CN116361728B - Civil aircraft system level abnormal precursor identification method based on real-time flight data - Google Patents
Civil aircraft system level abnormal precursor identification method based on real-time flight data Download PDFInfo
- Publication number
- CN116361728B CN116361728B CN202310244222.7A CN202310244222A CN116361728B CN 116361728 B CN116361728 B CN 116361728B CN 202310244222 A CN202310244222 A CN 202310244222A CN 116361728 B CN116361728 B CN 116361728B
- Authority
- CN
- China
- Prior art keywords
- precursor
- parameter
- flight
- data
- abnormal
- 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.)
- Active
Links
- 239000002243 precursor Substances 0.000 title claims abstract description 236
- 230000002159 abnormal effect Effects 0.000 title claims abstract description 74
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000012549 training Methods 0.000 claims abstract description 56
- 238000012360 testing method Methods 0.000 claims description 22
- 230000003993 interaction Effects 0.000 claims description 19
- 230000006870 function Effects 0.000 claims description 18
- 230000005856 abnormality Effects 0.000 claims description 16
- 230000008569 process Effects 0.000 claims description 13
- 230000015556 catabolic process Effects 0.000 claims description 11
- 230000001364 causal effect Effects 0.000 claims description 11
- 238000006731 degradation reaction Methods 0.000 claims description 11
- 230000006866 deterioration Effects 0.000 claims description 8
- 238000012795 verification Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 7
- 230000009471 action Effects 0.000 claims description 5
- 230000002776 aggregation Effects 0.000 claims description 5
- 238000004220 aggregation Methods 0.000 claims description 5
- 238000004140 cleaning Methods 0.000 claims description 5
- 238000000556 factor analysis Methods 0.000 claims description 5
- 238000007619 statistical method Methods 0.000 claims description 4
- 238000011176 pooling Methods 0.000 claims description 3
- 230000000306 recurrent effect Effects 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 claims description 2
- 238000002372 labelling Methods 0.000 claims description 2
- 230000017105 transposition Effects 0.000 claims description 2
- 238000001514 detection method Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 7
- 238000010586 diagram Methods 0.000 description 6
- 238000000540 analysis of variance Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 229920002430 Fibre-reinforced plastic Polymers 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 230000000903 blocking effect Effects 0.000 description 2
- 239000002131 composite material Substances 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 239000011151 fibre-reinforced plastic Substances 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 230000036962 time dependent Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000010339 dilation Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000001939 inductive effect Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000036961 partial effect Effects 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 230000002829 reductive effect Effects 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Classifications
-
- 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
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method for identifying system-level abnormal precursors of a civil aircraft based on real-time flight data, which comprises the following steps: constructing an abnormal precursor identification network, wherein the abnormal precursor identification network comprises a multi-head time domain convolution network self-encoder and a multi-example learning classifier; offline training is carried out on the abnormal precursor recognition network through historical flight data to obtain a precursor recognition network and a quantized risk value sequence; based on the quantized risk value sequence, obtaining precursor duration characteristics and precursor cause parameter sets; adjusting the precursor identification network based on the precursor duration characteristics and the precursor cause parameter set, and training the adjusted precursor identification network to obtain a precursor identifier; and identifying the real-time flight data through a precursor identifier to obtain a current precursor quantitative risk value so as to realize the identification of abnormal precursors of the civil aircraft system level. Through the technical scheme, means can be provided for preventing serious risks of civil aviation operation.
Description
Technical Field
The invention relates to the technical field of civil aviation safety technology and flight data application, in particular to a method for identifying system level abnormality of a civil aviation aircraft on line based on real-time flight data.
Background
As the size of civil aircraft fleets expands, the pressure of safe operation in civil aviation continues to increase. The flight quality monitoring is to acquire flight data after completing each flight mission, further screen abnormal events in flight, and manually analyze and summarize abnormal causes if necessary. Because the flight data is obtained after the flight mission is completed, the current flight quality monitoring is essentially a reactive safety management means based on post-flight analysis of the flight data, and is incapable of preventing sudden system-level anomalies of the civil aircraft such as engine air parking, control surface blocking, icing, stall and the like.
The real-time flight data is that the flight parameters are downloaded at high speed in real time by adopting an air-ground communication technology, so that the on-line recognition of the system-level abnormality of the civil aircraft and the adoption of further early warning prompt measures are possible. With the application of real-time flight data, the flight quality monitoring technology upgrades the data anomaly detection after the navigation to online real-time anomaly detection.
The abnormal precursor is the cause of flight abnormality, and the precursor characteristics presented by the abnormal precursor are obviously different and have uncertainty aiming at different civil aircraft system level abnormalities. The expression is as follows: (1) Different anomaly precursors, which have different associated flight parameters. (2) Precursor features may be characterized by a number of parameter morphology combinations; (3) The flight parameters associated with the precursors have uncertainties in the time of occurrence and the time span (duration), and so on.
On one hand, the uncertainty of the abnormal precursor can not be directly applied to online real-time detection of the abnormal precursor by various abnormal detection algorithms based on post-flight data; on the other hand, it is required to learn the characteristics of the anomaly precursor offline for the civil aircraft system level anomaly and to detect the anomaly precursor online from real-time flight data.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a system-level abnormal precursor identification method of a civil aircraft based on real-time flight data, and designs an offline/online combined precursor identification architecture. In an offline stage, precursor key characteristics of historical flight data are acquired through a precursor identification network, so that a precursor identification network architecture is adjusted, and a system-level abnormal precursor online identification network facing real-time flight data is acquired. After the offline training is completed, real-time flight data is intercepted through the obtained precursor evolution window in a long sliding mode, and the real-time flight data is used for online identification of whether abnormal precursors exist or not.
