintenance and mixture ratio limiting effects under fault conditions, exhibiting good steady-state and dynamic performance.

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

[0007] A multivariable fault-tolerant control method for maintaining thrust in a liquid rocket engine includes the following steps:

[0008] Step A) Based on the typical failure modes of liquid rocket engines, determine the failure mechanisms and characteristic parameters of oxygen pre-compression turbine blade ablation, oxygen main pump cavitation, and combustion chamber gas leakage, and introduce failure factors to simulate failures for different typical failures, and establish a typical failure model of liquid rocket engines.

[0009] Step B) Train a neural network prediction model with a dataset consisting of the output state vector and the input control vector obtained from the typical fault model of liquid rocket engine. The output state vector includes the mixing ratio. Based on the quadratic index of error and control quantity, a penalty function term that limits the mixing ratio is introduced to perform neural network optimization solution for thrust maintenance and design a thrust maintenance multivariate prediction optimization controller that ensures high safety.

[0010] Step C) Design a thrust holding fault-tolerant control architecture based on a multivariable predictive optimization controller. Under the rated conditions of the main stage, simulate the occurrence of typical faults such as oxygen pre-compression turbine blade ablation, oxygen main pump cavitation, and combustion chamber gas leakage in a liquid rocket engine, and study and verify the thrust holding fault-tolerant control method and control performance.

[0011] As a further optimization of the multivariable fault-tolerant control method for maintaining thrust in a liquid rocket engine according to the present invention, the specific steps of step A) are as follows:

[0012] Step A1) involves summarizing typical faults from a liquid rocket engine fault mode library, and analyzing the characteristics of fault-affecting parameters from the perspective of fault mechanism for faults such as oxygen pre-compression turbine blade ablation, oxygen main pump cavitation, and combustion chamber gas leakage. The analysis results are as follows: the characteristic parameters affecting oxygen pre-compression turbine blade ablation are a decrease in turbine efficiency and speed; the characteristic parameters affecting oxygen main pump cavitation are a decrease in pump head and downstream pressure; and the characteristic parameters affecting combustion chamber gas leakage are an increase in component outflow rate and a decrease in component pressure.

[0013] Step A2) is to establish a typical engine fault model by introducing a fault factor parameter F to correct the mathematical model parameter D of the engine components. F represents the severity of the fault in the engine components.

[0014] For oxygen preloaded turbine blade ablation failure, due to turbine efficiency η t The failure model mathematically describes the resulting decline as follows:

[0015]

[0016] 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.

[0017] For the cavitation failure of the oxygen main pump, since the pump head ΔP decreases, its mathematical expression is as follows:

[0018]

[0019] 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.

[0020] For combustion chamber gas leakage faults, since the component outflow rate equals the sum of the component outlet flow rate and the leakage flow rate in the mass conservation equation, meaning the component loses an additional portion of leakage flow rate, the mathematical expression of its fault model is as follows:

[0021]

[0022] 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.

[0023] As a further optimization of the multivariable fault-tolerant control method for maintaining thrust in a liquid rocket engine according to the present invention, the specific steps of step B) are as follows:

[0024] Step B1), select the control variables as the flow regulator valve core opening α and the fuel main valve opening τ. f The controlled variable is selected as the combustion chamber pressure P, which reflects the thrust. c The selected output constraint variables include the mixing ratio K. c The discretized system of the engine with multiple inputs and multiple outputs is then represented as:

[0025] x(t+1)=f(x(t),u(t))

[0026] Where, x(t)=[P c K c ] T As the output state vector, u(t) = [α τ f ] T Let P be the input control vector, where P is the input control vector. c Let it be denoted as x1(t), K c Let it be denoted as x2(t);

[0027] The neural network prediction model is trained using the current state vector x(t) and control vector u(t), with the output being the state vector x(t+1) at the next time step. The number of neurons in the input layer, hidden layer, and output layer are 4, 20, and 2, respectively. Training utilizes a typical fault model of a liquid rocket engine, generating 25,000 training samples near the rated operating conditions of the main stage, including the flow regulator valve core opening α and the main fuel valve opening τ. f and combustion chamber pressure P c Mixing ratio K c The normalized dataset.

