CN114611324A - Fault diagnosis method for high-power magnetic plasma electric propulsion system - Google Patents
Fault diagnosis method for high-power magnetic plasma electric propulsion system Download PDFInfo
- Publication number
- CN114611324A CN114611324A CN202210315939.1A CN202210315939A CN114611324A CN 114611324 A CN114611324 A CN 114611324A CN 202210315939 A CN202210315939 A CN 202210315939A CN 114611324 A CN114611324 A CN 114611324A
- Authority
- CN
- China
- Prior art keywords
- fault
- magnetic field
- current
- training
- factor
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000003745 diagnosis Methods 0.000 title claims abstract description 34
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000012549 training Methods 0.000 claims abstract description 47
- 238000013528 artificial neural network Methods 0.000 claims abstract description 26
- 238000004088 simulation Methods 0.000 claims abstract description 23
- 238000012360 testing method Methods 0.000 claims abstract description 5
- 210000002569 neuron Anatomy 0.000 claims description 25
- 239000003380 propellant Substances 0.000 claims description 20
- 238000004422 calculation algorithm Methods 0.000 claims description 17
- VSGPGYWZVPDDSK-UHFFFAOYSA-N 2-(5-methoxy-2-phenyl-1h-indol-3-yl)-n,n-dimethylethanamine Chemical compound CN(C)CCC=1C2=CC(OC)=CC=C2NC=1C1=CC=CC=C1 VSGPGYWZVPDDSK-UHFFFAOYSA-N 0.000 claims description 9
- 244000020998 Acacia farnesiana Species 0.000 claims description 6
- 235000010643 Leucaena leucocephala Nutrition 0.000 claims description 6
- 230000007613 environmental effect Effects 0.000 claims description 6
- 150000002500 ions Chemical class 0.000 claims description 5
- 230000035699 permeability Effects 0.000 claims description 5
- 238000006073 displacement reaction Methods 0.000 claims description 4
- 230000003628 erosive effect Effects 0.000 claims description 3
- 230000033001 locomotion Effects 0.000 description 8
- 238000010276 construction Methods 0.000 description 4
- 238000010891 electric arc Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000003993 interaction Effects 0.000 description 3
- 230000007257 malfunction Effects 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000000342 Monte Carlo simulation Methods 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000013178 mathematical model Methods 0.000 description 2
- 238000002679 ablation Methods 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000004323 axial length Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010494 dissociation reaction Methods 0.000 description 1
- 230000005593 dissociations Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000004451 qualitative analysis Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 239000002887 superconductor Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Plasma Technology (AREA)
Abstract
The invention provides a fault diagnosis method of a high-power magnetic plasma electric propulsion system, which belongs to the field of fault diagnosis and specifically comprises the following steps: firstly, establishing a high-power magnetic plasma propeller simulation model, and obtaining data in a fault-free state after operation; then setting a fault factor and adding the fault factor to a high-power magnetic plasma propeller simulation model to form a fault simulation simulator, and acquiring data of a fault state; building a cascade feedforward neural network, and performing supervised learning training by using fault-free data and fault data; finally, fault data serving as test signal state quantity are input into the trained neural network by using a fault simulation simulator, output results of the neural network are counted and compared with real fault states, and accuracy is counted; the invention has the characteristics of high diagnosis accuracy and strong adaptability.
Description
Technical Field
The invention belongs to the field of fault diagnosis, and relates to a fault diagnosis method of a high-power magnetic plasma electric propulsion system.
Background
With the continuous progress of the aerospace industry, various countries begin to put the attention to other planets of solar systems such as mars and saturns, and the manned aerospace technology is continuously developed; deep space exploration and manned space have become increasingly the subject of major concern.
The high-power electric propulsion technology has the characteristics of high thrust, high specific impulse, long-time controllable work and the like, and is the optimal selection for executing large-scale space tasks. A magnetic Plasma Dynamic thruster (MPDT, magnetic Plasma dynamics Thrusters) belongs to electromagnetic propulsion in the family of electric propulsion, which generates thrust by ionizing a propellant and injecting energy by a high-temperature arc of a large current, and accelerating Plasma by using lorentz force of interaction of a magnetic field and the current. The high-power MPD electric propulsion system has higher specific impulse, can provide larger thrust to complete complex tasks such as deep space exploration, and is a main thruster for future space exploration, such as document 1: voronov A S, Troitskiy A A, Egorov I D, et al.Magnetoplasmic driver with an applied field based on the second generation high-temperature superconductors [ C ]// Journal of Physics: reference series. IOP Publishing,2020,1686(1):012023.