In order to achieve the technical purpose, the invention provides the following technical scheme: a method for identifying abnormal precursors of a civil aircraft system level based on real-time flight data comprises the following steps:
constructing an abnormal precursor identification network, wherein the abnormal precursor identification network comprises a multi-head time domain convolution network self-encoder and a multi-example learning classifier;
acquiring historical flight data, and performing offline training on an abnormal precursor identification network through the historical flight data to obtain a precursor identification network and a quantized risk value sequence; carrying out statistical analysis based on the quantized risk value sequence to obtain precursor duration characteristics and precursor cause parameter sets;
adjusting the precursor identification network based on the precursor duration characteristics and the precursor incentive parameter set, and training the adjusted precursor identification network through the re-intercepted training sample to obtain a precursor identifier;
according to the precursor duration characteristics and the precursor cause parameter set, acquiring real-time flight data of the civil aircraft system, and identifying the real-time flight data through a precursor identifier to obtain a current precursor quantitative risk value so as to realize recognition of abnormal precursors of the civil aircraft system level.
Optionally, in the abnormal precursor identification network, the multi-head time domain convolutional network self-encoder includes a plurality of time domain convolutional network heads, wherein the time domain convolutional network heads include a first causal expansion convolutional layer, a first BN layer, a first ReLU layer, a first Dropout layer, a second causal expansion convolutional layer, a second BN layer, a second Dropout layer, and a second ReLU layer, and an addition result of an input sequence of the first causal expansion convolutional layer and an output sequence of the second Dropout layer is input to the second ReLU layer;
the multi-head time domain convolution network self-encoder is also connected with a corresponding decoder, wherein a first causal expansion convolution layer and a second causal expansion convolution layer in the encoder structure are respectively replaced by a transposition convolution layer to form the decoder;
and processing and combining the model input data through a multi-head time domain convolution network self-encoder to generate a multi-parameter feature map, and processing the multi-parameter feature map through a corresponding decoder to obtain a reconstruction parameter sequence.
Optionally, in the abnormal precursor recognition network, the multi-example learning classifier includes a recurrent neural network layer, a full-connection layer, a logic layer and an aggregation layer which are sequentially connected, wherein a sigmoid function is adopted as a threshold function of the logic layer, and the aggregation layer adopts a maximum pooling layer; and processing the multi-parameter feature map through the multi-example learning classifier to obtain a quantized risk value.
Optionally, after acquiring the historical flight data, further includes:
data cleaning is carried out on historical flight data: wherein the historical flight data includes normal and system level anomaly data;
wherein d= |λ 1 I, d is the number of parameters, lambda 1 For an initial set of parameters of historical flight data, F' i,j Recording the flight parameters of the jth leg at the moment i,and L is the recorded data length, wherein the recorded value of the kth parameter in the jth leg at the moment i is recorded.
And performing z-score centralization treatment on the data after data cleaning;
wherein,for the recorded values after z-score centering, μ is the overall mean of the recorded values, σ is its standard deviation, m is the point in time in the data record, </u >>The recorded value of the kth parameter in the jth leg at time m.
Dividing the processed data into a pre-training set, a pre-testing set and a pre-verifying set according to normal data and system level abnormal data;
and performing offline training on the abnormal precursor identification network through the pre-training set, the pre-test set and the pre-verification set.
Optionally, in the process of offline training the abnormal precursor recognition network, the training loss function of the abnormal precursor recognition networkIs that
Wherein w is the combining weight of the loss function;
loss function for training multi-head time domain convolutional network self-encoderThe method comprises the following steps:
wherein X is the parameter sequence of the input network, +.>Reconstructing a parameter sequence from the potential space;
loss function for training multiple example learning classifiersThe method comprises the following steps:
wherein y is a tag corresponding to flight data input to the network,>tags that are identified for the network.
Optionally, the acquiring process of the precursor time length feature includes:
outputting a quantized risk value sequence generated by a precursor recognition network in the offline training process;
finding S in a quantized risk value sequence p Precursor time > 0.5 and counting the duration of the precursor time;
and counting the historical flight data in the duration time, and acquiring the 25 th percentile value of the historical flight data in the duration time as a precursor duration characteristic of the flight abnormality.
Optionally, the process of acquiring the precursor incentive parameter set includes:
taking each flight parameter in the initial parameter set of the historical flight data as a control variable, taking the quantized risk value sequence as an observation variable, and decomposing the total variation of the precursor quantized risk value into:
SS T ＝SS A +SS B +SS C +SS AB +SS AC +SS BC +SS ABC +SS E ；
wherein the SS T Total variance of risk values for precursor vectorization; SS (support System) A 、SS B And SS (all-over-all) C Respectively different flight parameters P A 、P B P C Deterioration caused by independent action; SS (support System) AB For flight parameter P A 、P B Deterioration due to interaction of two by two, SS BC Flight parameter P B 、P C Deterioration due to interaction of two by two, SS AC Flight parameter P A 、P C Degradation due to pairwise interactions; SS (support System) ABC For flight parameter P A 、P B 、P C Degradation caused by interactions between; SS (support System) E Degradation due to random factors;
the flight parameters are horizontally divided into a plurality of types, and the variation caused by independent action is as follows:
wherein the flight parameter P A 、P B And P C Respectively have k, r and p levels, n ijl For flight parameter P A Ith level, flight parameters P B Jth level and flight parameter P C Number of samples at the first level; and->Respectively the flight parameters P A 、P B And P C Mean value of precursor quantized risk values at the ith level,/->Representative parameter P A 、P B And P C Precursor quantized risk value means at all levels;
the degradation caused by the random factor is:
wherein,for parameter P A Ith level, parameter P B Jth level and parameter P C Pre-vectorized risk value for the mth sample at the first level, +.>For parameter P A Ith level, parameter P B Precursor quantized risk value mean at jth level,/->For parameter P B Jth level, parameter P C Precursor quantized risk value mean at level l +.>For parameter P A Ith level, parameter P C Precursor quantized risk value mean at level l +.>For parameter P A Ith level, parameter P B Jth level and parameter P C Precursor quantitative risk value mean at the first level;
based on different types of variation, carrying out unitary multi-factor analysis on the initial parameter set; and obtaining a precursor incentive parameter set.