[0028] Step B2), construct the neural network optimization solver equation as follows:

[0029] u(t) = g(x(t), r(t), W)

[0030] Where r(t) is the system command input value at time t, and W is the solver weight matrix; the solver adopts a three-layer network structure, with x(t) and r(t) as input layers and u(t) as output layer, which is used to recursively calculate the predictive control quantity u' and the predictive output quantity x' in the prediction time domain;

[0031] For the quadratic performance metrics of the error vector and control vector:

[0032]

[0033] Where e(t) = x1(t) - r(t), P is the controller prediction time domain, and e(t1 + P) T Ze(t1+P) is the terminal error vector performance index, and Q, R, and Z are the weighted coefficient matrices of the error vector, control vector, and final value error vector in the quadratic performance index functional, respectively, and are taken as diagonal matrices.

[0034] Let e(t) T Qe(t)+u(t) T Ru(t) = L(t), considering that the initial value of the rolling optimization state x(t1) is equal to the final value of the rolling optimization state x'(t1'+P) in the previous stage, the increasing general function of the performance index is derived by introducing λ and q according to the Lagrange multiplier method:

[0035]

[0036] Where λ and q are Lagrange multipliers;

[0037] To obtain the control sequence that minimizes J, we need to recursively solve for λ and q:

[0038] For t = t1 + P:

[0039]

[0040] For t = t1 + P-1, ..., t1 + 2, t1 + 1:

[0041]

[0042]

[0043] Continuously update the weights of the neural network optimization solver until ΔW = 0:

[0044]

[0045] W'=W+ΔW

[0046] Where α is the learning rate and W' is the corrected weight matrix; after the weights are updated, the controller calculates the equations through the neural network optimization solver and applies only the actual control quantity at the current moment to the liquid rocket engine to form rolling optimization.

[0047] Step B3) In the controller solution, the corresponding maximum and minimum value constraint penalty function terms are introduced into the quadratic performance index of the error vector and control vector. The performance index becomes:

[0048]

[0049] At this point, L(t) becomes:

[0050] L(t) = e(t) T Qe(t)+u(t) T Ru(t)+σ1[max(x2(t)-x max ,0)] T [max(x2(t)-x max ,0)]

[0051] +σ2[-min(x2(t)-x min ,0)] T [-min(x2(t)-x min ,0)]

[0052] Where σ1 and σ2 represent the penalty factors for exceeding the upper and lower thresholds, respectively; x max x min K represents the mixing ratio. c The upper and lower limit thresholds are set to x. max =1.08, x min =0.92.

[0053] As a further optimization of the multivariable fault-tolerant control method for maintaining thrust in a liquid rocket engine according to the present invention, the specific steps of step C) are as follows:

[0054] Step C1) Based on the proposed multivariate predictive optimization controller for thrust maintenance of liquid rocket engines, and combined with the established typical fault model, a multivariate predictive optimization fault-tolerant control structure for thrust maintenance is designed to address typical fault occurrences.

[0055] The control structure specifically includes three parts: engine typical fault model establishment, thrust maintenance multivariate prediction optimization controller and liquid rocket engine object. The thrust maintenance multivariate prediction optimization controller mainly includes thrust maintenance optimization solver, prediction model and feedback correction module.

[0056] Given a thrust command from a liquid rocket engine, the thrust maintenance optimization solver calculates the corresponding combustion chamber pressure command. Based on the predicted outputs x'(t+1), x'(t+2), ..., x'(t+P) within P steps given by the prediction model, it performs optimization with the error incorporating a penalty function and the quadratic index J of the control vector, calculating the optimal control sequence u'(t), u'(t+1), ..., u'(t+P-1) within P steps. This sequence is then input into the prediction model, and only the first control quantity u(t) in the sequence, i.e., the control quantity u(t) at the current moment, is applied. Regarding the engine; the liquid rocket engine actuator controls the propellant flow rate based on the flow regulator valve core opening and fuel main valve opening signals provided by the controller. With the typical fault model, the engine output state quantity x is fed back to the thrust maintenance optimization solver and the feedback correction module in real time. The feedback correction module corrects the prediction model by calculating the difference between the actual output x(t+1) and the predicted output x'(t+1), and then performs a new optimization solution, so that the optimization result is not only based on the model, but also utilizes the engine's feedback information.

[0057] Step C2) Under the rated operating conditions of the main stage of the liquid rocket engine, based on the established fault model, the injection of typical faults such as oxygen pre-compression turbine blade ablation, oxygen main pump cavitation, and combustion chamber gas leakage are simulated respectively. With the loop switching PI thrust control method as a reference, the thrust holding fault-tolerant control architecture and control performance are studied and verified.

