CN111414932A - Classification identification and fault detection method for multi-scale signals of aircraft - Google Patents

Classification identification and fault detection method for multi-scale signals of aircraft Download PDF

Info

Publication number
CN111414932A
CN111414932A CN202010015165.1A CN202010015165A CN111414932A CN 111414932 A CN111414932 A CN 111414932A CN 202010015165 A CN202010015165 A CN 202010015165A CN 111414932 A CN111414932 A CN 111414932A
Authority
CN
China
Prior art keywords
scale
classification
convolution
aircraft
signal
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.)
Granted
Application number
CN202010015165.1A
Other languages
Chinese (zh)
Other versions
CN111414932B (en
Inventor
李可
刘静怡
文东升
杨顺昆
刘猛
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beihang University
Original Assignee
Beihang University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Beihang University filed Critical Beihang University
Priority to CN202010015165.1A priority Critical patent/CN111414932B/en
Publication of CN111414932A publication Critical patent/CN111414932A/en
Application granted granted Critical
Publication of CN111414932B publication Critical patent/CN111414932B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)

Abstract

A classification identification and fault detection method for multi-scale signals of an aircraft is characterized in that historical data and real-time data are input into a multi-scale residual error convolution network to obtain feature extraction of the aircraft signals, and deep features are input into a classification algorithm to obtain a classification result. During feature extraction, aircraft signal samples containing labels are sent to a deep neuron convolution network module, an input classification algorithm is applied to output signal feature maps, and propagation parameters are updated by using results of the input classification algorithm. The method has the advantages that the original features are input into the multi-scale residual convolution module, the scale judgment module is used for sensing the features of different scales, when signal classification and identification are started, the neural network parameters of the classification algorithm are set, and the signal features extracted through the multi-scale residual dilation convolution module before are input.