Optionally, the process of unitary multi-factor analysis includes:
acquiring an occurrence moment point and a termination moment point according to the duration of the precursor moment;
intercepting a flight parameter sequence of historical flight data according to the occurrence time point and the termination time point; respectively calculating the average value corresponding to each flight parameter of the flight parameter sequence, and horizontally dividing the average value;
providing an original assumption for F test statistics, wherein under different levels of each flight parameter, the average value of each overall of the flight abnormal precursor is not significantly different, and interaction among the flight parameters does not significantly influence the precursor quantitative risk value;
f test statistic calculation is carried out on each flight parameter:
wherein F is A 、F B 、F C 、F AB 、F BC 、F AC 、F ABC Respectively are parameters P A 、P B 、P C 、P AB 、P BC 、P AC 、P ABC Is a test amount of (a).
If the F test statistic of the flight parameters or the flight parameter combination is larger than the critical value of the confidence interval, the flight parameters or the interaction of the flight parameters larger than the critical value are considered to belong to the abnormal inducement parameters;
and correcting the initial parameter set of the historical flight data through the abnormal incentive parameters to obtain a precursor incentive parameter set.
Optionally, the process of adjusting the precursor identification network includes:
re-acquiring a training parameter sequence based on the precursor time length characteristic and the precursor cause parameter set, wherein the length of the training parameter sequence is equal to the precursor time length characteristic, and the parameter dimension is equal to the precursor cause parameter set dimension; labeling the training parameter sequence according to the quantized risk threshold value to obtain a precursor parameter sequence data set and a normal parameter sequence data set, and dividing the precursor parameter sequence data set and the normal parameter sequence data set into a training set, a testing set and a verification set;
and adjusting the number of input parameters of the multi-head time domain convolutional network self-encoder into the number of parameters of a precursor incentive parameter set, and training the adjusted precursor identification network through a training set, a testing set and a verification set to obtain the precursor identifier.
Optionally, intercepting real-time flight data with the length of the precursor duration feature, wherein the flight data is the same as the parameter types and the number of the precursor incentive parameter sets.
The invention has the following technical effects:
(1) The conventional abnormality detection method represented by overrun detection can detect only an abnormality, and cannot detect a precursor. Since overrun detection uses an empirically set threshold, all anomalies cannot be found. The invention provides a real-time flight data-based online recognition method for system-level abnormal precursors of a civil aircraft, which adopts historical data to extract characteristics of the system-level abnormal precursors, and can further take corrective measures when the precursors are recognized online, so that the abnormality is prevented.
(2) Traditional self-growing online anomaly detection techniques are limited by computational resources and cannot extract precursor features from large amounts of data. The invention provides an on-line recognition method for abnormal precursors combining off-line training and on-line recognition, which utilizes historical data to extract system-level abnormal precursor characteristics of a civil aircraft in an off-line stage to obtain a precursor recognition network, thereby greatly reducing the difficulty of on-line recognition and improving the recognition precision of system-level anomalies.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings that are needed in the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic flow chart of a method for online identifying system-level abnormal precursors of a civil aircraft, which is provided by an embodiment of the invention;
fig. 2 is a schematic diagram of an overall structure of a precursor recognition network according to an embodiment of the present invention;
fig. 3 is a schematic diagram of an MHTCN-AE network structure provided in an embodiment of the present invention;
fig. 4 is a schematic diagram of a network structure of MILC according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The invention provides a method for identifying system-level abnormal precursors of a civil aircraft based on real-time flight data.
S1: an abnormal precursor identification network architecture based on a multi-head time domain convolutional network self-encoder and a multi-example learning classifier is constructed. Extracting abnormal precursor features from historical flight data based on a multi-head time domain convolution network self-encoder; a multi-example learning classifier is designed for identifying a precursor quantized risk value of a system level anomaly.
S2: offline training the abnormal precursor recognition network through historical flight data of normal and system-level abnormality of the civil aircraft to obtain the precursor recognition network, and outputting a quantized risk value S of precursor changing along with time p . Then extracting precursor time length characteristic W and precursor incentive parameter set lambda through statistical analysis of precursor 2 。
S3: based on the duration characteristics of the abnormal precursor and the precursor incentive parameter set, the precursor identification network architecture is adjusted, and the history data is intercepted again to train the precursor identification network, so that the precursor identifier is obtained.
S4: after the off-line training is finished, real-time data is intercepted by a fixed-length sliding window and is input into a precursor identifier, and a precursor quantitative risk value in the current window time is obtained. The invention provides an online recognition means for system level abnormality of a civil aviation aircraft based on real-time flight data, and provides a means for preventing significant risk of civil aviation operation.
The method for identifying the flight anomaly precursor provided by the invention can be suitable for online identification of system-level anomalies of the civil aircraft such as engine air parking, control surface blocking, icing, stall and the like by adjusting historical data required by training.
As shown in fig. 1, the invention provides a method for identifying system-level abnormal precursors of a civil aircraft based on real-time flight data, which comprises the following steps:
s1: as shown in fig. 2, an abnormal precursor identification network architecture based on a multi-head time domain convolutional network self-encoder and a multi-example learning classifier is constructed. Extracting abnormal precursor features from historical flight data based on a multi-head time domain convolution network self-encoder; a multi-example learning classifier is designed for identifying a precursor quantized risk value of a system level anomaly. The method comprises the following specific steps:
s1.1 a Multi-headed time domain convolutional self-encoder (Multi-Head Temporal Convolutional NetworkAutoEncoder, MHTCN-AE) architecture is constructed.
And designing an MHTCN-AE architecture, extracting form and time sequence characteristics of each parameter sequence through a multi-head time domain convolution network TCN, and ensuring that the extracted information can completely retain key characteristics of the parameter sequences by utilizing an automatic encoder AE architecture.