However, the development of high-power MPD still has great limitations, such as electrode ablation caused by Onset phenomenon, power sinking of anode, and some limitations of propellant influence, which cause that the system is easy to malfunction during operation, and if the malfunction of the electric propulsion system is not discovered and repaired in time, the malfunction may pollute other systems, and even cause task failure. Therefore, the development of a fault diagnosis method for the high-power MPD electric propulsion system is of great significance.
The following methods for fault diagnosis are mainly divided into three directions: analytical model-based methods, knowledge-based methods, and data-driven based methods. For example, a model-based fault diagnosis method of a linear discrete time-varying system is disclosed in document 2: zhong M, Xue T, Ding S X.A surfy on model-based fault diagnosis for linear discrete time-varying systems [ J ]. neurocompting, 2018,306: 51-60. The fault diagnosis problem based on expert knowledge, which combines experience knowledge and mechanism principle to solve various fault problems, is disclosed in document 3: xu S.A surfey of wireless-based automatic diagnosis technologies [ C ]// Journal of Physics: Conference series. IOP Publishing,2019,1187(3): 032006. For the data-driven fault diagnosis method, mechanical faults are identified and classified according to compressed sensing and improved multi-scale network, such as document 4: hu Z X, Wang Y, Ge M F, et al. data-driven fault diagnosis method based on compressed sensing and improved multiscale network [ J ]. IEEE Transactions on Industrial Electronics,2019,67(4): 3216-3225.
The fault diagnosis method based on the analytical model can go deep into the essence of the system and can effectively quantify the dynamic fluctuation of the process, but when the system model is unknown, inaccurate or nonlinear in the process, the method is difficult to realize. And the actual prototype aiming at the MPD electric propulsion system is few, and the fault sample and the empirical data are also few, so the accuracy rate is not high by adopting the expert knowledge mode. The neural network can perform nonlinear modeling and unknown mapping relation learning, and has been used as a new fault diagnosis processing method, as in document 5: ZHao H, Sun S, Jin B. sequential failure based on LSTM neural network [ J ]. Ieee Access,2018,6: 12929-.
Disclosure of Invention
The invention provides a fault diagnosis method of a high-power magnetic plasma electric propulsion system, aiming at the problem that a fault diagnosis algorithm aiming at the high-power MPD electric propulsion system does not exist currently, and the fault diagnosis method of the high-power magnetic plasma electric propulsion system is used for carrying out fault diagnosis on the high-power MPD electric propulsion system by adopting a method of a cascade neural network.
The method for diagnosing the fault of the high-power magnetic plasma electric propulsion system comprises the following specific steps:
step one, establishing a high-power magnetic plasma propeller simulation model, and obtaining data in a fault-free state after operation;
the thruster simulation model comprises a thrust module, an additional magnetic field module and an electric arc module;
the total thrust of the MPDT in the thrust module is expressed as:
TAFin order to add the thrust of the magnetic plasma thruster,
is the flow rate of the propellant, VsfFor propulsion of ion velocity, V, influenced by induced magnetic fieldapFor the velocity of the propelling ions influenced by the additional magnetic field, χ is the influence coefficient of the additional magnetic field, RaIs the anode radius, RcIs the radius of the cathode, w is the angular velocity of rotation of the current sheet, r' is the displacement of the current sheet, TSFTo induce thrust under the influence of a magnetic field.
For a single turn current-carrying circular coil with radius p in the additional magnetic field module, the magnetic field generated at any position (z, r) in space is:
r is the radial position of the spatial position; bρIs the radial magnetic field strength, mu0For vacuum permeability, I is current intensity, a is coil length, ρ is coil radius, k (k) is first ellipse integral, e (k) is second ellipse integral.
The differential expression of the arc module is:
gCfor conductance based on Cassie model, u is instantaneous arc voltage, g is current arc conductance, E0Is the rated voltage of the arc, tauIs a time constant.
