CN115205582A - Intelligent drawing-simulating aviation sensor fault detection and classification method - Google Patents
Intelligent drawing-simulating aviation sensor fault detection and classification method Download PDFInfo
- Publication number
- CN115205582A CN115205582A CN202210650735.3A CN202210650735A CN115205582A CN 115205582 A CN115205582 A CN 115205582A CN 202210650735 A CN202210650735 A CN 202210650735A CN 115205582 A CN115205582 A CN 115205582A
- Authority
- CN
- China
- Prior art keywords
- data
- image
- training
- aviation
- network
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
Abstract
The invention discloses a method for detecting and classifying faults of a drawing-simulating intelligent aviation sensor, which belongs to the technical field of aviation and comprises the following steps: acquiring training data acquired by a plurality of aviation sensors under different types of fault conditions; stacking and packaging training data into an image format to obtain a training image; inputting a training image into an image classification network, and then performing network training by adopting a transfer learning method to obtain an aviation sensor fault detection and classification network; acquiring to-be-detected data acquired by a plurality of aerosensors to be detected in a flight state; stacking and packaging data to be detected into an image format to obtain an image to be detected; and inputting the image to be detected into the network to obtain the fault type of the aeronautical sensor to be detected. The invention converts the problem of the fault detection and classification of the aviation sensor into the problem of the abnormal region detection and classification of the image, and develops the high-accuracy fault detection and classification deep neural network of the aviation sensor by adopting the pre-trained image classification network.
Description
Technical Field
The invention relates to the technical field of aviation, in particular to a method for detecting and classifying faults of a drawing-imitating intelligent aviation sensor.
Background
A large number of sensors are installed on civil airliners, military fighters, general aircrafts and the like, and the sensors are used for measuring the flight states of the aircrafts such as the speed, the attack angle and the like and are the key for guaranteeing the installation and operation of the aircrafts. However, sensors are generally mounted on the outer surface of the aircraft and are therefore subject to rain, ice and, therefore, to failure, which severely affects the operation of the aircraft. Therefore, aircraft sensor failure detection techniques are essential.
The existing aircraft sensor fault detection technology in the industry is developed based on hardware redundancy, namely, a plurality of sets of sensors are assembled on an aircraft, redundancy measurement is carried out on the same flight state, and each set of sensors is monitored through voting logic so as to detect faults. However, the technology needs to be provided with a plurality of sets of sensors, and the cost is high; in addition, in recent years, flight accidents caused by sensor faults such as a330 of european airbus company and B737MAX of american boeing company also indicate that the existing aircraft sensor fault detection technology in the industrial industry is still insufficient.
Unlike hardware redundancy, a number of aircraft sensor fault detection methods are currently deployed based on software redundancy. However, these methods mostly depend on dynamics, kinematics models and parameters of the aircraft, and need to adjust parameters respectively for different aircraft or different flight states of the same aircraft, and thus, the methods have the disadvantages of large workload, low engineering universality and no large-scale application.
In the study of fault detection for mechanical equipment, aircraft engines, and the like, a number of sensor fault detection methods employing deep neural networks have emerged. The method has good universality aiming at different equipment or different running states of the same equipment, and has high fault detection accuracy. However, in these methods, the original structure design and training of the deep neural network need to be developed respectively for different tasks, and the calculation cost is high; the related structural design mostly adopts trial and error and other modes, and the reliability of the network is questioned; these factors limit their engineering applicability.
In deep neural network research oriented to problems such as image classification in the field of computer vision, interpretable analysis and transfer learning methods are widely applied; the former is developed for a convolution depth neural network, and the interpretability analysis theoretical basis of the network is compacted by analyzing the mechanism of convolution kernel operation in the network, so that the reliability of the network is effectively improved. In addition, the transfer learning is based on the interpretability analysis theory, the pre-trained network is transferred to the research of different tasks, and the calculation cost required by network research and development is effectively reduced.