As shown in fig. 3, the input data is X j ＝[F 0,j ,F 1,j ,...,F L-1,j ] T Wherein X is j For the j-th segment of the parameter sequence,a parameter sequence reconstructed for the decoder; f (F) i,j For the j-th section of parameter sequence, the parameter record at the i-th moment is that y is the label of the normal or abnormal section of data,/for the j-th section of parameter sequence>Data identified for the network is normally or abnormally tagged. The morphology and timing characteristics of the parameters are extracted by the causal dilation convolutional layer of TCN. />To specifically input the input data for each TCN header, and (2)>For the recorded value of the kth parameter of the jth parameter sequence at time i,/for the jth parameter sequence>For reconstructing the parameter sequence via the decoder, < > is>And (3) reconstructing values of the kth parameter at the moment i for the jth parameter sequence. For the original parameter sequence, the convolution kernel extracts the current moment T through unidirectional sliding of a time domain windowPrevious history information. After passing through the convolution layer, the BN layer is utilized to carry out batch normalization operation on the sequence so as to reduce internal covariance offset and increase regularization effect. Then, by means of the ReLU activation layer, the nonlinear characteristics of MHTCN-AE are improved. In addition, the Dropout layer is added to randomly discard the information of part of the hidden units to prevent MHTCN-AE overfitting. To prevent the situation of overfitting and gradient disappearance, the model complexity is reduced by means of residual connection, i.e. directly connecting BN layer and ReLU layer after causally expanding the convolution layer, followed by outputting the second Dropout layer +.>Input sequence +.>Added and obtained by the ReLU layerIn order to enhance the capability of the model to extract depth features, a plurality of TCN residual error modules can be arranged to extract multi-level features. The parameter sequence passes through an encoder to obtain a single parameter characteristic diagram. Taking the time sequence and morphological characteristics of parameters to be captured and the coupling relation among the parameters into consideration, carrying out combination processing on the extracted single-parameter characteristic diagram to form an L multiplied by d-dimensional multi-parameter characteristic diagram with d parameters and L duration> For characterizing the mapping function of the encoder, the encoder internal parameter is +.>Reconstructing the parameter sequence from the potential space by means of a decoder, resulting in +.>Wherein p is θ (. Cndot.) is a mapping function characterizing the decoder, withinThe partial parameter is theta, and Z' is the expression of the input parameter sequence in potential space after the input parameter sequence passes through the encoder. C (C) ψ (Z) represents the classification and identification of the potential spatial feature Z after integration by the classifier, C ψ (. Cndot.) is a classification function, and ψ is its internal parameters. The decoder structure is consistent with the encoder, and the convolutional network is only required to be modified into a transposed convolutional network.
S1.2, constructing a Multi-example learning Classifier (MILC) architecture.
As shown in fig. 4, the extracted parameter fusion feature map is processed to obtain time-dependent features of parameters and coupling relations between the parameters, and further, abnormal precursors are obtained by multi-example learning. The time dependence between the parameters is further extracted using a gating loop unit (Gated Recurrent Unit, GRU) and a full connectivity layer (Full Connected Layer, FCL). The multiparameter feature map Z passes through the GRU layer according to the time step length to realize the model many-to-many capability, so that the j-th section of parameter sequence is realizedSegmentation into multiple examples { I } in time step 0,j ,I 1,j ,…,I L-1,j }, wherein->I.e. example I at time k k,j And the parameter sequence information from 0 to k time is contained, and f (·) represents the conversion of input data to an example by the GRU-FCL structure. The example input LOG logic layer is then used to implement the example-to-example probability transition, i.e., to obtain a time-dependent precursor quantized risk value S p . Considering that a plurality of precursors can exist for an abnormality in a section of flight data, and the probability of the precursor inducing the abnormality is changed along with the time, a sigmoid function is selected as a LOG layer threshold function. Finally, a maximum pooling MaxPool layer is selected as an MILC aggregation layer, and the example probability is converted into a packet probability.
S2: offline training the abnormal precursor recognition network through historical flight data of normal and system-level abnormality of the civil aircraft to obtain the precursor recognition network, and outputting the precursor which changes with timeQuantification of risk value S p . Then extracting precursor time length characteristic W and precursor incentive parameter set lambda through statistical analysis of precursor 2 。
S2.1 obtaining a precursor identification network through offline training, and outputting a quantized risk value S of precursor changing along with time p 。
After the MHTCN-AE and MILC are constructed, the historical flight data of the normal and system level abnormality of the civil aircraft are utilized to train the precursor identification network offline. Firstly, data cleaning is carried out on flight data, and an initial parameter set lambda is determined 1 The parameter sequences for obtaining the uniform length are as follows:
wherein d is the number of recording parameters, d= |λ 1 |，λ 1 For an initial set of parameters of historical flight data, F' i,j Recording the flight parameters of the jth leg at the moment i,the recorded value of the kth parameter at the time i is represented by L, which is the recorded data length. The parameter sequence was z-score centered,
for the recorded values after z-score centering, μ is the overall mean of the recorded values, σ is its standard deviation, m is the point in time in the data record, </u >>The recorded value of the kth parameter in the jth leg at time m. After the steps are completed, the normal data set and the system level abnormal data set are respectively divided into a training set, a testing set and a verification set according to the proportion of 8:1:1.
In the network training process, setting a network training loss functionThe method comprises the following steps:
where w is the combining weight of the loss function, typically 0.5 is chosen. Loss functionThe method is used for ensuring that the characteristics extracted by the MHTCN-AE can be reconstructed from potential space so as to avoid the characteristic information loss of original data, as shown in a formula (4),
where X is the sequence of parameters of the input network,is a sequence of parameters reconstructed from the potential space. Loss function->Is used for training the classifying ability of the MILC to normal and abnormal parameter sequences, as shown in a formula (5),
wherein y is a label with normal or abnormal flight data input into the network,tags that are identified for the network. Taking accuracy, F1 fraction and the like as network performance indexes, and after the indexes meet certain standards, considering that the offline training of the precursor recognition network is completed to obtainObtaining a precursor identification network and outputting a quantized risk value S of precursor changing along with time p 。
S2.2 obtaining a precursor quantitative risk value S of historical data through network offline training p And further acquiring the duration characteristics of the precursor and the precursor incentive parameter set by adopting unitary multi-factor analysis of variance.