Setting a fault factor, adding the fault factor to a high-power magnetic plasma propeller simulation model to form a fault simulation simulator, and collecting data of a fault state;
the failure factors include:
(1) power supply device (PPU) failure, affecting current:
I*＝(1+f1)Im
I*is the actual current of PPU, ImRated current of PPU, f1Is a fault factor of the PPU, f1∈(0,1)；
(2) Failure of the additional field device, affecting the additional field:
B*＝(1+f2)Bm
B*for the actual field strength of the additional magnetic field, BmFor the nominal field strength of the additional field, f2Fault factor for additional magnetic field, f2∈(0，10)；
(3) Cathode erosion and wear failure, affecting cathode length:
(4) Supply system failure, affecting propellant flow:
(5) Other environmental faults (temperature, concentration), influence the ionization degree:
α*＝(1+f5)αm
α*is the actual degree of ionization, alpha, in the propellermIs the nominal ionization degree in the propeller, f5As an environmental fault factor, f5∈(-0.5，0)；
Step three, building a cascade feedforward neural network, and performing supervised learning training by using fault-free data and fault data;
the supervised learning training is divided into an initial training stage and a compensation training stage, and a gradient descent BP algorithm is adopted;
firstly, initializing newly added hidden layer neuron input and output weights to be 0 in an initial training stage, and training the hidden layer weights by using a BP algorithm; when the error is not reduced to a preset value after the training times are reached, adding compensation stage training, namely adding Gaussian noise with the mean value of 0 and the variance of 1 into the weight of the hidden layer neuron; and optimizing the weight of the hidden layer by using a BP algorithm so as to reduce the error to be within a preset value.
And step four, inputting fault data serving as test signal state quantity into the trained neural network by using the fault simulation simulator, counting output results of the neural network, comparing the output results with the real fault state, and counting accuracy.
The invention has the advantages that:
the method is based on the cascade feedforward neural network, solves the problem of fault diagnosis of the high-power magnetic plasma electric propulsion system, and has the characteristics of high diagnosis accuracy and strong adaptability.
Drawings
FIG. 1 is a flow chart of a method for diagnosing a fault in a high power magnetic plasma electric propulsion system according to the present invention;
FIG. 2 is a schematic diagram illustrating the propulsion of a MPD according to the present invention;
FIG. 3 is a simulation model of the high power magnetic plasma thruster of the present invention;
FIG. 4 is a schematic diagram of a cascaded feedforward neural network structure according to the present invention;
FIG. 5 is a PPU step fault result of the present invention;
FIG. 6 is a failure result of the additional magnetic field device of the present invention;
FIG. 7 is an unstable fault diagnosis result of the present invention;
FIG. 8 shows the result of simultaneous fault diagnosis of the PPU and the cathode according to the present invention;
FIG. 9 shows statistical results of Monte Carlo simulation fault diagnosis according to the present invention.
Detailed Description
The technical scheme of the invention is described below by combining the drawings and the embodiment;
the invention relates to a fault diagnosis method of a high-power magnetic plasma electric propulsion system, which is based on the training of a cascade feedforward neural network, so that a nonlinear relation is established between the fault and the observed quantity of the high-power MPD propulsion system, and the purpose of diagnosing the fault of the high-power MPD propulsion system is achieved.
As shown in fig. 1, the following steps are divided:
establishing a high-power magnetic plasma propeller simulation model for simulation in a fault-free state, and obtaining data in the fault-free state after operation;
the propeller simulation model is divided into: the device comprises a thrust module, an additional magnetic field module and an electric arc module;
the present invention selects a magnetic plasma thruster (MPDT) as a research object to perform fault diagnosis, and as shown in fig. 2, generally, the MPDT is composed of a central cathode and a concentric anode, and a propellant is ionized and then accelerated by a large current arc between the electrodes, and in the thruster, the ionized propellant is accelerated by lorentz force and then ejected from the thruster at high speed, thereby forming thrust. In MPDT, an azimuthal magnetic field B is inducedsfGenerated by discharge current, with an axially applied magnetic field BapGenerated by an electromagnetic coil outside the anode.
Magnetic field andthe interaction between the currents mainly generates two force components: orientation (f)θ＝jr×Bap) And axial direction (f)z＝jr×Bsf) Because the radial currents dominate the axial and azimuthal components; acceleration of ionized propellant in a thruster is considered as movement of a continuous current sheet, which is divided into two parts: axial thrust FsfAxial movement of drive and azimuthal thrust FθA driven vortex motion.