[0058] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0059] (1) The present invention proposes a multivariable controller for maintaining thrust of a liquid rocket engine, which can be combined with the established typical engine fault model to form a thrust maintenance fault-tolerant control architecture based on a multivariable predictive optimization controller. This control architecture has strong fault tolerance for engine faults and good steady-state and dynamic performance.

[0060] (2) The present invention proposes a multivariable fault-tolerant control method for maintaining thrust of a liquid rocket engine. The mixture ratio is used as an additional constraint. A neural network is used to construct a prediction model and an optimization solver, which can eliminate coupling effects. It has a strong thrust maintenance and mixture ratio limitation effect after a fault occurs, and makes the engine specific impulse higher. Attached Figure Description

[0061] Figure 1 This is a structural diagram of a multivariable fault-tolerant control method for maintaining thrust in a liquid rocket engine.

[0062] Figure 2 This is a schematic diagram illustrating the method for establishing a liquid rocket engine oxygen preload turbine blade ablation failure model.

[0063] Figure 3 This is a schematic diagram illustrating the method for establishing a cavitation failure model of the oxygen main pump in a liquid rocket engine.

[0064] Figure 4 This is a schematic diagram illustrating the method for establishing a fault model of gas leakage in the combustion chamber of a liquid rocket engine.

[0065] Figure 5 This is a training error graph of a neural network prediction model;

[0066] Figure 6 This is a comparison chart of thrust maintenance control effects under oxygen pre-compressed turbine blade ablation failure in a liquid rocket engine. The chart includes (a) flow regulator valve core opening, (b) fuel main valve opening, (c) combustion chamber pressure, (d) thrust, (e) mixture ratio, (f) specific impulse, (g) liquid oxygen main valve flow rate and fuel main valve flow rate, (h) main turbine speed and oxygen main pump post-pump pressure, (i) fuel primary pump and fuel secondary pump post-pump pressure, and (j) gas generator pressure.

[0067] Figure 7 This is a comparison chart of thrust maintenance control effects under cavitation failure of the oxygen main pump in a liquid rocket engine. The chart includes (a) flow regulator valve core opening, (b) fuel main valve opening, (c) combustion chamber pressure, (d) thrust, (e) mixture ratio, (f) specific impulse, (g) liquid oxygen main valve flow rate and fuel main valve flow rate, (h) main turbine speed and oxygen main pump post-pump pressure, (i) fuel primary pump and fuel secondary pump post-pump pressure, and (j) gas generator pressure.

[0068] Figure 8 This is a comparison chart of thrust maintenance control effects under a liquid rocket engine combustion chamber gas leakage fault, where (a) flow regulator valve core opening, (b) fuel main valve opening, (c) combustion chamber pressure, (d) thrust, (e) mixture ratio, (f) specific impulse, (g) liquid oxygen main valve flow rate and fuel main valve flow rate, (h) main turbine speed and oxygen main pump post-pump pressure, (i) fuel primary pump and fuel secondary pump post-pump pressure, and (j) gas generator pressure. Detailed Implementation

[0069] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings.

[0070] The specific implementation of this invention takes the design of a multivariable fault-tolerant control method for thrust maintenance of a certain type of liquid oxygen / kerosene staged combustion cycle liquid rocket engine in the Long March series carrier rockets as an example. Its control structure is as follows: Figure 1 As shown, the specific steps include:

[0071] Step A) Based on the typical failure modes of liquid rocket engines, determine the failure mechanisms and characteristic parameters of oxygen pre-compression turbine blade ablation, oxygen main pump cavitation, and combustion chamber gas leakage, and introduce failure factors to simulate failures for different typical failures, and establish a typical failure model of liquid rocket engines.

[0072] Step B) Train a neural network prediction model with a dataset consisting of the output state vector and the input control vector obtained from the typical fault model of liquid rocket engine. The output state vector includes the mixing ratio. Based on the quadratic index of error and control quantity, a penalty function term that limits the mixing ratio is introduced to perform neural network optimization solution for thrust maintenance and design a thrust maintenance multivariate prediction optimization controller that ensures high safety.

[0073] Step C) Design a thrust holding fault-tolerant control architecture based on a multivariable predictive optimization controller. Under the rated conditions of the main stage, simulate the occurrence of typical faults such as oxygen pre-compression turbine blade ablation, oxygen main pump cavitation, and combustion chamber gas leakage in a liquid rocket engine, and study and verify the thrust holding fault-tolerant control method and control performance.