After offline training is completed based on historical flight data, precursor quantitative risk value S in parameter sequence is searched p Time of > 0.5 and counting the duration of this part of precursor time, i.e. for sequence [ F ] i,j ,F i+1,j ,...,F i+L,j ]Precursor quantized risk value sequence [ S p (F i,j ),S p (F i+1,j ),...,S p (F i+L,j )]If min [ S ] within the duration of i+l to i+m p (F i+l,j ),...,S p (F i+m,j )]> 0.5, the sequence precursor duration is |m-l|. And counting the part of data to obtain the 25 th percentile value of the data as a precursor duration characteristic W of the flight abnormality.
After the precursor quantized risk value sequence of the historical data is acquired, the initial parameter set lambda is calculated 1 And (3) performing one-way multi-factor ANOVA analysis of variance on each parameter. Lambda is shown as 1 The parameters of the system are used as control variables, and the precursor quantitative risk value sequence is used as an observation variable.
With 3 flight parameters (i.e. three control variables) P A 、P B And P C The precursor features of the composition are exemplified, where a, B, C e {1, 2..d }, the total variation of the precursor quantized risk values is decomposed into:
SS T ＝SS A +SS B +SS C +SS AB +SS AC +SS BC +SS ABC +SS E (6)
wherein the SS T Total variance of risk values for precursor vectorization; SS (support System) A 、SS B And SS (all-over-all) C For flight parameter P A 、P B Deterioration caused by independent action; SS (support System) AB For the second interaction term, i.e. flight parameter P A 、P B Deterioration due to interaction of two by two, SS BC And SS (all-over-all) AC And the same is done; SS (support System) ABC For three interactive items, i.e. flight parameters P A 、P B 、P C Degradation caused by interactions between; SS (support System) E Is a degradation caused by random factors. In general, SS is A 、SS B And SS (all-over-all) C The effect of the cause is called the main effect, SS AB 、SS BC 、SS AC And SS (all-over-all) ABC Called N-Way interaction effect, SS E Known as residual degradation.
For the continuous flight parameters, dividing the continuous flight parameters into 5 levels according to 0-5%, 5-25%, 25-75%, 75-95% and 95-100% quantiles of the parameters; for discrete flight parameters, the flight parameters are classified according to the nature types of the flight parameters. SS (support System) A 、SS B And SS (all-over-all) C Is defined as follows:
wherein the flight parameter P A 、P B And P C Respectively have k, r and p levels, n ijl For flight parameter P A Ith level, flight parameters P B Jth level and flight parameter P C Number of samples at the first level; and->Respectively the flight parameters P A 、P B And P C Mean value of precursor quantized risk values at the ith level,/->Representative parameter P A 、P B And P C The risk value means were premonited at all levels. SS (support System) E Is defined as:
wherein,for parameter P A Ith level, parameter P B Jth level and parameter P C Pre-vectorized risk value for the mth sample at the first level, +.>For parameter P A Ith level, parameter P B Precursor quantized risk value mean at jth level,/->For parameter P B Jth level, parameter P C Precursor quantized risk value mean at level l +.>For parameter P A Ith level, parameter P C Precursor quantized risk value mean at level l +.>For parameter P A Ith level, parameter P B Jth level and parameter P C The precursor quantized risk value mean at level l.
Next, a one-way multi-factor ANOVA analysis was performed, comprising the steps of:
first, for continuous flight parameters, the precursor risk value S is locked p Time period more than 0.5, the occurrence time point t of the precursor is obtained p(begin) And termination time point t p(end) The parameter sequence of the same period of the sample data is intercepted, the average value of each flight parameter in all the sample data is calculated, and the sample data is divided into 5 levels according to the quantiles of 0-5%, 5-25%, 25-75%, 75-95% and 95-100% of the parameters.
Secondly, the original assumption H0 is proposed: under different levels of all flight parameters, the average value of all the flight abnormal precursors is not obviously different, and the main effect of all the flight parameters and the interaction effect between the flight parameters are 0 at the same time, namely the interaction between the parameters does not obviously influence the precursor quantitative risk value.
Thirdly, establishing test statistics of each parameter F:
wherein F is A 、F B 、F C 、F AB 、F BC 、F AC 、F ABC Respectively are parameters P A 、P B 、P C 、P AB 、P BC 、P AC 、P ABC Is a test amount of (a).
Finally, if the F test statistic of the flight parameter or the flight parameter combination is greater than the critical value of the 5% confidence interval, the flight parameter or the interaction of the flight parameter is considered to belong to the abnormal inducement parameter. Based on the above steps, for the initial parameter set lambda 1 Correcting to obtain precursor incentive parameter set lambda 2 。
S3: based on the duration characteristics of the abnormal precursor and the precursor incentive parameter set, the precursor identification network architecture is adjusted, and the abnormal precursor identification network is trained by intercepting data again to obtain a precursor identifier.
Based on precursor time length characteristics and precursor incentive parameter set lambda 2 Re-extracting the parameter sequence, wherein the length of the extracted sequence is W, and the parameter dimension d= |lambda is the same as the length of the extracted sequence 2 | a. The invention relates to a method for producing a fibre-reinforced plastic composite. Intercepting a parameter sequence by adopting a sliding window with a fixed window length W, marking data in the window based on a precursor quantized risk threshold value eta=0.5, and when the precursor quantized risk value of any moment of the parameter sequence in the window exceeds eta, namely max ([ S ] p (F i,j ),...,S p (F i+W,j )]) > eta, where F i,j Recording parameters of the ith moment of the jth leg, F i+W,j Recording the parameters of the ith+W moment of the jth leg, and recording the sequence [ F ] i,j ,...,F i+W,j ]Marked as precursor sequence; when none of the pre-quantification risk values of the data within the window exceeds the threshold, i.e. max ([ S ] p (F i,j ),...,S p (F i+W,j )]) And (3) marking the sequence as a normal sequence. Thereby retrieving the precursor parameter sequence data set and the normal parameter sequence data set and dividing the data set into a training set, a test set and a validation set according to a ratio of 8:1:1.