The mathematical model is established based on the macroscopic interaction of the MPDT; thus, there are several assumptions:
(1) the acceleration of the propellant can be regarded as the movement of a continuous current sheet in the thruster;
(2) the magnetic line of force of the external magnetic field is parallel to the axis of the propeller;
(3) the current into the propeller is uniform;
(4) the particles in the propeller are completely ionized;
(5) the azimuthal and axial components of the total current in the thruster are ignored compared to the radial component;
(6) aerodynamic thrust is ignored compared to electromagnetic thrust;
for axial movement:
the axial motion of the current sheet in the thruster is dynamically described by newton's second law:
m (t) mass of current sheet, z (t) displacement of current sheet, FsfAn electromagnetic force that induces a magnetic field;
the second equation can be converted to a first order differential equation:
σ＝z(t)
Vsfthe velocity of the propelled ions as a function of the induced magnetic field; the initial conditions were:
axial force F acting on current sheetsfExpressed as:
to induce magnetic field strength.
The current density j is expressed as:
i is the current intensity and r' is the radius of the current sheet.
Induced self field BsfIs shown as
μ0The magnetic permeability is vacuum magnetic permeability, L is the axial length of the coil, and z' is the axial displacement of the current sheet;
thus:
Rais the anode radius, RcIs the cathode radius.
For the rotational movement:
the rotational movement of the current sheet in the thruster can be expressed as:
wherein J (t) and MθAt different axial positions as a result of the flared anode;
is the angular velocity of rotation of the current sheet;
the inertia moment j (t) is expressed as:
swirl torque MθExpressed as:
Bzis the axial magnetic field strength;
at the outlet of the propeller, the azimuth kinetic energy is partially converted into axial kinetic energy due to the action of the magnetic nozzle.
The total thrust of the MPDT can be expressed as
Model of the additional magnetic field:
the performance of the thruster can be improved by adding an additional magnetic field into the magnetic plasma thruster, and the magnetic field must be calculated by considering the influence of the additional magnetic field. And (4) deriving a calculation formula of the additional magnetic field, and calculating the magnetic field intensity generated by the coil when the current is different.
The additional magnetic field is generated by a coil which is added on the outer side of the throat part of the MPDT nozzle in the self-field, the axial direction of the coil is 5 turns, the length of the coil is 0.017m, the radial direction of the coil is 19 turns, the inner diameter of the coil is 0.05m, the outer diameter of the coil is 0.06m, and the current of the coil is 100A. For a single turn current-carrying circular coil with radius r, the magnetic field generated at any position in space (z, r) is:
r is the radial position of the spatial position;
an arc model:
when an arc is considered to be a component of a circuit, it can be considered a black box model. The Mayr model and the Cassie model are two classical arc black box models derived based on three basic balance principles of thermal balance, thermal inertia and thermal dissociation, and have great significance for qualitative analysis of the arc.
The Mayr model is suitable for the condition of small current and large resistance, including the working condition of electric arc with current flowing through a zero region, and the differential expression of the model is
Wherein u is the instantaneous arc voltage; i is the instantaneous arc current; p0Dissipating power for the arc; gMτ is the time constant and g is the current arc conductance for the arc conductance based on the Mayr model.
The Cassie model considers that the arc has a cylindrical gas passage shape, the diameter of which varies with the arc current to keep the arc current density constant. The Cassie model is suitable for the condition of large current and small resistance, and the differential expression of the model is
gCFor Cassie model-based conductance, E0Is the rated voltage of the arc;
by utilizing the mathematical model, a built high-power magnetic plasma propeller simulation model is shown in fig. 3.
Setting a fault factor, adding the fault factor to a high-power magnetic plasma propeller simulation model to form a fault simulation simulator, and collecting data of a fault state;
the design of the fault signal is the basis of a fault algorithm, especially under the condition that an actual running sample is insufficient, the design is of great importance for the simulation of a real fault, and the fault signal of the MPD is designed according to a statistical MPD fault mode and a fault reason.
