A method, device and medium for diagnosing rocket engine faults

Abstract

This invention discloses a method, apparatus, and medium for diagnosing rocket engine faults, relating to the field of rocket engine technology, to address the problem of low accuracy in existing engine fault diagnosis methods. The rocket engine fault diagnosis method includes: acquiring engine measurement data; inputting initial measured values ​​of covariates into a first recurrent neural network to obtain predicted values ​​of the covariates at each time step; inputting the predicted values ​​of the covariates at each time step into a second recurrent neural network to obtain hidden layer information at the corresponding time step; mapping the hidden layer information to a subarray of the state transition matrix of an augmented state-space model of the rocket engine to obtain a rocket engine fault model; and based on the measured values ​​of target parameters, using an observer to observe the fault factors in the rocket engine fault model to obtain estimated values ​​of the fault factors, thereby achieving fault diagnosis of the engine. The rocket engine fault diagnosis method provided by this invention is used to improve the accuracy of fault diagnosis for liquid rocket engines.

Description

Technical Field

[0001] This invention relates to the field of rocket engine technology, and in particular to a method, apparatus and medium for diagnosing rocket engine faults. Background Technology

[0002] Liquid rocket engines are among the most prone to failure on launch vehicles, and malfunctions severely impact launch reliability. The operation of a liquid rocket engine can be divided into two phases: the unsteady phase and the steady-state phase. The steady-state phase refers to when engine controls cease operation and the engine operates under predetermined conditions.

[0003] Existing engine fault diagnosis methods observe fault factors by establishing a state-space model containing fault factors. The parameters representing the state in the observed model remain unchanged. However, due to the inherent characteristics of the engine, after startup and entering the steady-state phase, the important parameters of the engine exhibit trend-like changes. At this time, using the method of observing a pre-determined model will lead to incorrect diagnostic results and cannot meet the higher requirements for launch reliability of major space launch missions. Summary of the Invention

[0004] The purpose of this invention is to provide a method, apparatus and medium for diagnosing rocket engine faults, so as to improve the accuracy of fault diagnosis of liquid rocket engines.

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

[0006] On one hand, the present invention provides a method for diagnosing rocket engine faults, comprising:

[0007] Acquire engine measurement data; the measurement data includes initial measurement values ​​of covariates and measurement values ​​of target parameters at various time points;

[0008] The initial measured values ​​of the covariates are input into the first recurrent neural network to obtain the predicted values ​​of the covariates at each time step.

[0009] The predicted values ​​of the covariates at each time step are input into the second recurrent neural network to obtain the hidden layer information at the corresponding time step.

[0010] The hidden layer information is mapped to a subarray of the state transition matrix of the rocket engine augmented state space model to obtain the rocket engine fault model; the rocket engine augmented state space model is a model with fault factors as system state parameters.

[0011] Based on the measured values ​​of the target parameters, an observer is used to observe the fault factors in the rocket engine fault model, and the estimated values ​​of the fault factors are obtained, thereby realizing the fault diagnosis of the engine.

[0012] Compared with existing technologies, the rocket engine fault diagnosis method provided by this invention avoids data contamination caused by faults by inputting the initial measured values ​​of covariates into a first recurrent neural network to obtain predicted values ​​of covariates at each time step. Simultaneously, the predicted values ​​of covariates at each time step are input into a second recurrent neural network to obtain hidden layer information at the corresponding time step. By mapping the hidden information to a subarray of the state transition matrix of the rocket engine augmented state space model, a rocket engine fault model is obtained. Since the hidden information is determined in real time, the linear relationship describing the parameter changes under normal engine conditions can be determined in real time. The rocket engine augmented state space model can reflect the relationship between engine faults and target parameters; therefore, the obtained rocket engine fault model has the ability to describe both steady-state state parameter changes and engine faults. Finally, an observer is used to observe the fault factors in the rocket engine fault model and estimate the engine fault factors in real time, thereby achieving engine fault diagnosis. This method determines the linear relationship of parameter changes under normal conditions in the steady-state stage in real time, providing a more reliable fault model for fault diagnosis, improving the accuracy of fault diagnosis, and thus meeting the higher requirements for launch reliability in space launch missions.

[0013] Secondly, the present invention provides a rocket engine fault diagnosis device, comprising:

[0014] An engine measurement data acquisition module is used to acquire engine measurement data; the measurement data includes initial measurement values ​​of covariates and measurement values ​​of target parameters at various times.

[0015] The covariate prediction module is used to input the initial measurement value of the covariate into the first recurrent neural network to obtain the predicted value of the covariate at each time step.