At the same time, modify MHTCN-AE architecture to adjust the number of multi-head TCNs to |lambda 2 | a. The invention relates to a method for producing a fibre-reinforced plastic composite. And then training the network again based on the newly divided training set, test set and verification set, and obtaining the precursor identifier after the network performance index meets the requirement.
S4: after the off-line training is finished, real-time data is intercepted by a fixed-length sliding window and is input into a precursor identifier, and a precursor quantitative risk value in the current window time is obtained.
After the offline training is completed, the obtained precursor identifier can be used for real-time precursor identification, and the online identification stage only needs to carry out forward propagation because the precursor identifier has completed the reverse propagation super-parameter training in the offline stage. Intercepting real-time flight data by a fixed-length sliding window W, and respectively converting the data into I lambda 2 The sequences of the I parameters are respectively input into the multi-head network, and meanwhile, the fixed-length window W is slid to intercept data at the later moment, so that whether the data in the current window is a precursor or not is recognized online.
The foregoing has shown and described the basic principles, principal features and advantages of the invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, and that the above embodiments and descriptions are merely illustrative of the principles of the present invention, and various changes and modifications may be made without departing from the spirit and scope of the invention, which is defined in the appended claims. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (7)
1. The method for identifying the system-level abnormal precursor of the civil aircraft based on the real-time flight data is characterized by comprising the following steps of:
constructing an abnormal precursor identification network, wherein the abnormal precursor identification network comprises a multi-head time domain convolution network self-encoder and a multi-example learning classifier;
acquiring historical flight data, and performing offline training on an abnormal precursor identification network through the historical flight data to obtain a precursor identification network and a quantized risk value sequence; carrying out statistical analysis based on the quantized risk value sequence to obtain precursor duration characteristics and precursor cause parameter sets;
adjusting the precursor identification network based on the precursor duration characteristics and the precursor incentive parameter set, and training the adjusted precursor identification network through the re-intercepted training sample to obtain a precursor identifier;
according to the precursor duration characteristics and the precursor cause parameter set, acquiring real-time flight data of the civil aircraft system, and identifying the real-time flight data through a precursor identifier to obtain a current precursor quantitative risk value so as to realize system-level abnormal precursor identification of the civil aircraft;
the acquisition process of the precursor time length feature comprises the following steps:
outputting a quantized risk value sequence generated by a precursor recognition network in the offline training process;
finding S in a quantized risk value sequence p Precursor time > 0.5 and counting the duration of the precursor time;
counting the historical flight data in the duration time, and acquiring a 25 th percentile value of the historical flight data in the duration time as a precursor duration characteristic of the flight abnormality;
the acquisition process of the precursor incentive parameter set comprises the following steps:
taking each flight parameter in the initial parameter set of the historical flight data as a control variable, taking the quantized risk value sequence as an observation variable, and decomposing the total variation of the precursor quantized risk value into:
SS T ＝SS A +SS B +SS C +SS AB +SS AC +SS BC +SS ABC +SS E ；
wherein the SS T Total variance of risk values for precursor vectorization; SS (support System) A 、SS B And SS (all-over-all) C Respectively different flight parameters P A 、P B P C Deterioration caused by independent action; SS (support System) AB For flight parameter P A 、P B Deterioration due to interaction of two by two, SS BC Flight parameter P B 、P C Deterioration due to interaction of two by two, SS AC Flight parameter P A 、P C Degradation due to pairwise interactions; SS (support System) ABC For flight parameter P A 、P B 、P C Degradation caused by interactions between; SS (support System) E Degradation due to random factors;
the flight parameters are horizontally divided into a plurality of types, and the variation caused by independent action is as follows:
wherein the flight parameter P A 、P B And P C Respectively have k, r and p levels, n ijl For flight parameter P A Ith level, flight parameters P B Jth level and flight parameter P C Number of samples at the first level;and->Respectively the flight parameters P A 、P B And P C Mean value of precursor quantized risk values at the ith level,/->Representative parameter P A 、P B And P C Precursor quantized risk value means at all levels;
the degradation caused by the random factor is:
wherein,for parameter P A Ith level, parameter P B Jth level and parameter P C Pre-vectorized risk value for the mth sample at the first level, +.>For parameter P A Ith level, parameter P B Precursor quantized risk value mean at jth level,/->For parameter P B Jth level, parameter P C Precursor quantized risk value mean at level l +.>For parameter P A Ith level, parameter P C Precursor quantized risk value mean at level l +.>For parameter P A Ith level, parameter P B Jth level and parameter P C Precursor quantitative risk value mean at the first level;
based on different types of variation, carrying out unitary multi-factor analysis on the initial parameter set; obtaining a precursor incentive parameter set;
the process of unitary multi-factor analysis includes:
acquiring an occurrence moment point and a termination moment point according to the duration of the precursor moment;
intercepting a flight parameter sequence of historical flight data according to the occurrence time point and the termination time point; respectively calculating the average value corresponding to each flight parameter of the flight parameter sequence, and horizontally dividing the average value;
providing an original assumption for F test statistics, wherein under different levels of each flight parameter, the average value of each overall of the flight abnormal precursor is not significantly different, and interaction among the flight parameters does not significantly influence the precursor quantitative risk value;
f test statistic calculation is carried out on each flight parameter:
wherein F is A 、F B 、F C 、F AB 、F BC 、F AC 、F ABC Respectively are parameters P A 、P B 、P C 、P AB 、P BC 、P AC 、P ABC Is a test amount of (2);
if the F test statistic of the flight parameters or the flight parameter combination is larger than the critical value of the confidence interval, the flight parameters or the interaction of the flight parameters larger than the critical value are considered to belong to the abnormal inducement parameters;
and correcting the initial parameter set of the historical flight data through the abnormal incentive parameters to obtain a precursor incentive parameter set.