In view of the development of deep neural network research aiming at the problems of image classification and the like in the field of computer vision, a technology which can be applied to the fault detection and classification of the aviation sensor is urgently needed to be developed, so that the purpose of effectively reducing the calculation cost required by network research and development can be realized while the excellent performance of the fault detection and classification of the aviation sensor is ensured.
Disclosure of Invention
Aiming at the problems that the reliability of an aviation sensor fault detection technology based on hardware redundancy development is poor, the universality of the aviation sensor fault detection technology based on software redundancy development is low, and the parameter adjustment workload is large in the prior art, the invention aims to provide a method for detecting and classifying faults of a drawing-like intelligent aviation sensor.
In order to achieve the purpose, the technical scheme of the invention is as follows:
a method for detecting and classifying faults of a drawing-simulating intelligent aviation sensor comprises the following steps:
acquiring training data acquired by a plurality of aviation sensors under different types of fault conditions;
stacking and packaging the training data into an image format to obtain a training image;
inputting the training image into a pre-trained image classification network, and then adopting a transfer learning method to perform network training to obtain an aviation sensor fault detection and classification network;
acquiring to-be-detected data acquired by a plurality of to-be-detected aviation sensors in a flight state;
stacking and packaging the data to be detected into an image format to obtain an image to be detected;
and inputting the image to be detected into the aviation sensor fault detection and classification network to obtain the fault type of the aviation sensor to be detected.
Preferably, the step of stacking and packaging the training data into an image format to obtain a training image is as follows:
sampling training data acquired by N aviation sensors by using a window with the time length of M in the time dimension, stacking the sampled data to obtain an NxM dimensional matrix, and then performing normalization processing on the NxM dimensional matrix to obtain the training image.
Further, before inputting the training image into the pre-trained image classification network, the method further comprises the following steps:
and performing data enhancement processing on the training image to enable the pixel dimension of the training image subjected to data enhancement to meet the input requirement of the image classification network.
Preferably, the training image is subjected to data enhancement processing by using a mosaic method.
Preferably, the pre-trained image classification network is a pre-trained VGG16 image classification network, and all parameters inside the network are trained by adopting an Adams algorithm.
Preferably, learning efficiency parameters of the attenuation situation are adopted in the Adams algorithm so as to avoid the problem of numerical value oscillation in the training process.
Preferably, the training data collected by the aviation sensor includes air momentum data, overload data, euler angle data, euler angular velocity data and acceleration data.
Preferably, the aerodynamic quantity data, the overload data, the euler angle data, the euler angular velocity data and the acceleration data are acquired by 3 aviation sensors respectively and are coupled through the following equations:
wherein { V, α, β } respectively characterize aerodynamic-quantity-related speed, angle of attack, and sideslip angle data, { A x ,A y ,A z Characterize 3 acceleration data, { ψ, θ, φ } characterize 3 Euler angle data, { w x ,w y ,w z Characterize 3 Euler angular velocity data, { g x ,g y ,g z Characterize 3 overload data, g is the acceleration of gravity.
In another aspect, the present invention also provides an electronic device comprising a memory storing executable program code and a processor coupled to the memory; wherein the processor calls the executable program code stored in the memory to perform the method as described above.
In yet another aspect, the present invention also provides a computer-readable storage medium storing a computer program which, when executed by a processor, performs the method as described above.