[0016] The hidden layer information calculation module is used to input the predicted value of the covariate at each time step into the second recurrent neural network to obtain the hidden layer information at the corresponding time step.

[0017] The rocket engine fault model establishment module is used to map the hidden layer information into a subarray of the state transition matrix of the rocket engine augmented state space model to obtain the rocket engine fault model.

[0018] The fault factor observation module is used to observe the fault factors in the rocket engine fault model based on the measured values ​​of the target parameters, obtain the estimated values ​​of the fault factors, and realize the fault diagnosis of the engine.

[0019] Compared with the prior art, the beneficial effects of the rocket engine fault diagnosis device provided by the present invention are the same as the beneficial effects of the rocket engine fault diagnosis method described in the above technical solution, and will not be repeated here.

[0020] Thirdly, the present invention provides a computer-readable storage medium, comprising: instructions stored in the computer-readable storage medium, wherein when the instructions are executed, the above-mentioned rocket engine fault diagnosis method is implemented. Attached Figure Description

[0021] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0022] Figure 1 A flowchart of a rocket engine fault diagnosis method provided by the present invention;

[0023] Figure 2 This is a schematic diagram of the first initial recurrent neural network training provided by the present invention;

[0024] Figure 3 This is a schematic diagram of the first recurrent neural network prediction provided by the present invention;

[0025] Figure 4 This is a schematic diagram of the second initial recurrent neural network training provided by the present invention;

[0026] Figure 5 This is a schematic diagram of the second recurrent neural network prediction provided by the present invention;

[0027] Figure 6 This is a schematic diagram of the structure of a rocket engine fault diagnosis device provided by the present invention. Detailed Implementation

[0028] To facilitate a clear description of the technical solutions in the embodiments of the present invention, the terms "first" and "second" are used to distinguish identical or similar items with essentially the same function and effect. For example, the first threshold and the second threshold are merely used to distinguish different thresholds and do not limit their order. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that the terms "first" and "second" are not necessarily different.

[0029] It should be noted that in this invention, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0030] In this invention, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can represent: a, b, c, a combination of a and b, a combination of a and c, a combination of b and c, or a, b, and c, where a, b, and c can be single or multiple.

[0031] Before introducing the embodiments of the present invention, the relevant terms involved in the embodiments of the present invention are first defined as follows:

[0032] The extended Kalman filter, also known as an autoregressive filter, is a highly efficient recursive filter that can estimate the state of a dynamic system from a series of measurements that are not completely free of noise.

[0033] For liquid rocket engines, timely identification of engine faults through fault diagnosis algorithms and the implementation of fault-tolerant control measures can reduce the impact of engine faults on launch missions. Existing observer-based fault diagnosis methods generally combine measurement results with state-space models to observe fault factors characterizing the fault. When applying this method to fault diagnosis of liquid rocket engines, the model is predetermined, but the engine state is constantly changing, which can lead to inaccurate or even erroneous diagnostic results.

[0034] To address the aforementioned problems, this invention provides a rocket engine fault diagnosis method, device, and medium. It improves upon traditional modeling methods by employing a neural network to determine a model representing engine parameter changes under normal conditions in real time. Simultaneously, based on a system dynamics model of the injected fault, an augmented state-space model of the rocket engine representing the fault's impact is determined. These two models are fused to determine a rocket engine fault model representing the engine's linear state at each moment in real time. Furthermore, an observer is introduced to achieve online observation and fault diagnosis of fault factors. A detailed description follows with reference to the accompanying drawings.

[0035] Figure 1 A flowchart of a rocket engine fault diagnosis method provided by the present invention is shown below. Figure 1 As shown, it includes the following steps:

[0036] Step 101: Obtain engine measurement data.

[0037] The measurement data includes initial measurements of covariates and measurements of target parameters at various time points. The initial measurements of covariates are taken at the first moment after the engine enters steady state. All measurements are performed by corresponding sensors. Both covariates and target parameters include one or more parameters, and there is a mapping relationship between covariates and target parameters. Covariates are used to generate the state-space equations, while target parameters are parameters related to engine faults. The measurement data may contain irrelevant quantities, and the measurement results from such sensors need to be removed. To avoid data contamination, only the initial measurements of covariates are needed; the target parameters, as the system output parameters of the rocket engine fault model, need to be measured in real time.