2. The abnormal precursor recognition method according to claim 1, wherein:
in an abnormal precursor identification network, a multi-head time domain convolution network self-encoder comprises a plurality of time domain convolution network heads, wherein the time domain convolution network heads comprise a first causal expansion convolution layer, a first BN layer, a first ReLU layer, a first Dropout layer, a second causal expansion convolution layer, a second BN layer, a second Dropout layer and a second ReLU layer which are sequentially connected, and the addition result of an input sequence of the first causal expansion convolution layer and an output sequence of the second Dropout layer is input to the second ReLU layer;
the multi-head time domain convolution network self-encoder is also connected with a corresponding decoder, wherein a first causal expansion convolution layer and a second causal expansion convolution layer in the encoder structure are respectively replaced by a transposition convolution layer to form the decoder;
and processing and combining the model input data through a multi-head time domain convolution network self-encoder to generate a multi-parameter feature map, and processing the multi-parameter feature map through a corresponding decoder to obtain a reconstruction parameter sequence.
3. The abnormal precursor recognition method according to claim 2, wherein:
in an abnormal precursor recognition network, the multi-example learning classifier comprises a recurrent neural network layer, a full-connection layer, a logic layer and an aggregation layer which are sequentially connected, wherein a sigmoid function is adopted as a threshold function of the logic layer, and the aggregation layer adopts a maximum pooling layer; and processing the multi-parameter feature map through the multi-example learning classifier to obtain a quantized risk value.
4. The abnormal precursor recognition method according to claim 1, wherein:
the historical flight data acquisition further comprises the following steps:
data cleaning is carried out on historical flight data: wherein the historical flight data includes normal and system level anomaly data;
wherein d= |λ 1 I, d is the number of parameters, lambda 1 For an initial set of parameters of historical flight data, F' i,j Recording the flight parameters of the jth leg at the moment i,the recorded value of the kth parameter in the jth leg at the moment i is recorded, and L is the length of recorded data;
and performing z-score centralization treatment on the data after data cleaning;
wherein,for the recorded values after z-score centering, μ is the overall mean of the recorded values, σ is its standard deviation, m is the point in time in the data record, </u >>The recorded value of the kth parameter in the jth leg at the moment m is used as a record value;
dividing the processed data into a pre-training set, a pre-testing set and a pre-verifying set according to normal data and system level abnormal data;
and performing offline training on the abnormal precursor identification network through the pre-training set, the pre-test set and the pre-verification set.
5. The abnormal precursor recognition method according to claim 1, wherein:
in the process of offline training of an abnormal precursor recognition network, training loss function of the abnormal precursor recognition networkIs that
Wherein w is the combining weight of the loss function;
loss function for training multi-head time domain convolutional network self-encoderThe method comprises the following steps:
wherein X is the parameter sequence of the input network, +.>Reconstructing a parameter sequence from the potential space;
loss function for training multiple example learning classifiersThe method comprises the following steps:
wherein y is a tag corresponding to flight data input to the network,>tags that are identified for the network.
6. The abnormal precursor recognition method according to claim 1, wherein:
the process of adjusting the precursor identification network comprises:
re-acquiring a training parameter sequence based on the precursor time length characteristic and the precursor cause parameter set, wherein the length of the training parameter sequence is equal to the precursor time length characteristic, and the parameter dimension is equal to the precursor cause parameter set dimension; labeling the training parameter sequence according to the quantized risk threshold value to obtain a precursor parameter sequence data set and a normal parameter sequence data set, and dividing the precursor parameter sequence data set and the normal parameter sequence data set into a training set, a testing set and a verification set;
and adjusting the number of input parameters of the multi-head time domain convolutional network self-encoder into the number of parameters of a precursor incentive parameter set, and training the adjusted precursor identification network through a training set, a testing set and a verification set to obtain the precursor identifier.