Description

Classification identification and fault detection method for multi-scale signals of aircraft
Technical Field
The invention relates to a classification identification and fault detection method for multi-scale signals of an aircraft.
Background
The identification and classification of the aircraft signals are the core of an aircraft health management strategy, the current health state of a complex system is judged, the cause and the source of the fault are effectively found, and a series of suggestions or decisions related to maintenance and guarantee are provided. The health management strategy of the aircraft is widely accepted and applied in the field of aerospace industry, provides great guarantee for the safety and reliability of the spacecraft, and becomes a necessary trend for the development of the aerospace industry.
Disclosure of Invention
An aircraft signal classification and identification method, comprising:
A) when the aircraft signal fault is to be judged, firstly, the data of the complex equipment of the aircraft, including power supply equipment, aircraft attitude control equipment and aircraft radio measurement and control equipment, is subjected to signal acquisition, and the acquired signals are transmitted to a sensor,
B) the signals received by the sensor are pre-processed,
C) then, a judgment process is performed to judge the source of the signal, wherein:
if the signal source is historical data, reading the historical data, performing signal clustering analysis on the historical data so as to effectively assist expert labeling work, and constructing the historical data and corresponding labels into an expert database;
if the signal source is real-time data, reading the real-time data,
D) inputting historical data and real-time data into a neural network together,
E) the deep feature extraction of the aircraft signal is realized through the training of the multi-scale residual convolution network,
F) carrying out signal classification algorithm operation on deep features of the extracted aircraft signals to obtain aircraft signal classification results,
G) performing convergence judgment, wherein:
if the loss function of the classification result reaches convergence in the training, outputting a real-time fault diagnosis result;
if the loss function of the classification result is not converged in the training, updating the weight parameters of the neural network and returning to the step D), and repeating the step E) and the step F) until the loss function is converged,
wherein the step E) comprises:
E1) initializing a random value for a parameter in the neural network, and setting an iteration step number s to be 0,
E2) updating the iteration step number s according to the previous step, and if the previous step is E1), updating s to be 0; if the previous step is no branch in the decision block of E8), updating the iteration step s to s +1,
E3) sampling the aircraft signal x containing the tagsFeeding the signal into a deep neuron convolution network module,
E4) a multi-scale residual convolution operation is performed,
E5) the forward propagation calculates the network output, the output content is the aircraft signal characteristic diagram,
E6) inputting the output content, namely the aircraft signal characteristic diagram, into a cost function module, and reversely updating the propagation parameters by using a cost function result;
E7) judging whether to traverse the training set, if not, returning to the step E2); if the whole training set is traversed, making s equal to s + 1;
E8) judging the number of iteration steps, if the number of iteration steps is s<If the value of epoch is 1000, returning to the step E2); if the number of iteration steps s>Finishing the training of the multi-scale residual convolution network, and extracting deep features of a sample through the trained multi-scale residual convolution network, wherein the sample is an aircraft signal sample x containing a labels，
The step E4), namely the multi-scale residual convolution operation, includes:
inputting original features, entering a convolution kernel scale judgment step, and judging the convolution kernel scale, wherein:
when the scale of the convolution kernel is judged to be small scale, entering the small scale feature extraction branch flow, and sensing the small scale feature, wherein the formula of the convolution kernel is as follows:
Figure RE-GDA0002514305840000021
Figure RE-GDA0002514305840000022
wherein
Figure RE-GDA0002514305840000023
The value of the intermediate profile of the jth channel of convolutional layer l,
Figure RE-GDA0002514305840000024
the values corresponding to the input feature map and the output feature map of the previous layer,
Figure RE-GDA0002514305840000025
is the value of the final output characteristic map of the jth channel of convolutional layer l, f is the activation function, MjRepresentation for computing
Figure RE-GDA0002514305840000026
I.e. the position of the original feature map covered by the convolution kernel,
Figure RE-GDA0002514305840000027
are elements in the weight matrix of the convolution kernel,
Figure RE-GDA0002514305840000028
is an element in the deviation matrix of the convolved feature map;
and sending the result after the convolution kernel into an excitation function, wherein the excitation function adopts a Re L U function, and the formula is expressed as follows:
Figure RE-GDA0002514305840000031
then expansion convolution is carried out, wherein the expansion convolution, also called cavity convolution, adds cavities with a plurality of elements (expansion rates) among each element of the convolution kernel, thereby expanding the receptive field of the convolution layer under the condition of not increasing the parameter number of the convolution kernel,
when the convolution kernel scale is judged to be large scale in the step, the branch flow enters the large scale feature extraction branch flow for sensing the large scale feature, passes through the excitation function and the expansion convolution,
and (5) performing scale fusion, fusing the small-scale features and the large-scale features together, outputting a scale fusion result, and ending the process.
Drawings
Figure 1 shows a flow chart of an algorithm for aircraft signal classification and identification according to one embodiment of the invention,
figure 2 shows a flow diagram of feature extraction for an aircraft signal based on a multi-scale residual convolutional network module according to one embodiment of the present invention,