[0038] Step 102: Input the initial measurement values ​​of the covariates into the first recurrent neural network to obtain the predicted values ​​of the covariates at each time step;

[0039] As an alternative approach, to obtain the first recurrent neural network, it is necessary to train the recurrent neural network based on engine normal test data during the steady-state phase. This can be combined with... Figure 2 Explanation:

[0040] Specifically, such as Figure 2 As shown, the first initial recurrent neural network is first established;

[0041] Then, the measured value Z of the covariate at time n-1 in the engine normal test data during the steady-state phase is obtained. n-1 The input is fed into the first initial recurrent neural network to predict the value of the covariate at time n. The measured value Z of the covariate at time n n The input is fed into the initial recurrent neural network to predict the value of the covariate at time n+1. The measured value Z of the covariate at time n+1 n+1 The input is fed into the initial recurrent neural network to predict the value of the covariate at time n+2. And so on. For ease of distinction, the predicted values ​​of the covariates obtained in this stage are called the first predicted values.

[0042] Then, the measured value Z of the covariate at time n in the engine normal test data during the steady-state phase is obtained. n Compared with the predicted value For comparison, the measured value Z of the covariate at time n+1 is used. n+1 Compared with the predicted value For comparison, the measured value Z of the covariate at time n+2 is used. n+2 Compared with the predicted value The parameters of the first initial recurrent neural network are adjusted based on the comparison results until the training is completed and the first recurrent neural network is obtained.

[0043] When making predictions using the first recurrent neural network, only the initial measured values ​​of the covariates need to be input to obtain the predicted values ​​of the covariates at each time step.

[0044] Specifically, in combination Figure 3 Please provide an explanation, such as Figure 3 As shown, the initial measured value Z0 of the covariate is input into the first RNN neuron in the first recurrent neural network to obtain the predicted value of the covariate at the first time step. Predicted values The input is fed into the next RNN neuron to predict the value of the covariate at the second time step. Predicted values The input is fed into the next RNN neuron to predict the value of the covariate at time 3. This process continues, with the prediction result of the previous RNN neuron being input into the next RNN neuron to predict the covariates at the next time step, thus obtaining the predicted values ​​of the covariates at each time step.

[0045] Step 103: Input the predicted value of the covariate at each time step into the second recurrent neural network to obtain the hidden layer information at the corresponding time step.

[0046] As an alternative approach, to obtain the second recurrent neural network, it is necessary to train the recurrent neural network based on the engine's normal test data during the steady-state phase and the rocket engine's state-space model under normal operating conditions.

[0047] Specifically, firstly, a second initial recurrent neural network and a state-space model of the rocket engine under normal operating conditions are established;

[0048] The state-space model of a rocket engine under normal operating conditions is shown in equations (1) and (2):

[0049] x k+1 =A·x k +JJ+e k (1)

[0050] y k =C·x k +KK+ε k (2)

[0051] Where k is the time step, and the value of k is 1, 2, 3, ..., x k y is a system state parameter. k For system output parameters, e k and ε k For an uncorrelated zero-mean white noise sequence, e k The covariance matrix is ​​Q,ε kThe covariance matrix is ​​R, and the JJ and KK matrices are the compensation matrices. Let x∈y, then the C matrix is ​​known. Therefore, the A matrix represents the changes in state parameters, i.e., the state transition.

[0052] To describe the time series using a state-space model, the hidden layer information output by the second initial recurrent neural network is mapped to the state transition matrix A of the rocket engine state-space model. Simultaneously, to prevent data corruption caused by engine failure, the "update" step of Kalman filtering is omitted; that is, the predicted values ​​of the system state parameters at the next time step are predicted solely based on the measured values ​​of the rocket engine state-space model system state parameters from the previous time step in the Kalman filtering technique. The following section combines... Figure 4 Please provide a detailed explanation.

[0053] like Figure 4 As shown, the measured value Z of the covariate at time n-1 in the steady-state normal test data is... n-1 The information is input into the second initial recurrent neural network to obtain the initial hidden layer information at time n-1; this initial hidden layer information is then mapped to the initial state transition matrix A of the rocket engine state space model. n-1 Simultaneously, the measured value x of the target parameter at time n-1 is... n-1 The values ​​are input into the rocket engine state-space model to obtain the predicted values ​​of the target parameters at time n. The predicted value of the target parameter at time n The measured value x of the target parameter at time n n A comparison was made, and the parameters of the second initial recurrent neural network were adjusted based on the comparison results; then, the measured value Z of the covariate at time n in the steady-state normal test data was obtained. n The information is input into the second initial recurrent neural network to obtain the initial hidden layer information at time n; this initial hidden layer information is then mapped to the initial state transition matrix A of the rocket engine state-space model. n Simultaneously, the measured value x of the target parameter at time n n The values ​​are input into the rocket engine state-space model to obtain the predicted values ​​of the target parameters at time n+1. The predicted value of the target parameter at time n+1 The measured value x of the target parameter at time n+1 n+1 A comparison is performed, and the parameters of the second initial recurrent neural network are adjusted based on the comparison results until the adjusted parameters meet the requirements, resulting in a trained second recurrent neural network. Here, n is a positive integer. For ease of distinction, the predicted values ​​of the target parameters are referred to as the second predicted values.