By adopting the technical scheme, the invention has the beneficial effects that:
1. the method comprises the steps that data collected by an aerial sensor are stacked and packaged in a quasi-photographic image format, and after data enhancement is carried out by adopting a mosaic image method, the fault detection and classification problem of the aerial sensor can be converted into the abnormal region detection and classification problem on an image with the dimension of 224 x 224 pixels, so that the abnormal region detection and classification problem can be conveniently input into an image classification network;
2. based on a pre-trained VGG16 image classification network with excellent performance, a migration learning method is adopted to be applied to the problem of detecting and classifying faults of the aviation sensor, the method for detecting and classifying faults of the aviation sensor with excellent performance is developed, the migration learning is based on an interpretable analysis theory, the pre-trained network is migrated to the research of different tasks, the excellent performance of detecting and classifying faults of the aviation sensor is guaranteed, meanwhile, the calculation cost required by network research and development is effectively reduced, and the method has the advantages of good network reliability and wide engineering application prospect.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a schematic diagram of data collected by an airborne sensor of the present invention;
FIG. 3 is a schematic diagram of a data format after stacking and packaging the data collected by the aviation sensor in FIG. 2;
FIG. 4 is a schematic diagram of data obtained by normalizing speed, angle of attack, and sideslip angle data collected by an aviation sensor;
FIG. 5 is a graphical illustration of the sideslip angle data of FIG. 4 corresponding to a stuck fault in the sampling;
FIG. 6 is a diagram illustrating images corresponding to the sampled data in FIG. 4 when no fault exists;
FIG. 7 is a diagram illustrating an image of the angle of attack data in the graph when noise occurs during sampling;
FIG. 8 is a schematic diagram of a 15 × 31 dimensional image being mosaic enhanced to a 224 × 224 dimensional image;
FIG. 9 is a schematic diagram of learning efficiency decay in network training;
FIG. 10 is a schematic diagram of a network architecture for an aviation sensor fault detection and classification network according to the present invention;
FIG. 11 is a schematic diagram of an interpretable analysis of the aviation sensor fault detection and classification network proposed by the present invention using the CAM method;
fig. 12 is a schematic structural diagram of a third embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made with reference to the accompanying drawings. It should be noted that the description of the embodiments is provided to help understanding of the present invention, and is not intended to limit the present invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
It should be noted that in the description of the present invention, the terms "upper", "lower", "left", "right", "front", "rear", and the like indicate orientations or positional relationships based on structures shown in the drawings, and are only used for convenience in describing the present invention, but do not indicate or imply that the referred device or element must have a specific orientation, be constructed in a specific orientation, and be operated, and thus, should not be construed as limiting the present invention.
In the technical scheme, the terms "first" and "second" are only used for referring to the same or similar structures or corresponding structures with similar functions, and are not used for ranking the importance of the structures, or comparing the sizes or other meanings.
In addition, unless expressly stated or limited otherwise, the terms "mounted" and "connected" are to be construed broadly, e.g., the connection may be a fixed connection, a removable connection, or an integral connection; can be mechanically or electrically connected; the two structures can be directly connected or indirectly connected through an intermediate medium, and the two structures can be communicated with each other. To those skilled in the art, the specific meanings of the above terms in the present invention can be understood in light of the present general concepts, in connection with the specific context of the scheme.
Example one
A method for detecting and classifying faults of a drawing-like intelligent aviation sensor is shown in figure 1 and comprises the following steps:
s1, acquiring training data acquired by a plurality of aviation sensors under different types of fault conditions.
In this embodiment, before S1 is performed, a related database needs to be established first, so as to facilitate training and testing of the image classification network. Firstly, models such as aerodynamic parameters, dynamic characteristics and the like of the aircraft aiming at a type 2 aircraft and 2 flight states are collected based on flight simulation; and the pilot performs low-altitude take-off and landing and the cruise at the typical altitude by the control law to acquire the flight data of the aircraft, and then collects the real flight data of a large passenger plane, a medium fighter plane and the like based on the real aircraft test flight. In flight simulation and real-aircraft test flight, 15 state quantities of 3 air momentum, 3 overload, 3 Euler angles, 3 Euler angular velocities and 3 acceleration of the aircraft are collected. Injecting colored noise to simulate sensor measurement noise in flight simulation data, and testing flight data in flight simulation data and real aircraftAnd fault data of sensor drifting, jamming and the like are injected into the model to simulate the faults of the aviation sensor. In order to avoid the problem of overfitting of the network, collected data are divided into training data and test data, all real aircraft test flight data are placed in the test data without participating in network training in order to test the performance of the network aiming at real flight data in an important mode, and the division of the training data and the test data is shown in table 1. In the embodiment, data of 5 different airplanes under 6 flight states and 5 sensor fault types are collected in total, and are detailed in table 1, wherein B 1 、B 2 The airplane is a large passenger plane, F is a fighter plane, Y is a transport plane, and D is a general airplane. B is 1 Cruise and Y-person operation as simulation data, B 2 Taking off and landing, F people operating, D cruising and B 1 Human operations are real flight data. Referring to the literature, the category 5 sensor failures include sensor jamming, drifting, abnormal noise, etc., and all of these failures have caused serious flight accidents.