figure 3 shows a schematic diagram of the multi-scale residual convolution module principle,
FIG. 4 shows a flow diagram of an aircraft signal classification algorithm according to one embodiment of the invention.
Detailed Description
The traditional machine learning algorithm in the aircraft signal identification and classification method lacks the feature extraction capability of high-dimensional data, and has limitations on classification accuracy, classification speed and other performance indexes in the aircraft signal classification problem.
The invention adopts a neural network learning method based on a multi-scale residual error expansion convolution network, which is used as a class of emerging leading-edge machine learning methods, is widely applied to the fields of computer vision, natural language processing and the like related to data science, and has stronger robustness and universality. The deep learning method is based on a deep neural network model, can automatically extract features from complex aircraft signal samples, gradually combines simple features into more complex features, and solves the recognition problem through the complex features. The method solves the problem of high difficulty in processing high-dimensional complex data by a traditional machine learning method through a multi-scale residual expansion convolution network method, and improves the feature extraction capability and the generalization capability.
Therefore, the invention relates to a multi-scale residual error expansion convolution network algorithm and a signal classification algorithm.
According to one aspect of the invention, a multi-scale residual error expansion convolution network method for aircraft signal identification and classification is provided, and is characterized by comprising the following steps:
a training threshold judgment module;
a multi-scale residual convolutional network module;
and a signal classification module.
In order to solve the problem of signal scale identification limitation of the existing aircraft signal classification and identification method, the invention provides an aircraft signal classification and identification method based on a multi-scale residual error neural network, which effectively solves the problems of shallow layer feature extraction, gradient disappearance, single feature scale and the like, and obviously improves the accuracy of aircraft multi-scale signal classification and identification.
A flow diagram of an aircraft signal classification and identification method according to one embodiment of the invention is shown in fig. 1, which includes:
when the aircraft signal needs to be subjected to fault judgment work (101), firstly, data of complex equipment (generally comprising power supply equipment, aircraft attitude control equipment and aircraft radio measurement and control equipment) in the aircraft are subjected to signal acquisition and transmitted to a sensor (102), and an original signal obtained by the sensor is subjected to signal preprocessing (103). Then, a judging process is carried out, the signal source is judged (104), if the signal source is historical data, a historical data reading link (105) is carried out, signal clustering analysis (106) is carried out on the data, so that expert labeling work (107) is effectively assisted, and finally the historical data and corresponding labels are constructed into an expert database (108); if the signal source is real-time data, a real-time data reading step (109) is performed. Inputting historical data and real-time data into a neuron network (110) together, obtaining characteristic extraction (111) of the aircraft signal through learning of a multi-scale residual convolution network, and sending deep characteristics obtained through extraction into a classification algorithm (112) to obtain an aircraft signal classification result. Entering a judging process (113) again, and entering a final link to output a real-time fault diagnosis result if the loss function of the classification result reaches convergence in the trainingIf the loss function of the classification result is not converged in the training process, returning to the neural network (110), updating the neural network weight parameters in the neural network (110), performing the feature extraction (111) again, and performing the signal classification (112) process until the loss function is converged, wherein the overall training process adopts an Adam training method for training, so that the updating process of the parameters is more efficient and more robust, the learning rate of the training is set to be 0.001, and the hyper-parameter β of an Adam optimizer is used1And β2Set to 0.9 and 0.999 respectively.
Fig. 2 is a flow chart of a training process for feature extraction (111) of the multi-scale residual convolutional network in fig. 1. The feature extraction flow chart of the multi-scale residual convolution network is shown in fig. 2, and comprises the following steps:
when the aircraft signal is subjected to feature extraction (201), initializing a random value for a parameter in a neural network, and setting an iteration step number s to be 0 (202); updating the iteration step number s (203) according to the formula of the previous operation, wherein if the previous operation is (202), the updating s is equal to 0; if the former operation is a no branch in the judgment module (211), updating the iteration step number s to be s +1, and obtaining the aircraft signal sample x containing the labelsThe signal is sent to a deep neuron convolution network module (204) and a multi-scale residual convolution module (205), network output (206) is calculated through forward propagation of the multi-scale residual convolution module, the output content, namely an aircraft signal feature map, is input to a cost function module (207), and a propagation parameter (208) is updated by using the result of a classification algorithm; the judging module (209) judges whether to traverse the training set, if not, the step (204) of inputting the labeled sample is returned; if the whole training set has been traversed, updating the iteration step number s +1 (210); an iteration step number judging module (211) for judging if the iteration step number s<If the epoch is not the training set, the training set returning module (203) is retrained; if the number of iteration steps s>And (5) ending the training of the neural convolution network (212) and extracting deep features of the sample (213).