[0054] The prediction processes of the trained first and second recurrent neural networks can be combined. Figure 5 Please provide an explanation, such as Figure 5 As shown: The initial measured value Z0 of the covariate is input into the first recurrent neural network to predict the predicted value of the covariate. On the one hand, the predicted value As the input to the next RNN neuron in the first recurrent neural network, the predicted values ​​of the covariates are obtained. On the other hand, the predicted value The input is fed into the second recurrent neural network, and the output of the second recurrent neural network can be mapped to the state transition matrix A1. Thus, by inputting the predicted value of the slope variable at each time step obtained by the first recurrent neural network into the second recurrent neural network, the hidden layer information at the corresponding time step can be obtained, and this hidden layer information can be mapped to the state transition matrix.

[0055] Step 104: Map the hidden layer information to a subarray of the state transition matrix of the rocket engine augmented state space model to obtain the rocket engine fault model.

[0056] Specifically, the first step is to establish an object-oriented rocket engine model;

[0057] An object-oriented rocket engine model can be built using Simulink software. Specifically, based on the system dynamics equations, subsystem component-level models of the engine are established, including models of centrifugal pumps, turbines, piping, gas generators, thrust chambers, control systems, and physical property models. Then, these subsystem component-level models are connected to obtain the object-oriented rocket engine model.

[0058] In the object-oriented rocket engine model, fault factors representing the component-level models of each subsystem are constructed and linearized to obtain a discrete linearized state-space model of the rocket engine after fault injection.

[0059] Fault injection modifies an object-oriented rocket engine model to enable it to simulate faults. The principle of fault injection is to inject common parameters that can significantly affect the engine's normal operation and state parameters. Fault injection can be achieved through multiplicative and additive fault methods based on fault mode analysis.

[0060] The discrete linearized state-space model of the rocket engine after fault injection is shown in equations (3) and (4):

[0061] x k+1 =f(k,x k ,p k )+e k (3)

[0062] y k =g(k,x k ,pk )+ε k (4)

[0063] Where k is the time step, x k y is a system state parameter. k For system output parameters, e k and ε k For an uncorrelated zero-mean white noise sequence, e k The covariance matrix is ​​Q,ε k The covariance matrix is ​​R,p k Engine failure factors can characterize the engine's health status and can include factors such as pipeline leakage, turbopump efficiency failure factors, or other characteristic parameters of failures.

[0064] The nonlinear discrete dynamic model of a rocket engine containing a fault factor can be linearized at a certain reference point to obtain the discrete linearized state-space model of the rocket engine, as shown in Equations (5) and (6):

[0065] Δx k+1 =A′Δx k +L′Δp k +e k (5)

[0066] Δy k =C′Δx k +M′Δp k +ε k (6)

[0067] The state transition matrix is ​​A′, the output matrix is ​​C′, and the fault matrices are L′ and M′. Δx is the residual of the system state parameters, and Δy is the residual of the system output parameters. Therefore, the discrete linearized state space model of the rocket engine uses the residual of the system state parameters as the state parameters of the model and the residual of the system output parameters as the system output parameters. In order to match the second recurrent neural network, it is necessary to transform the residual of the state parameters in the discrete linearized state space model of the rocket engine into state parameters, the residual of the output parameters into output parameters, and expand the dimension of the fault factor residual into system state parameters to obtain the augmented state space model of the rocket engine.

[0068] Specifically, substituting Δx = x - x0 and Δy = y - y0 into formulas (5) and (6) yields formulas (7) and (8):

[0069] (x k+1 -x0)=A′(x k -x0)+L′Δp+e k (7)

[0070] (y k -y0)=C′(x k -x0)+M′Δp+ε k (8)

[0071] By splitting and merging formulas (7) and (8), we obtain formulas (9) and (10):

[0072] x k+1 =A′x k +L′Δp+(Ix0-A′x0)+e k (9)

[0073] y k =C′x k +M′Δp+(Iy0-C′x0)+ε k (10)

[0074] Then, the fault factors are expanded to system state parameters, resulting in formulas (11) and (12):

[0075]

[0076]