[0074] The detailed steps of step A) are as follows:

[0075] Step A1) Select turbine blade ablation, pump cavitation, and thermal component gas leakage failures occurring in the turbopump system and thermal component system from the liquid rocket engine common failure mode library for study. Analyze the failure mechanism and influencing parameter characteristics. The results are shown in the table below:

[0076] Table 1. Characteristics of Influence Parameters for Typical Failure Modes of Liquid Rocket Engines

[0077]

[0078] As shown in the table, the characteristic parameters affecting the oxygen preload turbine blade ablation fault are a decrease in turbine efficiency and speed; the characteristic parameters affecting the oxygen main pump cavitation fault are a decrease in pump head and downstream pressure; and the characteristic parameters affecting the combustion chamber gas leakage fault are an increase in component outflow and a decrease in component pressure. Based on the impact of each typical fault mode on engine parameter changes, a typical engine fault model is established.

[0079] Step A2) is to establish a typical engine fault model by introducing a fault factor parameter F to correct the mathematical model parameter D of the engine components. F represents the severity of the fault in the engine components.

[0080] For oxygen preloaded turbine blade ablation failure, based on the analysis of failure impact parameters, due to turbine efficiency η t The failure model is established as follows: Figure 2As shown, the mathematical description of the fault model is:

[0081]

[0082] 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.

[0083] For the cavitation failure of the oxygen main pump, based on the characteristic analysis of the failure impact parameters, since the pump head ΔP decreases, the failure model is established as follows: Figure 3 As shown, the mathematical description of the fault model is:

[0084]

[0085] 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.

[0086] For combustion chamber gas leakage faults, based on the fault impact parameter characteristic analysis, since the component outflow rate in the mass conservation equation equals the sum of the component outlet flow rate and the leakage flow rate, meaning the component loses an additional portion of leakage flow, the fault model is established as follows: Figure 4 As shown, the mathematical description of the fault model is:

[0087]

[0088] 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.

[0089] The detailed steps of step B) are as follows:

[0090] Step B1) Adjust the engine flow regulator valve core opening α and the fuel main valve opening τ. f As a control variable; the combustion chamber pressure P, which reflects the thrust. c As a controlled variable; mixing ratio Kc As output constraint variables, for a multi-input, multi-output discretized engine system:

[0091] x(t+1)=f(x(t),u(t))

[0092] Where, x(t)=[P c K c ] T As the output state vector, u(t) = [α τ f ] T Let P be the input control vector, where P is the input control vector. c Let it be denoted as x1(t), K c Let it be denoted as x2(t);

[0093] The neural network prediction model training adopts a single hidden layer feedforward BP neural network structure. The network input is selected from the current state vector x(t) and control vector u(t) in the above formula, and the output is the state vector x(t+1) at the next time step. The number of neurons in the input layer, hidden layer and output layer are 4, 20 and 2, respectively. The training first uses a typical fault model of liquid rocket engine to generate flow regulator valve core opening α and fuel main valve opening τ with 25,000 training samples near the rated state of the main stage. f and combustion chamber pressure P c Mixing ratio K c The normalized dataset was used; the learning rate was set to 0.05 and the momentum coefficient to 0.1 for training, and the neural network training error was as follows: Figure 5 As shown in the figure, the training error reaches 9.985 × 10⁻⁶ after 378 iterations. -9 They believe that neural network training converges;

[0094] Step B2), establish the neural network optimization solver equation as follows:

[0095] u(t) = g(x(t), r(t), W)

[0096] Where r(t) is the system command input value at time t, and W is the solver weight matrix; the solver adopts a single hidden layer feedforward BP neural network structure, with input layers x(t) and r(t) and output layer u(t), used to recursively calculate the predicted control quantity u' and predicted output quantity x' in the prediction time domain;

[0097] For the quadratic performance metrics of the error vector and control vector:

[0098]

[0099] Where e(t) = x1(t) - r(t), P is the controller prediction time domain, and e(t1 + P) TZe(t1+P) is the terminal error vector performance index, and Q, R, and Z are the weighted coefficient matrices of the error vector, control vector, and final value error vector in the quadratic performance index functional, respectively, and are taken as diagonal matrices.