A schematic diagram of the multi-scale residual convolution module (205) according to one embodiment of the present invention is shown in fig. 3, where:
starting a multi-scale residual convolution operation (301), inputting original features (302), entering a convolution kernel scale judgment step (303), and judging the scale of a convolution kernel;
when the convolution kernel scale is judged to be 3, entering a small-scale feature extraction branch flow (304), wherein the convolution kernel size of the small-scale feature extraction branch flow is 3(305) and sensing the small-scale features, wherein the formula of the convolution kernel is as follows:
Figure RE-GDA0002514305840000051
Figure RE-GDA0002514305840000052
wherein
Figure RE-GDA0002514305840000054
The value of the intermediate profile of the jth channel of convolutional layer l,
Figure RE-GDA0002514305840000055
for the corresponding value of the input feature map (the output feature map of the upper layer),
Figure RE-GDA0002514305840000056
is the value of the final output characteristic map of the jth channel of convolutional layer l, f is the activation function, MjRepresentation for computing
Figure RE-GDA0002514305840000057
The portion of the feature map of (a), i.e. the position of the original feature map covered by the convolution kernel,
Figure RE-GDA0002514305840000058
are elements in the weight matrix of the convolution kernel,
Figure RE-GDA0002514305840000059
is an element in the deviation matrix of the convolved feature map;
and then the result after the convolution kernel (305) is sent into an excitation function (306), wherein the excitation function adopts a Re L U function, and the formula is as follows:
Figure RE-GDA0002514305840000053
then expansion convolution (307) is carried out, the expansion convolution (or called cavity convolution) means that cavities with a plurality of elements (expansion rates) are added among each element of a convolution kernel, so that the receptive field of the convolution layer is expanded under the condition that the parameter number of the convolution kernel is not increased, and the expansion rate of the expansion convolution of the small-scale tributary is 2;
and when the convolution kernel scale is judged to be 5 in the step (303), entering a large-scale feature extraction branch flow (308), wherein the convolution kernel size of the large-scale feature extraction branch flow is 5(309), and the large-scale feature extraction branch flow is used for sensing large-scale features, passes through an excitation function (310) and then passes through expansion convolution (311) as same as that of the small-scale branch flow, and the expansion rate of the large-scale feature extraction branch flow is 5.
And then, the features obtained by the small-scale tributaries and the large-scale tributaries are fused together through scale fusion (312), the result of the scale fusion is output, and the process (313) is ended.
The aircraft signal classification algorithm, i.e. the signal classification (112) flow chart in fig. 1, is shown in fig. 4:
when signal classification and identification is started (401), neural network parameters of a classification algorithm are set (402), signal features previously extracted by a multi-scale residual dilation convolution module are input (403), then input to a global average pooling layer (404), then input to a random discard layer (405), the formula of which is as follows:
xl＝randomp(xl-1) (4)
wherein p is discard xl-1Random is expressed as a random behavior.
Then the full connection layer (406) is input:
xl＝f(ul) (5)
ul＝wlxl-1+bl(6)
wherein u islIs an intermediate quantity of the linear output of the full connection layer, which is equal to the output characteristic diagram x of the previous layerl-1With a fully-connected layerWeight matrix w oflMultiplication and then addition of the deviation, wlWeight matrix being a full connection layer, blIs a deviation matrix of the fully connected layers.
The output of the fully-connected layer is fed into a classification function (407) and a loss function calculation is performed. The formula of the classification function is:
the standard classification function softmax function can be expressed as:
Figure RE-GDA0002514305840000061
k is the classification number of the data aggregation; z is the output of the pre-stage unit of the classifier; i is the category of the index and e is the representation of the exponential function.
When the loss function convergence judging module (408) judges that the convergence is not carried out, returning to the parameter setting (402) and updating the weight parameter of the network; if the loss function convergence judging module (408) judges that convergence occurs, the classification result (409) is directly output. In practical multi-classification problems, a cross-entropy function is often used as a loss function, measuring the amount of effort required to remove the uncertainty of the system given a true distribution using a strategy specified by a non-true distribution, and possibly reducing gradient dispersion to some extent. The standard cross entropy loss function can be expressed as:
Figure RE-GDA0002514305840000071
wherein the content of the first and second substances,
Figure RE-GDA0002514305840000072
calculating labels according to the label classifier; y isiIs a true tag of the aircraft signal; i is the category of the index.
The advantages and beneficial effects of the invention include:
1) the deep learning method for aircraft signal identification and classification has good robustness and universality.
2) Compared with the traditional classification method, the method effectively solves the problem of extracting multi-scale features from high-dimensional signals, can identify and classify the signals with different scale features, and overcomes the gradient disappearance in a deep network.
3) The expansion convolution is effectively adopted, the number and complexity of parameters of the network are effectively controlled, and the operation efficiency is further improved.
4) The method remarkably improves the classification and identification accuracy of the aircraft signals, and makes a remarkable contribution to the core fault identification of the aircraft health management strategy.