[0077] Since the second recurrent neural network is trained using the rocket engine state space model under normal operating conditions as represented by formulas (1) and (2), and (Ix0-A′x0) and (Iy0-C′x0) are both constant matrices, in order to match the second recurrent neural network, (Ix0-A′x0) and (Iy0-C′x0) are replaced with JJ and KK matrices respectively, thus obtaining the rocket engine augmented state space model with the fault factor as the system state parameter, as shown in formulas (13) and (14):

[0078]

[0079]

[0080] Finally, the hidden layer information output by the second recurrent neural network is mapped to a submatrix A′ in the state transition matrix of the rocket engine augmented state space model, thus obtaining the rocket engine fault model. The rocket engine fault model at the corresponding time is shown in Equations (15) and (16):

[0081]

[0082]

[0083] make C aug= [C′ M′]. Therefore:

[0084] x aug,k+1 =A aug x aug,k +JJ+e k (17)

[0085] y k =C aug x aug,k +KK+ε k (18)

[0086] Among them, e k The covariance matrix is ​​Q aug ,ε k The covariance matrix is ​​R.

[0087] Step 105: Based on the measured values ​​of the target parameters, an observer is used to observe the fault factors in the rocket engine fault model to obtain the estimated values ​​of the fault factors, thereby realizing the fault diagnosis of the engine.

[0088] The observers include extended Kalman filters, particle filters, and unscented Kalman filters. The following section will use Kalman filtering as an example to explain fault diagnosis in detail:

[0089] When diagnosing rocket engine faults, the measured values ​​of target parameters are acquired in real time, and the measured values ​​of target parameters are input into the rocket engine fault model, and the initial measured values ​​of covariates are input into the first recurrent neural network in the rocket engine fault model.

[0090] Using Kalman filtering, the state parameters of the rocket engine fault model are predicted based on the state equations in the model, and the mean and variance of the state parameters are obtained.

[0091] Based on the prediction results and the measured values ​​of the target parameters, the mean and variance of the state parameters are updated to obtain the estimated results of the state parameters; the estimated results include the estimated values ​​of the fault factors.

[0092] Specifically, the prediction equations for the Kalman filter are shown in equations (19), (20), and (21):

[0093]

[0094]

[0095]

[0096] in, For the state parameter x augIn the prior knowledge at time k+1, P aug The variance of the state parameters, Let A be the prior of P at the corresponding time step k+1. aug,k+1 Let be the Jacobian matrix of the state parameters at time k+1.

[0097] Update: The update equations for the Kalman filter are shown in equations (22), (23), and (24):

[0098]

[0099]

[0100]

[0101] Where I is the standard matrix, K aug Kalman gain. Through Kalman gain K aug This determines the bias in the system's prediction and measurement results when weighted summing. When K... aug The smaller the value, the more the weighted sum is biased towards the system's prediction result. aug The larger the size, the opposite is true.

[0102] Specifically, the change in state estimation error can be observed from the change of P over time. First, given the initial state value, the prior prediction results of the mean and variance of the state parameters at k=1 are obtained based on formulas (19), (20), and (21). Then, based on formulas (22), (23), and (24) and according to the measured value of the target parameter at k=1, the mean and variance of the state parameters at k=1 are updated to obtain the posterior results of the mean and variance of the state parameters at k=1, completing one prediction and update process. Then, based on the posterior results at k=1, the prior prediction results of the mean and variance of the state parameters at k=2 are obtained. Based on the measured value of the target parameter at k=2 and the prior prediction results at k=2, the mean and variance of the state parameters at k=2 are updated to obtain the posterior results of the mean and variance of the state parameters at k=2. By analogy, the estimation results of the mean and variance of the state parameters at any time can be obtained iteratively.

[0103] Since the fault factor is a system state parameter, the estimation result of the state parameter includes the estimated value of the fault factor. Based on the estimated value of the fault factor, it can be determined whether the rocket engine is faulty.

[0104] The above mainly describes the solution provided by this invention from the perspective of interaction between various network elements. It is understood that, to achieve the above functions, it includes corresponding hardware structures and / or software modules for executing each function. Those skilled in the art should readily recognize that, in conjunction with the units and algorithm steps of the examples described in the embodiments disclosed herein, this invention can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed by hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention.

[0105] The embodiments of the present invention can divide functional modules according to the above method examples. For example, each function can be divided into its own functional module, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware or as a software functional module. It should be noted that the module division in the embodiments of the present invention is illustrative and only represents one logical functional division; other division methods may be used in actual implementation.