[0100] Let e(t) T Qe(t)+u(t) T Ru(t) = L(t), considering that the initial value of the rolling optimization state x(t1) is equal to the final value of the rolling optimization state x'(t1'+P) in the previous stage, the increasing general function of the performance index is derived by introducing λ and q according to the Lagrange multiplier method:

[0101]

[0102] Where λ and q are Lagrange multipliers;

[0103] To obtain the control sequence that minimizes J, we need to recursively solve for λ and q:

[0104] For t = t1 + P:

[0105]

[0106] For t = t1 + P-1, ..., t1 + 2, t1 + 1:

[0107]

[0108]

[0109] Continuously update the weights of the neural network optimization solver until ΔW = 0:

[0110]

[0111] W'=W+ΔW

[0112] Where α is the learning rate and W' is the corrected weight matrix; after the weights are updated, the controller calculates the equations through the neural network optimization solver and applies only the actual control quantity at the current moment to the liquid rocket engine to form rolling optimization.

[0113] Step B3), in order to handle the output constraint, make the mixing ratio K c The predicted output is constrained by a threshold in the prediction time domain. In the controller solution, the corresponding maximum and minimum value constraint penalty function terms are introduced into the quadratic performance index of the error vector and control vector. The performance index then becomes:

[0114]

[0115] At this point, L(t) becomes:

[0116] L(t) = e(t) T Qe(t)+u(t) T Ru(t)+σ1[max(x2(t)-x max ,0)] T [max(x2(t)-x max ,0)]+σ2[-min(x2(t)-x min ,0)] T [-min(x2(t)-x min ,0)]

[0117] Where σ1 and σ2 represent the penalty factors for exceeding the upper and lower thresholds, respectively, and are large positive constants, typically two orders of magnitude larger than other weighting coefficients; x max x min K represents the mixing ratio. c The upper and lower limit thresholds are determined by taking x based on the adjustable range of the mixture ratio in the liquid rocket engine. max =1.08, x min =0.92. After introducing a penalty function term to the performance index, at each time point in the entire prediction time domain, if the predicted output is within the limit threshold, the penalty function term is 0, and it has no impact on the optimization of the performance index J; if the predicted output is outside the limit threshold, the penalty function term increases significantly, causing the performance index J to increase significantly. The controller aims to find the control quantity that minimizes the performance index J, thereby limiting the output to within its threshold. Thus, the controller can monitor the mixing ratio K in real time. c Limiting control is implemented, meaning that when the controller receives a trend signal that the mixture ratio exceeds the limit threshold, it will calculate the control quantity to avoid the mixture ratio exceeding the threshold, so that the mixture ratio deviates little from the design value, thus ensuring the engine's performance stability and operational safety.

[0118] The detailed steps of step C) are as follows:

[0119] Step C1) Based on the proposed multivariate predictive optimization controller for thrust maintenance in liquid rocket engines, and combined with the established typical fault model, a multivariate predictive optimization fault-tolerant control structure for thrust maintenance is designed to address typical fault occurrences, such as... Figure 1 As shown;

[0120] As shown in the figure, the thrust holding fault-tolerant control architecture based on the multivariable predictive optimization controller consists of three main parts: the establishment of typical engine fault models, the thrust holding multivariable predictive optimization controller, and the liquid rocket engine object. The thrust holding multivariable predictive optimization controller mainly includes a thrust holding optimization solver, a prediction model, and a feedback correction module.

[0121] Given a thrust command from a liquid rocket engine, the thrust maintenance optimization solver calculates the corresponding combustion chamber pressure command. Based on the predicted outputs x'(t+1), x'(t+2), ..., x'(t+P) within P steps given by the prediction model, it performs optimization with the error incorporating a penalty function and the quadratic index J of the control vector, calculating the optimal control sequence u'(t), u'(t+1), ..., u'(t+P-1) within P steps. This sequence is then input into the prediction model, and only the first control quantity u(t) in the sequence, i.e., the control quantity u(t) at the current moment, is applied. Regarding the engine; the liquid rocket engine actuator controls the propellant flow rate based on the flow regulator valve core opening and fuel main valve opening signals provided by the controller. With the typical fault model, the engine output state quantity x is fed back to the thrust maintenance optimization solver and the feedback correction module in real time. The feedback correction module corrects the prediction model by calculating the difference between the actual output x(t+1) and the predicted output x'(t+1), and then performs a new optimization solution, so that the optimization result is not only based on the model, but also utilizes the engine's feedback information.