TABLE 1 overview of training data and test data used in the present invention
Note: "-" indicates that the altitude data in real flight is not recorded; the failure category "0" represents no failure; failure class "1-5"
And respectively representing the fault types of sensor drift, abnormal noise and the like.
In this embodiment, the data collected by the aviation sensors are specifically divided into 5 types, each type of data is collected 3 in practical application, and 15 data are collected for training, which are respectively the aerodynamic quantity data (speed, attack angle, sideslip angle) collected by 3 aerodynamic measurement sensors, and 3 overload data, 3 euler angle data, 3 euler angular velocity data, and 3 acceleration data collected by 12 inertial measurement sensors.
The above 15 training data, and are coupled by the following equation:
wherein { V, α, β } characterize aerodynamic-quantity-related speed, angle of attack, and angle of sideslip data, respectively, { A x ,A y ,A z Denotes 3 acceleration data, { ψ, θ, φ } denotes 3 Euler angle data, { w x ,w y ,w z Characterize 3 Euler angular velocity data, { g x ,g y ,g z Characterize 3 overload data, g is the acceleration of gravity. The above equations (1) - (3) describe the coupling relationship of the flight state quantities measurable by the 15 sensors of the aircraft, and completely determine the motion of the aircraft, so that the measurement data of the 15 sensors are used as the input.
And S2, stacking and packaging the training data into an image format to obtain a training image.
In this embodiment, training data acquired by N (15) aviation sensors is sampled in a time dimension with a window with a duration of M (30 seconds) (sampling frequency is 1 Hz), the sampled data is normalized and then stacked to obtain a 15 × 31-dimensional matrix (as shown in fig. 3), a training image similar to a gray image is obtained after the matrix is visualized (as shown in fig. 5-7), and then a fault detection and classification problem of the aviation sensors is converted into an abnormal region detection and classification problem on the image.
For example, as shown in fig. 2, the sampled data of the speed V, the attack angle α, and the sideslip angle β are stacked and packaged to obtain a3 × n dimensional matrix as shown in fig. 3, and usually for convenience of calculation, the training data obtained in S1 is first normalized. After the matrix in fig. 3 is visualized, an image as shown in fig. 4 is obtained, from which it can be seen that the sensor sample data of the sideslip angle β has a failure, i.e., is stuck to 0.5 in 21-31 seconds, and correspondingly, in the training image shown in fig. 5, the 3 rd row thereof has a sudden change stripe. And when the sampled data of the aviation sensor has no fault, the corresponding image is as shown in fig. 6, and the image changes smoothly without abnormal areas. When the sampled data (such as the attack angle α) of the aerial sensor has noise, so that the data is uneven, the corresponding image is as shown in fig. 7, and the 2 nd line on the image has noise.
And S3, inputting the training images into a pre-trained image classification network, and then performing network training by adopting a transfer learning method to obtain an aviation sensor fault detection and classification network.
In this embodiment, the pre-trained image classification network is configured as a pre-trained VGG16 network, and input data of the original VGG16 network is 224 × 224 pixel dimensions, so that the training image with 15 × 31 pixel dimensions needs to be enhanced to an abnormal region detection and classification problem on the image with 224 × 224 pixel dimensions, as shown in fig. 8. There are various ways to perform enhancement processing on the training images, as shown in table 2.