(1) Power supply device (PPU) failure, affecting current:
I*＝(1+f1)Im
I*is the actual current of PPU, ImRated current of PPU, f1Is a fault factor of the PPU, f1∈(0,1)；
(2) Failure of the additional magnetic field device, affecting the additional magnetic field;
B*＝(1+f2)B
B*for the actual field strength of the additional magnetic field, B is the nominal field strength of the additional magnetic field, f2Fault factor for additional magnetic field, f2∈(0，10)；
(3) Cathode erosion and wear failure, affecting cathode length:
(4) Supply system failure, affecting propellant flow:
(5) Other environmental faults (temperature, concentration), influence the ionization degree:
α*＝(1+f5)α
α*is the actual ionization degree in the propeller, alpha is the rated ionization degree in the propeller, f5As an environmental fault factor, f5∈(-0.5，0)；
Step three, building a cascade feedforward neural network, and performing supervised learning training by using fault-free data and fault data;
as shown in fig. 4, a hybrid training method is adopted in the process of constructing the cascade feedforward neural network, which is an initial training stage and a compensation training stage, and both stages are trained by using a gradient descent algorithm (BP). In the initial training phase, the input and output weights of the newly added hidden layer neurons are initialized to 0, and the initial weights of the neurons need to be randomly generated in a small range except for the initial hidden layer neuron. And then training the weight of the hidden layer by using a BP algorithm, and if the error is not reduced to a preset value after training for many times, considering that the error falls into a local minimum value. To solve this problem, compensation phase training is added, and gaussian noise with a mean value of 0 and a variance of 1 is added to the weights of the hidden layer neurons that have been initially trained. Then, the adjusted weights are optimized by using a BP algorithm, so that the error is reduced to be within a preset value, and theories and experiments show that the convergence and generalization characteristics of the weights of the neural network can be improved by adding noise.
The construction steps are as follows:
1) selecting the simplest three-layer neural network structure, comprising: a hidden layer, an input layer, and an output layer.
The number of neurons in the input-output layer is related to the number of input-output variables which need to be diagnosed by the fault diagnosis system. In the initialization stage, the network only has one hidden layer, the inside of the hidden layer contains one neuron, all weights of the neural network are randomly generated in a small range, the hidden layer is represented by a label C, and a label I represents a single neuron in the hidden layer.
2) Creating a new training set corresponding to the neuron I in the hidden layer C;
wherein the first neuron I in the first hidden layer C does not need to create a training set, and the original training set is set as a new training set of the hidden neuron.
3) Setting the training times of a training set; and training the neuron I in the hidden layer C in a new training set by using a gradient descent algorithm, wherein the training stage is initial part training.
4) And judging the termination standard of the network, and if the construction of the neural network structure is completed according with the standard, not conforming to the next step.
5) Calculating error E of a neural network training set, if the error is reduced to an expected value after training for multiple times, turning to step 3), representing that the initial training reaches the standard, and performing subsequent training; otherwise, go to the next step.
6) And adding a small amount of noise to all the connection weights of the neuron I in the hidden layer C, and then further training.
The noise is set to gaussian noise with an average value of 0 and a variance of 1, and the training number is set, and this time training is final stage training.
7) And (5) judging the termination standard of the network, if the construction of the neural network structure is completed according with the standard, turning to the step 11), and if the construction is not completed, continuing the next step.
8) Calculating error E of the training set, and if the decrement of E after tau times of training is eta, turning to step 6) to further train the neuron I; if the requirement is not met, fixing the input and output connection weight of the neuron, deleting the neuron label, and going to the next step
9) Checking whether a new hidden layer neuron is required to be added, if so, adding a new neuron I in a hidden layer C, initializing the input and output connection weight of the new neuron I to 0, starting further training by using the previous error value, and then turning to the step 2); if not, the hidden layer tag C is deleted and the next step is continued.
10) A new hidden layer is added on the basis of the existing hidden layer of the neural network, and only one neuron is contained in the new hidden layer. And (3) representing the hidden layer by using a label C, representing a neuron by using a label I, initializing the neuron connection weight to be 0, and turning to the step 2) to train again.
11) And outputting the network structure.
And step four, inputting fault data serving as test signal state quantity into the trained neural network by using the fault simulation simulator, counting output results of the neural network, comparing the output results with the real fault state, and counting accuracy.
The results show that:
the results of the PPU step fault and the additional magnetic field device fault for the single fault step detection are shown in fig. 5 and 6. The method can be found to be capable of accurately diagnosing the single-step fault. The pulse signal is added in the fault adding module to obtain a diagnosis result under an unstable fault, so that the fault diagnosis sensitivity can be found to be very high, and the situation of continuous error reporting is avoided, as shown in fig. 7.