[0122] Step C2), under the rated operating conditions of the main stage of the liquid rocket engine, based on the established fault model, such as Figure 2 , Figure 3 , Figure 4 The injection of typical faults such as oxygen preload turbine blade ablation, oxygen main pump cavitation, and combustion chamber gas leakage is shown in the figure. With the loop switching PI thrust control method as a reference, the thrust holding fault-tolerant control architecture and control performance are studied and verified.

[0123] The loop-switching PI control method used in the comparison adopts a distributed loop to form a multivariable control system, which consists of a thrust PI single-variable control loop and a mixture ratio PI single-variable control loop. The loop switching logic is that when the mixture ratio exceeds the limit, the thrust single-variable control loop is switched to the thrust and mixture ratio distributed control loop.

[0124] like Figure 2 The simulation results show the ablation failure of the turbine blades during oxygen injection preloading in the engine at 60s. The failure factor F1 is set to 0.9. Figure 6 As shown in Table 2, the performance indicators are as follows. It can be seen that after the engine experiences an oxygen preload turbine blade ablation failure, both this invention and loop switching control can reduce the steady-state thrust loss under fault conditions. However, since the mixture ratio is not exceeded, loop switching control only uses a flow regulator to control the thrust as a single variable. In contrast, this invention optimizes based on the thrust maintenance target, eliminating coupling effects and solving for two control variables as engine inputs. Therefore, it suffers less transient thrust loss, requires less time to recover to the rated value, and has a smaller absolute value of the maximum change in the mixture ratio, resulting in a higher specific impulse.

[0125] Table 2 Comparison of thrust retention effects under typical faults in the main stage of a certain type of liquid rocket engine under rated operating conditions.

[0126]

[0127] like Figure 3 The simulation results show that at 60s, the main oxygen injection pump of the engine experiences cavitation failure. The failure factor F2 is set to 0.95, and the simulation results are as follows: Figure 7 As shown in Table 2, the performance index results are as follows. It can be seen that after the main oxygen pump cavitation failure, both the present invention and the loop switching control can reduce the steady-state thrust loss under fault conditions. The loop switching control switches to a limiting loop to control the mixture ratio after it exceeds a threshold. The present invention, through online rolling optimization with mixture ratio constraints, calculates a control output that is more optimal for thrust maintenance and mixture ratio limitation, which is then applied to the engine. Therefore, the transient thrust loss is smaller, the thrust recovery is faster, the mixture ratio always maintains a distance from the boundary, the absolute value of the maximum change in mixture ratio is smaller, and the specific impulse is also higher.

[0128] like Figure 4 The simulation results for a combustion chamber gas leakage fault in the engine at 60s are shown below. The fault factor F3 is set to 0.3. Figure 8 As shown in Table 2, the performance index results are as follows. It can be seen that after the engine experiences a combustion chamber gas leakage fault, the proposed control method demonstrates strong fault tolerance when the engine experiences a severe thrust decline fault. This is because the present invention uses a neural network to construct the prediction model and optimization solver. The neural network's robustness, feedback correction of the prediction model based on the neural network, and the rolling implementation of the optimization solver enable timely compensation for model mismatch and uncertainties caused by external disturbances. Specifically, compared to loop switching control, the present invention exhibits smaller transient and steady-state thrust losses, a smaller maximum deviation of the mixture ratio from the design value, and a higher specific impulse.

[0129] Therefore, thanks to the establishment of the neural network prediction model and the optimization solver, as well as the constrained online rolling optimization strategy, the multivariable fault-tolerant control method for maintaining thrust of liquid rocket engines proposed in this invention is more effective than conventional methods in maintaining thrust under typical engine failures. It reduces the transient thrust loss by 10% to 40%, and makes the deviation between the engine's mixture ratio and the design value smaller, resulting in a higher specific impulse. It can achieve performance maintenance and operational safety under typical engine failures.

[0130] This invention establishes a typical engine failure model based on the influence parameter characteristic analysis of typical failure modes of liquid rocket engines; using the combustion chamber pressure P, which reflects thrust, as an example. c The controlled variable is the mixing ratio K. cAs constrained quantities, the flow regulator valve core and the fuel main valve opening α, τ f To control the input, a thrust-holding multivariate predictive optimization controller was established through offline training of a neural network prediction model and optimization of the control input based on the error with a penalty function and a quadratic index of the control input using a neural network. By combining a typical fault model with the controller, a multivariate fault-tolerant control architecture for thrust holding was proposed. Simulations show that this invention eliminates the effects of coupling, achieves superior thrust holding and mixture ratio limiting effects under fault conditions, minimizes transient and steady-state thrust losses, demonstrates strong fault tolerance, and results in a relatively high engine specific impulse.