Table 2 mosaic data enhancement method compared with the present embodiment and its description
In this embodiment, an image with 224 × 224 pixel dimensions obtained by enhancing a training image with 15 × 31 dimensions is defined as a "mosaic image". The mosaic completely reserves the correlation and coupling characteristics of all data acquired by the aviation sensors and is compatible with the input data structure of the VGG16 network, so that the fault detection and classification problem of the aviation sensors is converted into the abnormal region detection and classification problem on the mosaic.
The VGG16 convolutional deep neural network applied in this embodiment is pre-trained based on the ImageNet database, and can accurately classify 1000 different types of images. After the training image is enhanced by the mosaic image data, the fault detection and classification problem of the aerial sensor can be converted into an abnormal area detection and classification problem on the mosaic image.
In addition, the invention adopts a transfer learning method to transfer the VGG16 network into a fault detection and classification network of the aviation sensor. Firstly, the output of the original VGG16 network is 1000 × 1 dimension, that is, corresponding to 1000 types of image classification problems, for the problem of detecting and classifying faults of the aviation sensor of the present invention, see table 1 and formula (1) -formula (3), it needs to be expanded for 1 no fault and 5 faults, that is, 6 total situations, so that the output of the VGG16 network is modified to 6 × 1 dimension, the training image enhanced by the mosaic image and the corresponding 6 fault types are respectively installed at the input end and the output end of the VGG16 network, and the training data obtained in S1 is called, the pre-trained VGG16 image classification network is loaded, and the needed aviation sensor fault detecting and classifying network for detecting the fault type of the aviation sensor can be obtained after all parameters in the network are trained by adopting Adams algorithm. In order to avoid the problems of numerical value oscillation and the like which may occur in the network training process, learning efficiency parameters of the attenuation situation are adopted in the Adams algorithm, as shown in fig. 9. The network structure of the aviation sensor fault detection and classification network finally obtained by the embodiment is shown in fig. 10.
And S4, acquiring the data to be detected acquired by the plurality of aeronautical sensors to be detected in the flying state.
And S5, stacking and packaging the data to be detected into an image format to obtain an image to be detected.
And S6, inputting the image to be detected into an aviation sensor fault detection and classification network to obtain the fault type of the aviation sensor to be detected.
The processing mode of the data to be detected in the step S5 is the same as that in the step S2, and when the pixel dimension of the image to be detected is not 224 × 224, the same mosaic image mode is applied for enhancement processing, and then the data is input into the trained aerosensor fault detection and classification network, so that an abnormal region in the input image (the image to be detected) can be accurately and quickly determined, the aerosensor represented by the region can be known through the position of the abnormal region in the image, and the corresponding fault type can be obtained through the form of the abnormal region.
In order to enhance the reliability of the network and the interpretability analysis basis of the aviation sensor fault detection and classification network obtained in the compaction step S3, the embodiment further performs interpretability analysis on the network. As shown in fig. 11, a Class Activation Mapping (CAM) method is used to weight the convolutional kernel nodes in the trained aircraft sensor fault detection and classification network to the mosaic image input by the network. In fig. 11, the upper left is the original 15 × 31 dimension input image, and the box indicates that the aviation sensor has drift fault; the mosaic image after enhancement processing is arranged on the upper right, and a box represents an aviation sensor fault occurrence area; the lower graph shows the region of the model which is focused by the CAM method in a highlighting mode, the region is consistent with the failure occurrence region of the mosaic image, the failure of the aviation sensor is vividly characterized as the sudden change of pixel point values on the image, and the CAM method shows the region of the network which is focused on the image in a highlighting mode. As can be seen from fig. 11, the concerned area of the network coincides with the break point of the fault on the graph, so that the operation of the convolution kernel node inside the network is reliable.