[0106] When dividing each function into modules according to its corresponding function. Figure 6 A schematic diagram of the structure of a rocket engine fault diagnosis device provided by the present invention is shown. Figure 6 As shown, the rocket engine fault diagnosis device includes:

[0107] Engine measurement data acquisition module 601 is used to acquire engine measurement data; the measurement data includes initial measurement values ​​of covariates and measurement values ​​of target parameters at various times;

[0108] The covariate prediction module 602 is used to input the initial measurement value of the covariate into the first recurrent neural network to obtain the predicted value of the covariate at each time step.

[0109] The hidden layer information calculation module 603 is used to input the predicted value of the covariate at each time step into the second recurrent neural network to obtain the hidden layer information at the corresponding time step.

[0110] The rocket engine fault model establishment module 604 is used to map the hidden layer information into a subarray of the state transition matrix of the rocket engine augmented state space model to obtain the rocket engine fault model.

[0111] The fault factor observation module 605 is used to observe the fault factors in the rocket engine fault model based on the measured values ​​of the target parameters, obtain the estimated values ​​of the fault factors, and realize the fault diagnosis of the engine.

[0112] Optionally, the device further includes: a first recurrent neural network model, specifically a first initial recurrent neural network model;

[0113] The measured value of the covariate at time n-1 in the engine normal test data during the steady state phase is input into the first initial recurrent neural network to predict the first predicted value of the covariate at time n.

[0114] The measured value of the covariate at time n in the steady-state normal test data is compared with the first predicted value. The parameters of the first initial recurrent neural network are adjusted according to the comparison result. The first recurrent neural network is obtained after training is completed, where n is a positive integer.

[0115] Optionally, the device further includes: a second recurrent neural network for model building, specifically used for:

[0116] Establish a second initial recurrent neural network and a state-space model of the rocket engine under normal operating conditions;

[0117] The measured values ​​of the covariates at time n-1 in the steady-state normal test data are input into the second initial recurrent neural network to obtain the initial hidden layer information;

[0118] Based on the measured values ​​of the target parameters at time n-1 in the steady-state normal test data, and mapping the initial hidden layer information to the initial state transition matrix of the rocket engine state space model, the predicted values ​​of the target parameters at time n are obtained through the rocket engine state space model.

[0119] The second predicted value is compared with the measured value of the target parameter at time n in the normal test data during the steady-state phase. The parameters of the second initial recurrent neural network are adjusted according to the comparison result, and the trained second recurrent neural network is obtained after training is completed; n is a positive integer.

[0120] Optionally, the device may further include: a rocket engine augmented state space model establishment module, specifically used for:

[0121] Establish a discrete linearized state-space model of the rocket engine after fault injection;

[0122] The residuals of the state parameters in the discrete linearized state-space model of the rocket engine are transformed into state parameters, the residuals of the output parameters are transformed into output parameters, and the residuals of the fault factors are expanded to become system state parameters, thus obtaining the augmented state-space model of the rocket engine.

[0123] Optionally, the device further includes: a module for establishing a discrete linearized state-space model of the rocket engine after fault injection, specifically used for:

[0124] Establish an object-oriented rocket engine model;

[0125] In the object-oriented rocket engine model, fault factors representing the component-level models of each subsystem are constructed and linearized to obtain a discrete linearized state-space model of the rocket engine after fault injection.

[0126] Optionally, the fault factor observation module 605 can be specifically used for:

[0127] The measured values ​​of the target parameters are input into the rocket engine fault model;

[0128] Using Kalman filtering technology, the state parameters of the rocket engine fault model are predicted based on the state equation in the rocket engine fault model, and the mean and variance of the state parameters are obtained.

[0129] Based on the prediction results and the measured values ​​of the target parameters, the mean and variance of the state parameters are updated to obtain the estimated results of the state parameters; the estimated results include the estimated values ​​of the fault factors.

[0130] The rocket engine is determined to be faulty based on the estimated value of the fault factor.

[0131] Optionally, the covariate prediction module 602 can be specifically used for:

[0132] The initial measurement value of the covariate is input into the first RNN neuron in the first recurrent neural network to obtain the predicted value of the covariate at the first time step.

[0133] The prediction result of the previous RNN neuron is input into the next RNN neuron to predict the covariates at the next time step, thus obtaining the predicted values ​​of the covariates at each time step.

[0134] Optionally, the observer may include an extended Kalman filter, a particle filter, or a tasteless Kalman filter.