[0131] 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 of multivariable fault-tolerant control of liquid rocket engine thrust holding, characterized in that, The method comprises the following steps: Step A), according to the typical failure mode of the liquid rocket engine, the characteristic parameters of the oxygen pre-pressurization turbine blade ablation, the oxygen main pump cavitation, the combustion chamber gas leakage failure mechanism and the failure influence are determined, the failure factor is introduced to simulate the failure for different typical failures, and the typical failure model of the liquid rocket engine is established; Step B), a neural network prediction model is trained by using a data set composed of an output state vector and an input control vector obtained according to the typical failure model of the liquid rocket engine, the output state vector comprises a mixture ratio, on the basis of a quadratic form index of an error and a control amount, a penalty function term for limiting the mixture ratio is introduced, a neural network optimization solver is solved in the direction of thrust holding, and a thrust holding multivariable prediction optimization controller with high safety is designed; the specific steps are as follows: Step B1), the control variable is selected as the flow regulator valve core opening α, the fuel main valve opening τ f ; the controlled variable is selected as the combustion chamber pressure P reflecting the thrust c ; the output constraint variable includes the mixture ratio K c , and the engine multi-input and multi-output discrete system is represented as: x (t + 1) = f (x (t), u (t) ) where x(t) = [P c K c ] T as output state vector, u(t) = [a τ f ] T as input control vector, where P c denoted as x1(t), K c denoted as x2(t); The neural network prediction model selects the current time state vector x(t) and control vector u(t) as input, and outputs the next time state vector x(t+1); the training utilizes the typical fault model of liquid rocket engine to generate a normalized dataset containing several groups of training samples of flow regulator valve core opening α, fuel main valve opening τ f and combustion chamber pressure P c , mixture ratio K c near the rated state of the main stage working condition. Step B2), the equation of the neural network optimization solver is constructed as u (t) = g (x (t), r (t), W) Wherein, r (t) is the system instruction input value at t time, and W is the solver weight matrix; the solver adopts a single hidden layer feedforward BP neural network structure, the input layer is x (t) and r (t), and the output layer is u (t), which is used to recursively calculate the predicted control amount u' and the predicted output x' in the prediction time domain; For the quadratic form performance index of the error vector and the control vector: where e(t) = x1(t) - r(t), P is the prediction horizon of the controller, e(t1+P) T Ze(t1+P) is the terminal error vector performance index, Q, R, Z are the error vector, control vector and terminal error vector weighting coefficient matrix in the quadratic performance index functional, respectively, and are taken as diagonal matrices; Let e(t) T Qe(t) + u(t) T Ru(t) = L(t), considering the rolling optimization state initial value x(t1) is equal to the last stage rolling optimization state final value x'(t1' + P), according to the Lagrange multiplier method, λ and q are introduced to obtain the performance index augmented functional: Wherein, λ and q are Lagrange multipliers; In order to obtain the control sequence that makes J as small as possible, λ and q need to be recursively solved: For t = t1 + P: For t = t1 + P-1,…, t1 + 2, t1 + 1: The weight of the neural network optimization solver is updated constantly until ΔW = 0: W' = W + ΔW Wherein, α is the learning rate, and W' is the corrected weight matrix; after the weight update is completed, the controller calculates through the neural network optimization solver equation, only the actual control amount at the current time is applied to the liquid rocket engine, and rolling optimization is formed; Step B3), the corresponding maximum and minimum value constraint penalty function term is introduced into the error vector and the control vector quadratic form performance index in the controller solving, and the performance index becomes: At this time, L (t) becomes: L(t) = e(t) T Qe(t) + u(t) T Ru(t) + σ1[max(x2(t) - x T , 0)] max [max(x2(t) - x min , 0)] + σ2[-min(x2(t) - x min , 0)] T - min(x2(t) - x min , 0)] wherein σ1, σ2 represent the penalty factors for exceeding the upper threshold and the lower threshold, respectively; x max , x min represent the upper limit threshold and the lower limit threshold for the mixing ratio K c , respectively. Step C), a thrust holding fault-tolerant control architecture based on the multivariable prediction optimization controller is designed, the occurrence of the typical failure of the oxygen pre-pressurization turbine blade ablation, the oxygen main pump cavitation and the combustion chamber gas leakage of the liquid rocket engine is simulated under the rated state of the main stage working condition, and the thrust holding fault-tolerant control method and control performance are researched and verified.