Example two
As shown in table 2, in the first embodiment, 7 mosaic image enhancement modes are proposed, and for these 7 data enhancement modes, in this embodiment, the training data shown in table 1 (database) are called, 30 sets of training are performed with the learning efficiency parameters shown in fig. 9, 5 sets of worst and optimal data are discarded, and then the remaining 20 sets of data are averaged to obtain the training process accuracy of the 7 data enhancement modes, as shown in table 3. And continuing to call the test data (completely separated from the training data) shown in table 1 (database), respectively obtaining the mosaic image by 7 data enhancement methods shown in table 2, and performing a test on the model trained in S3 to respectively obtain the network speed measurement accuracy of the 7 data enhancement methods, as shown in table 3.
It can be seen that the network performance obtained by using the data enhancement method of "All _ Repeat" is the best (the highest accuracy in the training process), and the highest accuracy of 97.704% is obtained in the testing process. Therefore, this method was selected as the data enhancement method ultimately employed by the present invention.
TABLE 3 training and testing results for different mosaic data enhancement modes
EXAMPLE III
An electronic device, as shown in fig. 12, includes a memory storing executable program code and a processor coupled to the memory; wherein the processor calls the executable program code stored in the memory to execute the method as disclosed in embodiment one or embodiment two.
Example four
A computer storage medium, in which a computer program is stored, which, when being executed by a processor, performs the method as disclosed in the first or second embodiment.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The embodiments of the present invention have been described in detail with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. It will be apparent to those skilled in the art that various changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, and the scope of protection is still within the scope of the invention.
Claims (10)
1. A method for detecting and classifying faults of a drawing-simulating intelligent aviation sensor is characterized by comprising the following steps: the method comprises the following steps:
acquiring training data acquired by a plurality of aviation sensors under different types of fault conditions;
stacking and packaging the training data into an image format to obtain a training image;
inputting the training image into a pre-trained image classification network, and then adopting a transfer learning method to perform network training to obtain an aviation sensor fault detection and classification network;
acquiring to-be-detected data acquired by a plurality of to-be-detected aviation sensors in a flight state;
stacking and packaging the data to be detected into an image format to obtain an image to be detected;
and inputting the image to be detected into the aviation sensor fault detection and classification network to obtain the fault type of the aviation sensor to be detected.
2. The method of claim 1, wherein: the step of stacking and packaging the training data into an image format to obtain a training image is as follows:
sampling training data acquired by N aviation sensors by using a window with the time length of M in the time dimension, stacking the sampled data to obtain an NxM dimensional matrix, and then performing normalization processing on the NxM dimensional matrix to obtain the training image.
3. The method of claim 1, wherein: before inputting the training image into the pre-trained image classification network, the method further comprises the following steps:
and performing data enhancement processing on the training image to enable the pixel dimension of the training image subjected to data enhancement to meet the input requirement of the image classification network.
4. The method of claim 3, wherein: and performing data enhancement processing on the training image by adopting a mosaic method.
5. The method of claim 1, wherein: the pre-trained image classification network is a pre-trained VGG16 network, and all parameters in the network are trained by adopting an Adams algorithm.
6. The method of claim 5, wherein: learning efficiency parameters of attenuation situations are adopted in the Adams algorithm, so that the problem of numerical value oscillation in the training process is avoided.
7. The method of claim 1, wherein: the training data collected by the aviation sensor comprises air momentum data, overload data, euler angle data, euler angular velocity data and acceleration data.
8. The method of claim 7, wherein: the air momentum data, the overload data, the Euler angle data, the Euler angular velocity data and the acceleration data are acquired by 3 aviation sensors respectively and are coupled through the following equations:
wherein { V, α, β } respectively characterize aerodynamic-quantity-related speed, angle of attack, and sideslip angle data, { A x ,A y ,A z Characterize 3 acceleration data, { ψ, θ, φ } characterize 3 Euler angle data, { w x ,w y ,w z Characterize 3 Euler angular velocity data, { g x ,g y ,g z Characterize 3 overload data, g is the acceleration of gravity.