For the case where the PPU fails simultaneously with the cathode (with some delay), as shown in fig. 8, it can be found that the algorithm can diagnose multiple failures. 10000 Monte Carlo simulations are performed to obtain the statistical result of fault diagnosis, and the diagnosis accuracy can reach 99.7%, as shown in FIG. 9.
To test the superiority of the algorithm of the present invention, the algorithm of the present invention was compared with other algorithms, as shown in the following table:
TABLE 1 comparison of different algorithms
It can be found that the method of the invention has more advantages in accuracy.
Claims (4)
1. A fault diagnosis method for a high-power magnetic plasma electric propulsion system is characterized by comprising the following specific steps:
firstly, establishing a high-power magnetic plasma propeller simulation model, and obtaining data in a fault-free state after operation;
then setting a fault factor and adding the fault factor to a high-power magnetic plasma propeller simulation model to form a fault simulation simulator, and acquiring data of a fault state;
then, building a cascade feedforward neural network, and performing supervised learning training by using the fault-free data and the fault data;
and finally, inputting fault data serving as test signal state quantity into the trained neural network by using a fault simulation simulator, counting output results of the neural network, comparing the output results with the real fault state, and counting accuracy.
2. The method for diagnosing the fault of the high-power magnetic plasma electric propulsion system as claimed in claim 1, wherein the propeller simulation model comprises a thrust module, an additional magnetic field module and an arc module;
the total thrust of the MPDT in the thrust module is expressed as:
TAFin order to add the thrust of the magnetic plasma thruster,
is the flow rate of the propellant, VsfFor propulsion of ions by influence of an induced magnetic field, VapFor the velocity of the propelling ions influenced by the additional magnetic field, χ is the influence coefficient of the additional magnetic field, RaIs the anode radius, RcIs the radius of the cathode, w is the angular velocity of rotation of the current sheet, r' is the displacement of the current sheet, TSFThrust under the influence of an induced magnetic field;
the magnetic field generated at any position (z, r) in space for a single-turn current-carrying circular coil with the radius of rho in the additional magnetic field module is as follows:
r is the radial position of the spatial position; bρIs the radial magnetic field strength, mu0The magnetic permeability is vacuum magnetic permeability, I is current intensity, a is coil length, rho is coil radius, K (k) is first ellipse integral, and E (k) is second ellipse integral;
the differential expression of the arc module is:
gCfor Cassie model-based conductance, u is the instantaneous arc voltage, g is the current arc conductance, E0τ is the time constant for the rated voltage of the arc.
3. The method for diagnosing the fault of the high-power magnetic plasma electric propulsion system as recited in claim 1, wherein the fault factor includes:
(1) power device failure, affecting current:
I*＝(1+f1)Im
I*is the actual current of PPU, ImRated current of PPU, f1Is a fault factor of the PPU, f1∈(0,1)；
(2) Failure of the additional field device, affecting the additional field:
B*＝(1+f2)Bm
B*for the actual field strength of the additional magnetic field, BmFor the nominal field strength of the additional field, f2Fault factor for additional magnetic field, f2∈(0，10)；
(3) Cathode erosion and wear failure, affecting cathode length:
(4) Supply system failure, affecting propellant flow:
(5) Other environmental faults, affecting the ionization degree:
α*＝(1+f5)αm
α*is the actual degree of ionization, alpha, in the propellermIs the nominal ionization degree in the propeller, f5As an environmental fault factor, f5∈(-0.5，0)。
4. The method for diagnosing the fault of the high-power magnetic plasma electric propulsion system as recited in claim 1, wherein the supervised learning training is divided into an initial training stage and a compensation training stage, both of which adopt a gradient descent BP algorithm;
firstly, initializing newly added hidden layer neuron input and output weights to 0 in an initial training stage, and training the hidden layer weights by using a BP algorithm; when the error is not reduced to a preset value after the training times are reached, adding compensation stage training, namely adding Gaussian noise with the mean value of 0 and the variance of 1 into the weight of the hidden layer neuron; and optimizing the weight of the hidden layer by using a BP algorithm so as to reduce the error to be within a preset value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210315939.1A CN114611324A (en) | 2022-03-28 | 2022-03-28 | Fault diagnosis method for high-power magnetic plasma electric propulsion system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210315939.1A CN114611324A (en) | 2022-03-28 | 2022-03-28 | Fault diagnosis method for high-power magnetic plasma electric propulsion system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114611324A true CN114611324A (en) | 2022-06-10 |