2. A multivariable fault-tolerant control method for liquid rocket engine thrust holding according to claim 1, characterized in that, The failure influence parameter characteristics of the oxygen pre-pressurization turbine blade ablation are the decrease of the turbine efficiency and the rotating speed, the failure influence parameter characteristics of the oxygen main pump cavitation are the decrease of the pump head and the pump pressure, and the failure influence parameter characteristics of the combustion chamber gas leakage are the increase of the outflow of the component and the decrease of the component pressure.

3. A multivariable fault-tolerant control method for liquid rocket engine thrust holding according to claim 2, characterized in that, The engine typical failure model is established by introducing the failure factor parameter F to modify the engine component mathematical model parameter D, and F represents the severity of the failure of the engine component; For the oxygen pre-pressurization turbine blade ablation failure, since the turbine efficiency η t occurs, the failure model is mathematically described as: wherein η t (t) is the normal value of turbine efficiency, η t ′(t) is the actual value of turbine efficiency, t is the running time, t f is the time of failure, F1 is the turbine failure factor, when 0≤F1<1, it represents that the turbine has blade ablation failure; when F1=1, the turbine works normally; For the oxygen main pump cavitation failure, since the pump head ΔP decreases, the failure model is mathematically described as: Wherein, ΔP(t) is the normal value of pump head, ΔP'(t) is the actual value of pump head, F2 is the pump failure factor, when 0≤F2<1, it represents that the pump has cavitation failure; when F2=1, the pump works normally; For the combustion chamber gas leakage fault, since the component outflow in the mass conservation equation is equal to the sum of the component outlet flow and the leakage flow, i.e. the component loses a part of the leakage flow, the fault model is mathematically described as: wherein q ig (t), q lo (t), q lf (t) are the normal values of the mass flow rates of the fuel gas, the liquid oxidizer and the liquid fuel flowing into the thermal component, respectively, q eg (t) is the normal value of the mass flow rate of the fuel gas flowing out of the thermal component, is the actual value of the mass change rate of the working medium of the thermal component, and F3 is a fault factor of the thermal component. When F3 > 0, it indicates that the thermal component has a fuel gas leakage fault. When F3 = 0, the thermal component is working normally.

4. A multivariable fault-tolerant control method for liquid rocket engine thrust holding according to claim 1, characterized in that, The specific steps of step C) are as follows: Step C1), on the basis of the proposed liquid rocket engine thrust holding multivariable predictive optimization controller, combined with the established typical fault model, a thrust holding multivariable predictive optimization fault-tolerant control structure for typical fault occurrence is designed. The control structure specifically includes three parts of engine typical fault model establishment, thrust holding multivariable predictive optimization controller and liquid rocket engine object, wherein the thrust holding multivariable predictive optimization controller includes a thrust holding optimization solver, a prediction model and a feedback correction module. By giving a liquid rocket engine thrust command, the thrust holding optimization solver solves the corresponding combustion chamber pressure command, and according to the P-step prediction output x'(t+1), x'(t+2), …, x'(t+P) given by the prediction model, the error of the penalty function and the quadratic index J of the control vector are optimized and solved to calculate the optimal control amount sequence u'(t), u'(t+1), …, u'(t+P-1) within P steps, which is input into the prediction model, and only the first control amount u(t) in the control amount sequence, i.e. the control amount at the current time, acts on the engine; the liquid rocket engine actuator realizes the regulation and control of the engine propellant flow according to the flow regulator valve core opening degree and fuel main valve opening degree signals given by the controller; under the access of the typical fault model, the engine output state x is fed back to the thrust holding optimization solver and the feedback correction module in real time; wherein the feedback correction module corrects the prediction model by calculating the difference between the actual output x(t+1) and the prediction output x'(t+1), and then performs new optimization and solving, so that the optimization result is not only based on the model, but also uses the feedback information of the engine; Step C2), under the rated state of the main stage working condition of the liquid rocket engine, based on the established fault model, the injection of typical faults such as oxygen pre-pressing turbine blade ablation, oxygen main pump cavitation and combustion chamber gas leakage is simulated, and the thrust holding fault-tolerant control architecture and control performance are researched and verified with the loop switching PI thrust control method as a reference.

Images