[0135] On the one hand, a computer-readable storage medium is provided, which stores instructions that, when executed, are used to implement:

[0136] Acquire engine measurement data; the measurement data includes initial measurement values ​​of covariates and measurement values ​​of target parameters at various time points;

[0137] The initial measured values ​​of the covariates are input into the first recurrent neural network to obtain the predicted values ​​of the covariates at each time step.

[0138] The predicted values ​​of the covariates at each time step are input into the second recurrent neural network to obtain the hidden layer information at the corresponding time step.

[0139] The hidden layer information is mapped to a subarray of the state transition matrix of the rocket engine augmented state space model to obtain the rocket engine fault model; the rocket engine augmented state space model is a model with fault factors as system state parameters.

[0140] Based on the measured values ​​of the target parameters, an observer is used to observe the fault factors in the rocket engine fault model, and the estimated values ​​of the fault factors are obtained, thereby realizing the fault diagnosis of the engine.

[0141] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instructions are loaded and executed on a computer, the processes or functions described in this invention are performed entirely or partially. The computer can be a general-purpose computer, a special-purpose computer, a computer network, a terminal, a user equipment, or other programmable device. The computer program or instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. For example, the computer program or instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center integrating one or more available media. The available medium can be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; it can also be an optical medium, such as a digital video disc (DVD); or it can be a semiconductor medium, such as a solid-state drive (SSD).

[0142] Although the invention has been described herein, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings, the disclosure, and the appended claims in carrying out the claimed invention. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple components. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0143] Although the invention has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made therein without departing from the spirit and scope of the invention. Accordingly, this specification and drawings are merely exemplary descriptions of the invention as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if such modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include such modifications and modifications.

Claims

1. A method for diagnosing rocket engine faults, characterized in that, include: Acquire engine measurement data; the measurement data includes initial measurement values ​​of covariates and measurement values ​​of target parameters at various time points; The initial measured values ​​of the covariates are input into the first recurrent neural network to obtain the predicted values ​​of the covariates at each time step. The predicted values ​​of the covariates at each time step are input into the second recurrent neural network to obtain the hidden layer information at the corresponding time step. The hidden layer information is mapped to a subarray of the state transition matrix of the rocket engine augmented state space model to obtain the rocket engine fault model; the rocket engine augmented state space model is a model with fault factors as system state parameters. Based on the measured values ​​of the target parameters, an observer is used to observe the fault factors in the rocket engine fault model, and the estimated values ​​of the fault factors are obtained, thereby realizing the fault diagnosis of the engine. The step of inputting the initial measurement values ​​of the covariates into the first recurrent neural network to obtain the predicted values ​​of the covariates at each time step also includes: Establish a second initial recurrent neural network and a state-space model of the rocket engine under normal operating conditions; The measured values ​​of the covariates at time n-1 in the steady-state normal test data The information is input into the second initial recurrent neural network to obtain the initial hidden layer information at time n-1; this initial hidden layer information is then mapped to the initial state transition matrix of the rocket engine state-space model. Simultaneously, the measured values ​​of the target parameters at time n-1 will be... The values ​​are input into the rocket engine state-space model to obtain the predicted values ​​of the target parameters at time n. The predicted value of the target parameter at time n The measured value of the target parameter at time n A comparison was made, and the parameters of the second initial recurrent neural network were adjusted based on the comparison results; then, the measured values ​​of the covariates at time n in the steady-state normal test data were obtained. The information is input into the second initial recurrent neural network to obtain the initial hidden layer information at time n; this initial hidden layer information is then mapped to the initial state transition matrix of the rocket engine state-space model. Simultaneously, the measured values ​​of the target parameters at time n The values ​​are input into the rocket engine state-space model to obtain the predicted values ​​of the target parameters at time n+1. The predicted value of the target parameter at time n+1 The measured values ​​of the target parameters at time n+1 The parameters of the second initial recurrent neural network are adjusted based on the comparison results until the adjusted parameters meet the requirements, thus obtaining the trained second recurrent neural network; n is a positive integer.

2. The method according to claim 1, characterized in that, The step of inputting the initial measurement values ​​of the covariates into the first recurrent neural network to obtain the predicted values ​​of the covariates at each time step also includes: Establish the first initial recurrent neural network; The measured value of the covariate at time n-1 in the engine normal test data during the steady state phase is input into the first initial recurrent neural network to predict the first predicted value of the covariate at time n. The measured value of the covariate at time n in the steady-state normal test data is compared with the first predicted value. The parameters of the first initial recurrent neural network are adjusted according to the comparison result. The first recurrent neural network is obtained after training is completed, where n is a positive integer.

3. The method according to claim 1, characterized in that, The step of mapping the hidden layer information to a submatrix of the state transition matrix of the rocket engine augmented state space model to obtain the rocket engine fault model also includes: Establish a discrete linearized state-space model of the rocket engine after fault injection; The residuals of the state parameters in the discrete linearized state-space model of the rocket engine are transformed into state parameters, the residuals of the output parameters are transformed into output parameters, and the residuals of the fault factors are expanded to become system state parameters, thus obtaining the augmented state-space model of the rocket engine.

4. The method according to claim 1, characterized in that, The measured values ​​based on the target parameters are used by an observer to observe the fault factors in the rocket engine fault model, obtaining estimated values ​​of the fault factors to achieve fault diagnosis of the engine, including: The measured values ​​of the target parameters are input into the rocket engine fault model; Using Kalman filtering technology, the state parameters of the rocket engine fault model are predicted based on the state equation in the rocket engine fault model, and the mean and variance of the state parameters are obtained. Based on the prediction results and the measured values ​​of the target parameters, the mean and variance of the state parameters are updated to obtain the estimated results of the state parameters; the estimated results include the estimated values ​​of the fault factors. The rocket engine is determined to be faulty based on the estimated value of the fault factor.

5. The method according to claim 3, characterized in that, The process of establishing a discrete linearized state-space model of the rocket engine after fault injection also includes, prior to: Establish an object-oriented rocket engine model; In the object-oriented rocket engine model, fault factors representing the component-level models of each subsystem are constructed and linearized to obtain a discrete linearized state-space model of the rocket engine after fault injection.

6. The method according to claim 1, characterized in that, The step of inputting the initial measured values ​​of the covariates into the first recurrent neural network to obtain the predicted values ​​of the covariates at each time step includes: The initial measurement value of the covariate is input into the first RNN neuron in the first recurrent neural network to obtain the predicted value of the covariate at the first time step. The prediction result of the previous RNN neuron is input into the next RNN neuron to predict the covariates at the next time step, thus obtaining the predicted values ​​of the covariates at each time step.

7. The method according to claim 1, characterized in that, The observers include extended Kalman filters, particle filters, and tasteless Kalman filters.

8. A rocket engine fault diagnosis device, characterized in that, include: An engine measurement data acquisition module is used to acquire engine measurement data; the measurement data includes initial measurement values ​​of covariates and measurement values ​​of target parameters at various times. The covariate prediction module is used to input the initial measurement value of the covariate into the first recurrent neural network to obtain the predicted value of the covariate at each time step. The hidden layer information calculation module is used to input the predicted value of the covariate at each time step into the second recurrent neural network to obtain the hidden layer information at the corresponding time step. The rocket engine fault model establishment module is used to map the hidden layer information into a subarray of the state transition matrix of the rocket engine augmented state space model to obtain the rocket engine fault model. The fault factor observation module is used to observe the fault factors in the rocket engine fault model based on the measured values ​​of the target parameters, and obtain the estimated values ​​of the fault factors to realize the fault diagnosis of the engine. The device also includes: a second recurrent neural network model, used to establish a second initial recurrent neural network and a rocket engine state space model under normal operating conditions; The measured values ​​of the covariates at time n-1 in the steady-state normal test data The information is input into the second initial recurrent neural network to obtain the initial hidden layer information at time n-1; this initial hidden layer information is then mapped to the initial state transition matrix of the rocket engine state-space model. Simultaneously, the measured values ​​of the target parameters at time n-1 will be... The values ​​are input into the rocket engine state-space model to obtain the predicted values ​​of the target parameters at time n. The predicted value of the target parameter at time n The measured value of the target parameter at time n A comparison was made, and the parameters of the second initial recurrent neural network were adjusted based on the comparison results; then, the measured values ​​of the covariates at time n in the steady-state normal test data were obtained. The information is input into the second initial recurrent neural network to obtain the initial hidden layer information at time n; this initial hidden layer information is then mapped to the initial state transition matrix of the rocket engine state-space model. Simultaneously, the measured values ​​of the target parameters at time n The values ​​are input into the rocket engine state-space model to obtain the predicted values ​​of the target parameters at time n+1. The predicted value of the target parameter at time n+1 The measured values ​​of the target parameters at time n+1 The parameters of the second initial recurrent neural network are adjusted based on the comparison results until the adjusted parameters meet the requirements, thus obtaining the trained second recurrent neural network; n is a positive integer.

9. A computer-readable storage medium, characterized in that, include: The computer-readable storage medium stores instructions that, when executed, implement the rocket engine fault diagnosis method according to any one of claims 1-7.

Images