Family
ID=81866101
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210315939.1A Pending CN114611324A (en) | 2022-03-28 | 2022-03-28 | Fault diagnosis method for high-power magnetic plasma electric propulsion system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114611324A (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100237715A1 (en) * | 2009-03-23 | 2010-09-23 | Rolls-Royce Plc | Magnetoplasmadynamic (mpd) generator |
| CN107143475A (en) * | 2017-07-11 | 2017-09-08 | 中国人民解放军国防科学技术大学 | The multistage discharge circuit for the magnetic plasma propeller supported for laser |
| CN112346893A (en) * | 2020-11-10 | 2021-02-09 | 深圳市康必达控制技术有限公司 | Fault prediction method, device, terminal and storage medium |
-
2022
- 2022-03-28 CN CN202210315939.1A patent/CN114611324A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100237715A1 (en) * | 2009-03-23 | 2010-09-23 | Rolls-Royce Plc | Magnetoplasmadynamic (mpd) generator |
| CN107143475A (en) * | 2017-07-11 | 2017-09-08 | 中国人民解放军国防科学技术大学 | The multistage discharge circuit for the magnetic plasma propeller supported for laser |
| CN112346893A (en) * | 2020-11-10 | 2021-02-09 | 深圳市康必达控制技术有限公司 | Fault prediction method, device, terminal and storage medium |
Non-Patent Citations (1)
| Title |
|---|
| JINXING ZHENG等: "Integrated study on the comprehensive magnetic-field configuration performance in the 150 kW superconducting magnetoplasmadynamic thruster", SCIENTIFIC REPORTS, 19 October 2021 (2021-10-19), pages 1 - 11 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Rausch et al. | Learning a deep neural net policy for end-to-end control of autonomous vehicles | |
| Li et al. | Learning the aerodynamic design of supercritical airfoils through deep reinforcement learning | |
| Aboutalebi et al. | A novel sensor fault detection in an unmanned quadrotor based on adaptive neural observer | |
| CN111177974B (en) | Structure small failure probability calculation method based on double-layer nested optimization and subset simulation | |
| Weatheritt et al. | A comparative study of contrasting machine learning frameworks applied to RANS modeling of jets in crossflow | |
| Gurumurthy et al. | Terminal sliding mode disturbance observer based adaptive super twisting sliding mode controller design for a class of nonlinear systems | |
| Valdes et al. | A pulsed plasma thruster fault detection and isolation strategy for formation flying of satellites | |
| US11842575B2 (en) | Method and system for vehicle analysis | |
| CN105606914A (en) | IWO-ELM-based Aviation power converter fault diagnosis method | |
| Zheng et al. | Aero-engine on-board model based on batch normalize deep neural network | |
| CN111415010A (en) | Bayesian neural network-based wind turbine generator parameter identification method | |
| CN112099351B (en) | Distributed fault diagnosis method based on centrosymmetric polyhedron | |
| CN114564000A (en) | Active fault tolerance method and system based on fault diagnosis of intelligent aircraft actuator | |
| Faieghi et al. | Design an optimized PID controller for brushless DC motor by using PSO and based on NARMAX identified model with ANFIS | |
| CN110412997B (en) | Spacecraft attitude control spray pipe fault intelligent diagnosis system and method based on neural network | |
| Taimoor et al. | Lyapunov theory based adaptive neural observers design for aircraft sensors fault detection and isolation | |
| Lin et al. | Engine components fault diagnosis using an improved method of deep belief networks | |
| Peterson et al. | Overview of NASA's Solar Electric Propulsion Project | |
| US20220229948A1 (en) | Simulation of the behavior of a vehicle in a fluid by Hamiltonian neural network | |
| CN111199105A (en) | Flapping wing motion parameter optimization method | |
| CN114611324A (en) | Fault diagnosis method for high-power magnetic plasma electric propulsion system | |
| Zhou et al. | Aero-engine thrust estimation based on ensemble of improved wavelet extreme learning machine | |
| Wu et al. | Study on the Smart Diagnosis of Gas Turbine With Distributed Computing Technique | |
| Guangcun et al. | Control of quadrotor drone with partial state observation via reinforcement learning | |
| Kuspijani et al. | Faults identification of induction motor based on vibration using backpropagation neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |