Rocket engine fault diagnosis method based on WGAN-MLP

Abstract

The invention discloses a rocket engine fault diagnosis method based on WGAN-MLP, and the method comprises the steps: S1, constructing a WGAN model, and training the WGAN model through the fault data of an engine; s2, generating a fault sample by using the trained WGAN model; s3, constructing an MLP model, and training the MLP model by using engine measured data and the fault sample generated in the step S2; and S4, realizing engine fault diagnosis by using the trained MLP model. According to the method, the fault sample is generated by using the WGAN model, the problem of insufficient fault data of the current engine can be overcome, fault diagnosis is realized based on the WGAN-MLP, and the method has the advantages of high accuracy and fast response.

Description

Technical Field

[0001] The present invention relates to the technical field of rocket engines, and more specifically, to a rocket engine fault diagnosis method based on WGAN-MLP. Background Art

[0002] From 1990 to 2015, there were 127 failures in foreign launch vehicles, most of which were concentrated in the propulsion systems, control systems, interstage separation mechanisms, electrical systems, payload fairings, etc. Among them, there were 64 failures due to propulsion system failures, accounting for 51% of all launch failures. Liquid rocket engines are the core components of the launch vehicle propulsion system, and their reliability is related to the success or failure of space missions. Therefore, it is necessary to further study engine fault diagnosis technology.

[0003] Engine fault diagnosis methods are divided into model-based, signal processing-based and artificial intelligence-based methods. Model-based methods such as fuzzy models, Kalman filtering, recursive structure recognition, etc. have strong interpretability and the ability to detect unknown faults, but they are highly targeted, have poor generalization performance, and are difficult to accurately model. Signal processing-based methods mainly extract features from measurement signals, and then make diagnostic decisions based on fault feature analysis and prior knowledge. They are simple and intuitive, but heavily rely on expert experience, have low accuracy, and focus more on fault diagnosis in the steady-state working stage of the engine. Artificial intelligence-based methods include expert systems, machine learning, etc., which have the advantages of fast response speed and high accuracy, but are greatly affected by the scale and quality of sensor data.

[0004] Therefore, a fault diagnosis method that is not affected by data size and has high accuracy is needed, which is suitable for engine test and actual flight. Summary of the invention

[0005] The purpose of the present invention is to provide a rocket engine fault diagnosis method based on WGAN-MLP to overcome the defects of the prior art.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] The rocket engine fault diagnosis method based on WGAN-MLP includes the following steps:

[0008] S1. constructing a WGAN model, and using engine failure data to train the WGAN model;

[0009] S2, generate fault samples using the trained WGAN model;

[0010] S3, constructing an MLP model, and training the MLP model using the engine measured data and the fault samples generated in step S2;

[0011] S4. Use the trained MLP model to diagnose engine faults.

[0012] Furthermore, the WGAN model in step S1 is composed of a generator and a discriminator, and the generator and the discriminator are respectively composed of independent MLP models;

[0013] The MLP model consists of multiple layers of neurons, including an input layer, a hidden layer, and an output layer. The input layer is composed of input signals, and the hidden layer and the output layer are fully connected layers. The calculation method is:

[0014]

[0015] Where, the subscript i represents the i-th neuron in this layer, j represents the j-th neuron in the previous layer, O is the output, x is the output of the previous layer, f is the activation function, W is the weight, and b is the bias vector.

[0016] Furthermore, the step of training the WGAN model using engine fault data includes:

[0017] Input random noise to the generator, output generated samples, input generated samples to the discriminator, output the discriminant results, and calculate the loss function L of the generator. G Calculate the error and update the network weights of the generator; input the generated samples and real samples to the discriminator, output the discrimination results, and D The error is calculated, the network weights of the discriminator are updated, the training of the generator and the discriminator is performed alternately, and the training is completed when a set number of training rounds or a set condition is reached.

[0018] Furthermore, the WGAN model uses Wasserstein distance to measure the training effect of the generator, and its mathematical form is as follows:

[0019]

[0020] Where P r and P θ For two sample distributions, Π(P r ,P θ ) is P r and P θ The set of all possible joint distributions composed of γ is the joint distribution of x and y, (x, y) ~ γ means sampling from the joint distribution γ to obtain samples x and y, ‖xy‖ is the distance between x and y, and the lower bound of the expected value in all possible joint distributions That is the Wasserstein distance. In order to adapt to the training of neural networks, the Wasserstein distance is approximated as:

[0021]

[0022] In the formula, ‖f‖ L Represents the Lipschitz constant of function f, and the loss function of the generator is calculated as:

[0023]

[0024] The loss function of the discriminator is calculated as:

[0025]

[0026] After the WGAN model training is completed, the generator generates new samples that conform to the original sample distribution, and uses them to generate engine fault data.

[0027] Furthermore, the training process of the MLP model is implemented by a back propagation algorithm. The learning process of the back propagation algorithm consists of forward propagation of signals and back propagation of errors. During forward propagation, the input signal is propagated to the output layer through the input layer and the hidden layer, and the error is calculated based on the actual output and the expected output. During the back propagation process, the error is propagated back along the original path, and the error is shared by each unit of each layer and the weight is corrected.

[0028] Furthermore, the WGAN model is replaced by a conditional GAN, a deep convolutional GAN, or a cyclic GAN model.

[0029] Compared with the prior art, the advantages of the present invention are as follows: the present invention utilizes the WGAN model to generate fault samples, which can overcome the problem of insufficient current engine fault data, and implements fault diagnosis based on WGAN-MLP, which has the advantages of high accuracy and fast response. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0031] Figure 1 It is a flow chart of the rocket engine fault diagnosis method based on WGAN-MLP of the present invention.

[0032] Figure 2 It is a schematic diagram of the WGAN training process of the present invention.

[0033] Figure 3Schematic diagram of the MLP model of the present invention.

[0034] Figure 4 It is a fault diagnosis result diagram of the present invention. DETAILED DESCRIPTION

[0035] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more definite definition of the protection scope of the present invention.

[0036] See also Figure 1-Figure 4 As shown, this embodiment discloses a rocket engine fault diagnosis method based on WGAN-MLP, comprising the following steps:

[0037] Step S1: construct a WGAN model, and use engine fault data to train the WGAN model.

[0038] Step S2: Generate fault samples using the trained WGAN model;

[0039] Step S3, constructing an MLP model, and training the MLP model using the engine measured data and the fault samples generated in step S2;

[0040] Step S4: Use the trained MLP model to implement engine fault diagnosis.

[0041] In this embodiment, the WGAN model in step S1 is composed of a generator and a discriminator, and the generator and the discriminator are respectively composed of independent MLP models.

[0042] The MLP model consists of multiple layers of neurons, including an input layer, a hidden layer, and an output layer. The input layer is composed of input signals, and the hidden layer and the output layer are fully connected layers. The calculation method is:

[0043]

[0044] Where, the subscript i represents the i-th neuron in this layer, j represents the j-th neuron in the previous layer, O is the output, x is the output of the previous layer, f is the activation function, W is the weight, and b is the bias vector.

[0045] The training process of the MLP model is implemented by the back propagation (BP) algorithm. The learning process of the back propagation algorithm consists of the forward propagation of the signal and the back propagation of the error. During the forward propagation, the input signal is propagated to the output layer through the input layer and the hidden layer, and the error is calculated based on the actual output and the expected output. During the back propagation process, the error is transmitted back along the original path, and the error is shared by each unit of each layer and the weight is corrected. This process is completed by the optimization algorithm. In order to speed up the convergence speed of the model and improve the computational efficiency, the Adam optimizer is used as the optimization algorithm, and the dropout method is used to reduce overfitting.

[0046] In this embodiment, the step of training the WGAN model using engine fault data includes:

[0047] Input random noise to the generator, output generated samples, input generated samples to the discriminator, output the discriminant results, and calculate the loss function L of the generator. G Calculate the error and update the network weights of the generator; input the generated samples and real samples to the discriminator, output the discrimination results, and D The error is calculated, the network weights of the discriminator are updated, the training of the generator and the discriminator is performed alternately, and the training is completed when a set number of training rounds or a set condition is reached.

[0048] In this embodiment, the WGAN model uses Wasserstein distance to measure the training effect of the generator, and its mathematical form is as follows:

[0049]

[0050] Where P r and P θ For two sample distributions, Π(P r ,P θ ) is P r and P θ The set of all possible joint distributions composed of γ is the joint distribution of x and y, (x, y) ~ γ means sampling from the joint distribution γ to obtain samples x and y, ‖xy‖ is the distance between x and y, and the lower bound of the expected value in all possible joint distributions That is the Wasserstein distance. In order to adapt to the training of neural networks, the Wasserstein distance is approximated as:

[0051]

[0052] In the formula, ‖f‖ L Represents the Lipschitz constant of function f, and the loss function of the generator is calculated as:

[0053]

[0054] The loss function of the discriminator is calculated as:

[0055]

[0056] After the WGAN model training is completed, the generator generates new samples that conform to the original sample distribution, and uses them to generate engine fault data.

[0057] In this embodiment, the WGAN model is replaced with a conditional GAN, a deep convolutional GAN, or a cyclic GAN model, which can realize the generation of fault samples and achieve basically the same effect.

[0058] In this embodiment, all steps S1 to S4 can be completed through Python programming to achieve rocket engine fault diagnosis based on WGAN-MLP.

[0059] The present invention utilizes the WGAN model to generate fault samples, which can overcome the problem of insufficient current engine fault data, and implements fault diagnosis based on WGAN-MLP, which has the advantages of high accuracy and fast response.

[0060] Although the embodiments of the present invention are described in conjunction with the accompanying drawings, the patent owner may make various variations or modifications within the scope of the appended claims. As long as they do not exceed the protection scope described in the claims of the present invention, they should be within the protection scope of the present invention.

Claims

1. The rocket engine fault diagnosis method based on WGAN-MLP is characterized by: The following steps are involved: S1. constructing a WGAN model, and using engine failure data to train the WGAN model; S2, generate fault samples using the trained WGAN model; S3, constructing an MLP model, and training the MLP model using the engine measured data and the fault samples generated in step S2; S4. Use the trained MLP model to diagnose engine faults.

2. The rocket engine fault diagnosis method based on WGAN-MLP according to claim 1 is characterized in that: The WGAN model in step S1 is composed of a generator and a discriminator, and the generator and the discriminator are respectively composed of independent MLP models; The MLP model consists of multiple layers of neurons, including an input layer, a hidden layer, and an output layer. The input layer is composed of input signals, and the hidden layer and the output layer are fully connected layers. The calculation method is: Where, the subscript i represents the i-th neuron in this layer, j represents the j-th neuron in the previous layer, O is the output, x is the output of the previous layer, f is the activation function, W is the weight, and b is the bias vector.

3. The rocket engine fault diagnosis method based on WGAN-MLP according to claim 2 is characterized in that: The step of training the WGAN model using engine fault data comprises: Input random noise to the generator, output generated samples, input generated samples to the discriminator, output the discriminant results, and calculate the loss function L of the generator. G Calculate the error and update the network weights of the generator; input the generated samples and real samples to the discriminator, output the discrimination results, and D The error is calculated, the network weights of the discriminator are updated, the training of the generator and the discriminator is performed alternately, and the training is completed when a set number of training rounds or a set condition is reached.

4. The rocket engine fault diagnosis method based on WGAN-MLP according to claim 3 is characterized in that: The WGAN model uses Wasserstein distance to measure the training effect of the generator, and its mathematical form is as follows: Where P r and P θ are two sample distributions, Π(Pr,P θ ) is P r and P θ The set of all possible joint distributions composed of γ is the joint distribution of x and y, (x, y) ~ γ means sampling from the joint distribution γ to obtain samples x and y, ‖xy‖ is the distance between x and y, and the lower bound of the expected value in all possible joint distributions That is the Wasserstein distance. In order to adapt to the training of neural networks, the Wasserstein distance is approximated as: In the formula, ‖f‖ L Represents the Lipschitz constant of function f, and the loss function of the generator is calculated as: The loss function of the discriminator is calculated as: After the WGAN model training is completed, the generator generates new samples that conform to the original sample distribution, and uses them to generate engine fault data.

5. The rocket engine fault diagnosis method based on WGAN-MLP according to claim 4 is characterized in that: The training process of the MLP model is implemented by a back propagation algorithm. The learning process of the back propagation algorithm consists of forward propagation of signals and back propagation of errors. During forward propagation, the input signal is propagated to the output layer through the input layer and the hidden layer, and the error is calculated based on the actual output and the expected output. During the back propagation process, the error is transmitted back along the original path, and the error is shared by each unit in each layer and the weight is corrected.

6. The rocket engine fault diagnosis method based on WGAN-MLP according to claim 1, characterized in that: The WGAN model is replaced with conditional GAN, deep convolutional GAN, and cyclic GAN models.

Images