Claims (6)

1. An aircraft signal classification and identification method, comprising:
A) when the aircraft signal fault is to be distinguished (101), firstly, the data of complex equipment including power supply equipment, aircraft attitude control equipment and aircraft radio measurement and control equipment in the aircraft is subjected to signal acquisition, and the acquired signals are transmitted to a sensor (102),
B) the signals received by the sensors are pre-processed (103),
C) then, a determination process is performed to determine the source of the signal (104), wherein:
if the signal source is historical data, reading the historical data (105), performing signal clustering analysis (106) on the historical data, thereby effectively assisting expert labeling work (107), and constructing the historical data and labels corresponding to the historical data into an expert database (108);
if the signal source is real-time data, the real-time data is read (109),
D) historical data and real-time data are input into a neural network (110) together,
E) deep feature extraction (111) of the aircraft signal is realized through training of a multi-scale residual convolution network,
F) performing a signal classification algorithm operation (112) on the deep features of the extracted aircraft signal to obtain aircraft signal classification results,
G) performing a convergence determination (113), wherein:
if the loss function of the classification result reaches convergence in the training, outputting a result of real-time fault diagnosis (114);
if the loss function of the classification result is not converged in the training, updating the weight parameters of the neural network and returning to the step D), and repeating the step E) and the step F) until the loss function is converged,
wherein the step E) comprises:
E1) initializing random values for parameters in the neural network, and setting an iteration step number s equal to 0(202),
E2) updating the iteration step number s (203) according to the previous step, and if the previous step is E1), updating s to be 0; if the previous step is no branch in the judgment module (211) of E8), updating the iteration step number s to s +1,
E3) sampling the aircraft signal x containing the tagsSending the signal into a deep neuron convolutional network module (204),
E4) a multi-scale residual convolution operation is performed (205),
E5) the forward propagation computes a network output (206), the content of which is an aircraft signal profile,
E6) inputting the output content, namely the aircraft signal characteristic diagram, into a cost function module (207), and reversely updating the propagation parameters by using the cost function result (208);
E7) judging whether to traverse the training set (209), if not, returning to the step E2); if the entire training set has been traversed, let s +1 (210);
E8) judging the number of iteration steps (211), if the number of iteration steps s<If the value of epoch is 1000, returning to the step E2); if the number of iteration steps s>And ending the training (212) of the multi-scale residual convolution network, and extracting deep features of the sample through the trained multi-scale residual convolution network, wherein the sample is the aircraft signal sample x containing the labels，
Said step E4), the multi-scale residual convolution operation (205), comprises:
inputting original features (302), entering a convolution kernel scale judgment step (303), and performing convolution kernel scale judgment, wherein:
when the scale of the convolution kernel is judged to be small scale, entering a small scale feature extraction branch flow (304), and sensing the small scale feature, wherein the formula of the convolution kernel is as follows:
Figure FDA0002358607190000021
Figure FDA0002358607190000022
wherein
Figure FDA0002358607190000023
The value of the intermediate profile of the jth channel of convolutional layer l,
Figure FDA0002358607190000024
the values corresponding to the input feature map and the output feature map of the previous layer,
Figure FDA0002358607190000025
is the value of the final output characteristic map of the jth channel of convolutional layer l, f is the activation function, MjRepresentation for computing
Figure FDA0002358607190000026
I.e. the position of the original feature map covered by the convolution kernel,
Figure FDA0002358607190000027
are elements in the weight matrix of the convolution kernel,
Figure FDA0002358607190000028
is an element in the deviation matrix of the convolved feature map;
the result after the convolution kernel (305) is fed into an excitation function (306) which takes the Re L U function and is formulated as:
Figure FDA0002358607190000029
then, performing a dilation convolution (307), wherein the dilation convolution, also called hole convolution, adds holes with several elements (dilation rate) between each element of the convolution kernel, thereby enlarging the receptive field of the convolution layer without increasing the parameter number of the convolution kernel,
when the convolution kernel scale is judged to be large scale in the step (303), entering a large scale feature extraction branch flow (308) for sensing large scale features, passing through an excitation function (310), and then passing through an expansion convolution (311),
and (3) performing scale fusion (312), fusing the small-scale features and the large-scale features together, outputting a scale fusion result, and ending the process (313).
2. The aircraft signal classification and identification method according to claim 1, characterized in that the classification algorithm operation (112) comprises:
C1) when signal classification and recognition is started (401), setting neural network parameters of a classification algorithm (402),
C2) inputting deep features of the aircraft signal previously extracted by a multi-scale residual dilation-convolution module (403),
C3) next, deep features of the aircraft signal are input into a global average pooling layer (404),
C4) next, a random discard layer (405) is input, whose formula is shown below:
xl＝randomp(xl-1) (4)
wherein p is discard xl-1Random, is expressed as a random behavior,
C5) then the full connection layer (406) is input:
xl＝f(ul) (5)
ul＝wlxl-1+bl(6)
wherein u islIs an intermediate quantity of the linear output of the full connection layer, which is equal to the output characteristic diagram x of the previous layerl-1Weight matrix w with full connection layerlMultiplication by then deviationAddition, wlWeight matrix being a full connection layer, blIs a deviation matrix of the fully connected layers,
C6) the output of the fully connected layer is sent to a classification function (407) and a loss function calculation is performed, wherein the formula of the classification function is a standard classification function softmax function, namely:
Figure FDA0002358607190000031
k is the classification number of the data aggregation; z is the output of the pre-stage unit of the classifier; i is the category of the index, e is the representation of the exponential function,
C7) a convergence decision (408) of the loss function is made, wherein: when it is judged that the loss function is not converged, the operation returns to the step C1), i.e., the parameter setting step (402), and updates the weight parameter of the network; when the loss function is judged to be converged, the classification result is directly output (409).
3. The aircraft signal classification and identification method of claim 2, wherein:
a cross-entropy function is used as the loss function,
wherein:
the standard cross entropy loss function is:
Figure FDA0002358607190000032
wherein the content of the first and second substances,
Figure FDA0002358607190000033
calculating labels according to the label classifier; y isiIs a true tag of the aircraft signal; i is the category of the index.
4. The aircraft signal classification and identification method of claim 1, wherein:
the convolution kernel size of the small scale feature extraction tributary is 3,
the swell-convolution expansion ratio of the small-scale tributary is 2,
the convolution kernel size of the large scale feature extraction tributary is 5,
the large scale feature extracts the expansion ratio of the substream 5.
5. The aircraft signal classification and identification method of claim 1, wherein:
the training of the multi-scale residual convolution network is carried out by adopting an Adam training method, the learning rate of the training is set to be 0.001, and the hyper-parameter β of an Adam optimizer1And β2Set to 0.9 and 0.999 respectively.
6. Storage medium having stored thereon a computer program enabling a processor to execute the method according to one of claims 1 to 5.
CN202010015165.1A 2020-01-07 2020-01-07 Classification identification and fault detection method for multi-scale signals of aircraft Active CN111414932B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010015165.1A CN111414932B (en) 2020-01-07 2020-01-07 Classification identification and fault detection method for multi-scale signals of aircraft

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010015165.1A CN111414932B (en) 2020-01-07 2020-01-07 Classification identification and fault detection method for multi-scale signals of aircraft

Publications (2)

Publication Number Publication Date
CN111414932A true CN111414932A (en) 2020-07-14
CN111414932B CN111414932B (en) 2022-05-31

Family

ID=71493969

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010015165.1A Active CN111414932B (en) 2020-01-07 2020-01-07 Classification identification and fault detection method for multi-scale signals of aircraft

Country Status (1)

Country Link
CN (1) CN111414932B (en)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112101374A (en) * 2020-08-01 2020-12-18 西南交通大学 Unmanned aerial vehicle obstacle detection method based on SURF feature detection and ISODATA clustering algorithm
CN112348124A (en) * 2021-01-05 2021-02-09 北京航空航天大学 Data-driven micro fault diagnosis method and device
CN112926547A (en) * 2021-04-13 2021-06-08 北京航空航天大学 Small sample transfer learning method for classifying and identifying aircraft electric signals
CN113361042A (en) * 2021-06-22 2021-09-07 石家庄铁道大学 Method for evaluating service life of rolling bearing
CN113470831A (en) * 2021-09-03 2021-10-01 武汉泰乐奇信息科技有限公司 Big data conversion method and device based on data degeneracy
CN113469286A (en) * 2021-07-27 2021-10-01 北京信息科技大学 Spacecraft multi-local component detection method based on regional convolutional neural network
CN114689122A (en) * 2022-03-31 2022-07-01 国网北京市电力公司 Equipment fault monitoring method, device, equipment and medium
CN116738339A (en) * 2023-06-09 2023-09-12 北京航空航天大学 Multi-classification deep learning recognition detection method for small-sample electric signals

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108710826A (en) * 2018-04-13 2018-10-26 燕山大学 A kind of traffic sign deep learning mode identification method
CN109141847A (en) * 2018-07-20 2019-01-04 上海工程技术大学 A kind of aircraft system faults diagnostic method based on MSCNN deep learning
CN109214250A (en) * 2017-07-05 2019-01-15 中南大学 A kind of static gesture identification method based on multiple dimensioned convolutional neural networks
CN110059582A (en) * 2019-03-28 2019-07-26 东南大学 Driving behavior recognition methods based on multiple dimensioned attention convolutional neural networks

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109214250A (en) * 2017-07-05 2019-01-15 中南大学 A kind of static gesture identification method based on multiple dimensioned convolutional neural networks
CN108710826A (en) * 2018-04-13 2018-10-26 燕山大学 A kind of traffic sign deep learning mode identification method
CN109141847A (en) * 2018-07-20 2019-01-04 上海工程技术大学 A kind of aircraft system faults diagnostic method based on MSCNN deep learning
CN110059582A (en) * 2019-03-28 2019-07-26 东南大学 Driving behavior recognition methods based on multiple dimensioned attention convolutional neural networks

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
YIXIN LIU: "MRD-Nets: Multi-Scale Residual Networks With Dilated Convolutions for Classification and Clustering Analysis of Spacecraft Electrical Signal", 《IEEE ACCESS》 *
YIXIN LIU: "MRD-Nets: Multi-Scale Residual Networks With Dilated Convolutions for Classification and Clustering Analysis of Spacecraft Electrical Signal", 《IEEE ACCESS》, 18 October 2019 (2019-10-18), pages 171584 - 171597 *