7. The abnormal precursor recognition method according to claim 1, wherein: and intercepting real-time flight data by the length of the precursor time length characteristic, wherein the flight data is the same as the parameter types and the number of the precursor incentive parameter sets.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310244222.7A CN116361728B (en) | 2023-03-14 | 2023-03-14 | Civil aircraft system level abnormal precursor identification method based on real-time flight data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310244222.7A CN116361728B (en) | 2023-03-14 | 2023-03-14 | Civil aircraft system level abnormal precursor identification method based on real-time flight data |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116361728A CN116361728A (en) | 2023-06-30 |
| CN116361728B true CN116361728B (en) | 2024-01-23 |
Family
ID=86927150
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310244222.7A Active CN116361728B (en) | 2023-03-14 | 2023-03-14 | Civil aircraft system level abnormal precursor identification method based on real-time flight data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116361728B (en) |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014093670A1 (en) * | 2012-12-12 | 2014-06-19 | University Of North Dakota | Analyzing flight data using predictive models |
| EP3460611A1 (en) * | 2017-09-25 | 2019-03-27 | The Boeing Company | System and method for aircraft fault detection |
| CN109738939A (en) * | 2019-03-21 | 2019-05-10 | 蔡寅 | A kind of Precursory Observational Data method for detecting abnormality |
| CN110427419A (en) * | 2019-07-29 | 2019-11-08 | 陕西千山航空电子有限责任公司 | A kind of flying quality anomalous identification model building method based on data mining |
| CN113033621A (en) * | 2021-03-05 | 2021-06-25 | 南京航空航天大学 | Method for identifying unstable approach and inducement thereof of civil aircraft |
| CN113761933A (en) * | 2021-05-18 | 2021-12-07 | 腾讯科技（深圳）有限公司 | Retrieval method, retrieval device, electronic equipment and readable storage medium |
| WO2022160902A1 (en) * | 2021-01-28 | 2022-08-04 | 广西大学 | Anomaly detection method for large-scale multivariate time series data in cloud environment |
| CN115169386A (en) * | 2022-06-17 | 2022-10-11 | 东北大学 | Weak supervision increasing activity identification method based on meta-attention mechanism |
| CN115270926A (en) * | 2022-06-27 | 2022-11-01 | 温州大学 | Airplane flight abnormity detection method and system |
| CN115510950A (en) * | 2022-08-26 | 2022-12-23 | 北京机电工程研究所 | Aircraft telemetry data anomaly detection method and system based on time convolution network |
| CN115550072A (en) * | 2022-11-30 | 2022-12-30 | 浙江省能源集团有限公司 | Network attack monitoring and early warning method and system |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200050182A1 (en) * | 2018-08-07 | 2020-02-13 | Nec Laboratories America, Inc. | Automated anomaly precursor detection |
-
2023
- 2023-03-14 CN CN202310244222.7A patent/CN116361728B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014093670A1 (en) * | 2012-12-12 | 2014-06-19 | University Of North Dakota | Analyzing flight data using predictive models |
| EP3460611A1 (en) * | 2017-09-25 | 2019-03-27 | The Boeing Company | System and method for aircraft fault detection |
| CN109738939A (en) * | 2019-03-21 | 2019-05-10 | 蔡寅 | A kind of Precursory Observational Data method for detecting abnormality |
| CN110427419A (en) * | 2019-07-29 | 2019-11-08 | 陕西千山航空电子有限责任公司 | A kind of flying quality anomalous identification model building method based on data mining |
| WO2022160902A1 (en) * | 2021-01-28 | 2022-08-04 | 广西大学 | Anomaly detection method for large-scale multivariate time series data in cloud environment |
| CN113033621A (en) * | 2021-03-05 | 2021-06-25 | 南京航空航天大学 | Method for identifying unstable approach and inducement thereof of civil aircraft |
| CN113761933A (en) * | 2021-05-18 | 2021-12-07 | 腾讯科技（深圳）有限公司 | Retrieval method, retrieval device, electronic equipment and readable storage medium |
| CN115169386A (en) * | 2022-06-17 | 2022-10-11 | 东北大学 | Weak supervision increasing activity identification method based on meta-attention mechanism |
| CN115270926A (en) * | 2022-06-27 | 2022-11-01 | 温州大学 | Airplane flight abnormity detection method and system |
| CN115510950A (en) * | 2022-08-26 | 2022-12-23 | 北京机电工程研究所 | Aircraft telemetry data anomaly detection method and system based on time convolution network |
| CN115550072A (en) * | 2022-11-30 | 2022-12-30 | 浙江省能源集团有限公司 | Network attack monitoring and early warning method and system |
Non-Patent Citations (4)
| Title |
|---|
| Flight Anomaly Detection via a Deep Hybrid Model;Kun Qin et al.;aerospace;全文 * |
| 基于深度学习的航空器异常飞行状态识别;吴奇;储银雪;;民用飞机设计与研究(03);全文 * |
| 基于稀疏自动编码器的飞参数据异常检测;杜辰飞;曲建岭;孙文柱;高峰;;计测技术(02);全文 * |
| 无人机飞行数据异常检测算法综述;祁超帅等;计算机应用;第43卷(第6期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116361728A (en) | 2023-06-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110555479B (en) | Fault feature learning and classifying method based on 1DCNN and GRU fusion | |
| CN108536123B (en) | Train control on board equipment method for diagnosing faults based on long Memory Neural Networks in short-term | |
| CN110334741B (en) | Radar one-dimensional range profile identification method based on cyclic neural network | |
| CN111368885B (en) | Gas circuit fault diagnosis method for aircraft engine | |
| CN111337768A (en) | Deep parallel fault diagnosis method and system for dissolved gas in transformer oil | |
| CN112200244A (en) | Intelligent detection method for anomaly of aerospace engine based on hierarchical countermeasure training | |
| CN111709292A (en) | Compressor vibration fault detection method based on recursive graph and deep convolutional network | |
| CN114176607B (en) | Electroencephalogram signal classification method based on vision transducer | |
| CN111783534A (en) | Sleep staging method based on deep learning | |
| Cao et al. | Bearing fault diagnosis method based on GMM and coupled hidden Markov model | |
| CN116894215B (en) | Gear box fault diagnosis method based on semi-supervised dynamic graph attention | |
| Xu et al. | Global attention mechanism based deep learning for remaining useful life prediction of aero-engine | |
| CN111310719B (en) | Unknown radiation source individual identification and detection method | |
| CN115017939A (en) | Intelligent diagnosis method and device for faults of aircraft fuel pump and storage medium | |
| CN113409213B (en) | Method and system for enhancing noise reduction of time-frequency diagram of fault signal of plunger pump | |
| CN116361728B (en) | Civil aircraft system level abnormal precursor identification method based on real-time flight data | |
| CN116977834B (en) | Method for identifying internal and external images distributed under open condition | |
| CN117592543A (en) | Aeroengine rolling bearing fault diagnosis method based on self-supervision learning | |
| CN112861443A (en) | Advanced learning fault diagnosis method integrated with priori knowledge | |
| CN110231165B (en) | Mechanical equipment fault diagnosis method based on expectation difference constraint confidence network | |
| CN117093924A (en) | Rotary machine variable working condition fault diagnosis method based on domain adaptation characteristics | |
| CN111461184A (en) | XGB multi-dimensional operation and maintenance data anomaly detection method based on multivariate feature matrix | |
| CN115048873B (en) | Residual service life prediction system for aircraft engine | |
| CN116502696A (en) | Rolling bearing residual life prediction method based on federal learning and model pruning | |
| CN114004144A (en) | Method for predicting remaining service life of aircraft engine |
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 |