9. An electronic device, characterized in that: comprising a memory storing executable program code and a processor coupled to the memory; wherein the processor invokes executable program code stored in the memory to perform the method of any of claims 1-8.
10. A computer-readable storage medium storing a computer program, characterized in that: the computer program, when executed by a processor, performs the method of any one of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210650735.3A CN115205582A (en) | 2022-06-09 | 2022-06-09 | Intelligent drawing-simulating aviation sensor fault detection and classification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210650735.3A CN115205582A (en) | 2022-06-09 | 2022-06-09 | Intelligent drawing-simulating aviation sensor fault detection and classification method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115205582A true CN115205582A (en) | 2022-10-18 |
Family
ID=83576437
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210650735.3A Pending CN115205582A (en) | 2022-06-09 | 2022-06-09 | Intelligent drawing-simulating aviation sensor fault detection and classification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115205582A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115824261A (en) * | 2023-01-15 | 2023-03-21 | 北京理工大学 | Control moment gyroscope fault detection method and device and related storage medium |
-
2022
- 2022-06-09 CN CN202210650735.3A patent/CN115205582A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115824261A (en) * | 2023-01-15 | 2023-03-21 | 北京理工大学 | Control moment gyroscope fault detection method and device and related storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US9650152B2 (en) | Flight envelope protection system for unmanned aerial vehicles | |
| CN108196532B (en) | Fault detection and separation method for longitudinal flight control system of unmanned aerial vehicle based on nonlinear adaptive observer | |
| US11472563B2 (en) | Method for detecting freezing conditions for an aircraft by supervised automatic learning | |
| KR20180039570A (en) | Aircraft monitoring system | |
| CN108609202B (en) | Flight bump prediction method and system | |
| CN108427400A (en) | A kind of aircraft airspeed pipe method for diagnosing faults based on neural network Analysis design | |
| Dong et al. | Research on inflight parameter identification and icing location detection of the aircraft | |
| Sevil et al. | Fault diagnosis in air data sensors for receiver aircraft in aerial refueling | |
| CN115205582A (en) | Intelligent drawing-simulating aviation sensor fault detection and classification method | |
| Wang et al. | Hybrid modeling based double-granularity fault detection and diagnosis for quadrotor helicopter | |
| CN111781946A (en) | Unmanned aerial vehicle control and control system based on link | |
| Garbarino et al. | Neural network based architecture for fault detection and isolation in air data systems | |
| Cheng et al. | Application of self-organizing map on flight data analysis for quadcopter health diagnosis system | |
| Battipede et al. | Neural networks for air data estimation: Test of neural network simulating real flight instruments | |
| Zhang et al. | A deep neural network-based fault detection scheme for aircraft IMU sensors | |
| Dong et al. | Deep neural networks-based air data sensors fault detection for aircraft | |
| Pan et al. | A model of fuel consumption estimation and abnormality detection based on airplane flight data analysis | |
| Raptis et al. | A particle filtering-based framework for real-time fault diagnosis of autonomous vehicles | |
| Anaya et al. | Damage classification based on machine learning applications for an un-manned aerial vehicle | |
| Moncayo et al. | Aircraft failure detection and identification over an extended flight envelope using an artificial immune system | |
| Kılıç | Deep Learning-Based Airspeed Estimation System for a Commercial Aircraft: Ticari Uçaklar için Derin Öğrenme Tabanlı Hava Hızı Tahmin Sistemi | |
| CN114757332A (en) | Intelligent fault detection method for aircraft pneumatic sensor | |
| Kılıç | Deep Learning-Based Airspeed Estimation System for a Commercial Aircraft. | |
| Sevil | Airdata sensor based position estimation and fault diagnosis in aerial refueling | |
| Dutta et al. | Machine-Learning Based Rotor Fault Diagnosis in a Multicopter with Strain Data |
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 |