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112101374A (en) * 2020-08-01 2020-12-18 西南交通大学 Unmanned aerial vehicle obstacle detection method based on SURF feature detection and ISODATA clustering algorithm
CN112348124A (en) * 2021-01-05 2021-02-09 北京航空航天大学 Data-driven micro fault diagnosis method and device
CN112348124B (en) * 2021-01-05 2021-04-20 北京航空航天大学 Data-driven micro fault diagnosis method and device
CN112926547A (en) * 2021-04-13 2021-06-08 北京航空航天大学 Small sample transfer learning method for classifying and identifying aircraft electric signals
CN112926547B (en) * 2021-04-13 2022-03-29 北京航空航天大学 Small sample transfer learning method for classifying and identifying aircraft electric signals
CN113361042A (en) * 2021-06-22 2021-09-07 石家庄铁道大学 Method for evaluating service life of rolling bearing
CN113469286A (en) * 2021-07-27 2021-10-01 北京信息科技大学 Spacecraft multi-local component detection method based on regional convolutional neural network
CN113470831A (en) * 2021-09-03 2021-10-01 武汉泰乐奇信息科技有限公司 Big data conversion method and device based on data degeneracy
CN113470831B (en) * 2021-09-03 2021-11-16 武汉泰乐奇信息科技有限公司 Big data conversion method and device based on data degeneracy
CN114689122A (en) * 2022-03-31 2022-07-01 国网北京市电力公司 Equipment fault monitoring method, device, equipment and medium
CN114689122B (en) * 2022-03-31 2023-11-10 国网北京市电力公司 Equipment fault monitoring method, device, equipment and medium
CN116738339A (en) * 2023-06-09 2023-09-12 北京航空航天大学 Multi-classification deep learning recognition detection method for small-sample electric signals

Also Published As

Publication number Publication date
CN111414932B (en) 2022-05-31

Similar Documents

Publication Publication Date Title
CN111414932B (en) Classification identification and fault detection method for multi-scale signals of aircraft
CN112926547B (en) Small sample transfer learning method for classifying and identifying aircraft electric signals
CN108596203B (en) Optimization method of parallel pooling layer for pantograph carbon slide plate surface abrasion detection model
CN110163069B (en) Lane line detection method for driving assistance
CN111985533B (en) Incremental underwater sound signal identification method based on multi-scale information fusion
CN111368920A (en) Quantum twin neural network-based binary classification method and face recognition method thereof
CN111768000A (en) Industrial process data modeling method for online adaptive fine-tuning deep learning
CN113486578A (en) Method for predicting residual life of equipment in industrial process
CN113157678B (en) Multi-source heterogeneous data association method
CN112766303A (en) CNN-based aeroengine fault diagnosis method
CN114544155A (en) AUV propeller multi-information-source fusion fault diagnosis method and system based on deep learning
CN112985830A (en) Abs result automatic judging algorithm
CN114754973A (en) Wind tunnel force measurement test data intelligent diagnosis and analysis method based on machine learning
CN115048537A (en) Disease recognition system based on image-text multi-mode collaborative representation
CN116263849A (en) Injection molding process parameter processing method and device and computing equipment
CN114974306A (en) Transformer abnormal voiceprint detection and identification method and device based on deep learning
CN116738332A (en) Aircraft multi-scale signal classification recognition and fault detection method combining attention mechanism
CN109583456B (en) Infrared surface target detection method based on feature fusion and dense connection
CN114357372A (en) Aircraft fault diagnosis model generation method based on multi-sensor data driving
CN112785479B (en) Image invisible watermark universal detection method based on few sample learning
CN113858566A (en) Injection molding machine energy consumption prediction method and system based on machine learning
CN116977834A (en) Method for identifying internal and external images distributed under open condition
CN116070985A (en) Dangerous chemical vehicle loading and unloading process identification method
CN116168235A (en) Hyperspectral image classification method based on double-branch attention network
CN114358189A (en) Hydraulic waterproof valve fault diagnosis method based on multi-mode deep residual shrinkage 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
GR01 Patent grant
GR01 Patent grant