CN111381584A - Aircraft cabin abnormal fault detection method based on two-stage gated loop network associative memory - Google Patents

Aircraft cabin abnormal fault detection method based on two-stage gated loop network associative memory Download PDF

Info

Publication number
CN111381584A
CN111381584A CN202010217602.8A CN202010217602A CN111381584A CN 111381584 A CN111381584 A CN 111381584A CN 202010217602 A CN202010217602 A CN 202010217602A CN 111381584 A CN111381584 A CN 111381584A
Authority
CN
China
Prior art keywords
observer
training
sequence
association
feedback
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202010217602.8A
Other languages
Chinese (zh)
Inventor
陶来发
马梁
杨帆
吕琛
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 CN202010217602.8A priority Critical patent/CN111381584A/en
Publication of CN111381584A publication Critical patent/CN111381584A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B23/00—Testing or monitoring of control systems or parts thereof
    • G05B23/02—Electric testing or monitoring
    • G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
    • G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
    • G05B23/0243—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults model based detection method, e.g. first-principles knowledge model
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • 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
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B64—AIRCRAFT; AVIATION; COSMONAUTICS
    • B64D—EQUIPMENT FOR FITTING IN OR TO AIRCRAFT; FLIGHT SUITS; PARACHUTES; ARRANGEMENTS OR MOUNTING OF POWER PLANTS OR PROPULSION TRANSMISSIONS IN AIRCRAFT
    • B64D13/00—Arrangements or adaptations of air-treatment apparatus for aircraft crew or passengers, or freight space, or structural parts of the aircraft
    • B64D13/06—Arrangements or adaptations of air-treatment apparatus for aircraft crew or passengers, or freight space, or structural parts of the aircraft the air being conditioned
    • B64D2013/0603—Environmental Control Systems
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B2219/00—Program-control systems
    • G05B2219/20—Pc systems
    • G05B2219/24—Pc safety
    • G05B2219/24065—Real time diagnostics

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Automation & Control Theory (AREA)
  • Testing And Monitoring For Control Systems (AREA)

Abstract

The invention discloses an aircraft cockpit abnormal fault detection method based on association memory of a two-stage gated loop network, which comprises the following specific steps of: the method comprises the following steps: acquiring time sequence data and delaying slicing processing; step two: establishing and training a primary GRU network association observer; step three: obtaining observation residual and processing slices; step four: establishing and training a secondary GRU network association observer; step five: and detecting abnormal faults of the airplane cockpit. According to the invention, the dynamic residual threshold output by the associated observer is adopted to replace a fixed threshold, so that the self-adaptability of fault detection is effectively improved; learning the front-back dependency relationship among various time sequence signals by using a gated cycle network with a cycle structure, and fully mining the time sequence fault characteristics; the multiple redundant sensors are adopted to judge the abnormality together, so that the robustness of a fault detection result is improved; the method can effectively detect the abnormal faults of the environmental parameters of the aircraft cabin, and has higher practical engineering application value.

Description

Aircraft cabin abnormal fault detection method based on two-stage gated loop network associative memory
Technical Field
The invention relates to the field of fault detection of an aircraft environmental control system, in particular to an aircraft cabin abnormal fault detection method based on association memory of a two-stage gated circulation network.
Background
The airplane environment control system is an important airborne system of an airplane, and is used for providing a comfortable living environment for onboard personnel, mainly adjusting parameters such as temperature, humidity and pressure in an airplane cabin. Whether the environment control system can work normally or not determines the working capacity and working efficiency of the personnel on the airplane.
The aircraft environmental control system mostly adopts a closed-loop control mode to perform feedback adjustment on environmental parameters in the aircraft cabin, so that the environmental parameters are maintained at a preset level and fluctuate within an allowable range. When the environmental control system fails, the environmental parameters in the cabin are abnormal. At this moment, the fault needs to be timely and accurately warned, so that appropriate emergency treatment measures are taken, and the safety of personnel in the cabin is ensured.
The cockpit abnormity detection method of the existing airplane environment control system is mainly divided into a fixed threshold class and a dynamic threshold class. The abnormal detection method based on the fixed threshold value monitors abnormal conditions of the cockpit by artificially setting upper and lower limit threshold values of each environmental parameter in the cockpit, and sends an abnormal alarm when an actually measured parameter exceeds a threshold value range. The fixed threshold values are derived from expert experience, or from analysis and recording of historical flight parameter data. The detection method based on the fixed threshold has the defects of poor self-adaptability, serious disturbance influence and the like, and is easy to cause the phenomena of false alarm, missing report and the like. The anomaly detection method based on the dynamic threshold realizes the tracking of the environmental parameters of the cabin by establishing an observer model, simultaneously considers the actual instruction and the actual feedback condition of the cabin, dynamically generates the threshold of the residual error between the tracking parameters and the actual measurement parameters, and realizes the detection of the anomaly of the cabin by comparing the actual residual error with the residual error threshold. The traditional method based on dynamic threshold only considers the incidence relation between the instruction and feedback at the same time and between the residual error and the instruction-feedback combination, and does not consider the existing time sequence dependence characteristics, so that certain defects exist.
Disclosure of Invention
In view of the above problems, the present application aims to provide an aircraft cabin abnormal fault detection method based on dual-stage gated loop network (GRU) associative memory, which learns the association between cabin parameter control instructions and feedback of an environmental control system, and the association between residual errors between actual feedback and observation feedback and instruction-feedback combinations by establishing a deep neural network based on a gated loop unit, and dynamically generates an adaptive residual error threshold, thereby implementing abnormal detection of a cabin.
The application discloses an aircraft cabin abnormal fault detection method based on two-stage gated loop network associative memory, which comprises the following steps:
the method comprises the following steps: time series data acquisition and delayed slicing processing
And acquiring a cabin environmental parameter control command signal and a feedback signal sequence in a normal state of the system, and delaying the feedback signal sequence by one step on a time axis to enable the feedback signal to lag behind the control command signal by one sampling moment. And slicing the processed command signal and the delayed feedback signal by using a single-step time sliding window with a preset length, so that each sample comprises the command signal and the feedback signal with the same sampling time. And taking each slice sample as a training input, taking a feedback signal value which is the same as the sampling time of the instruction signal in the last step in each slice sample as a training output, and finishing time sequence data acquisition and delayed slice processing to obtain a training sample of the primary GRU network association observer. For the case where multiple redundant sensors are present, the process is the same.
Step two: establishing and training one-level GRU network association observer
Constructing and initializing a gated cycle unit network with a multi-input single-output structure to serve as a primary GRU network association observer; and (4) inputting the training sample set constructed in the step one, and training the network by using an optimization algorithm until the loss value is converged. At the moment, the first-stage association observer can establish an association mapping between the multistep command signal, the multistep delay feedback signal and the current single-step feedback signal. For the case where multiple redundant sensors are present, the process is the same.
Step three: observation residual acquisition and slicing
And inputting all input sample matrixes in the training set into the trained primary GRU network association observer according to the time sequence to obtain an observation feedback sequence output by the observer. And comparing the values of the observation feedback sequence and the actual feedback sequence at the corresponding sampling time to obtain a residual sequence between the observation feedback sequence and the actual feedback sequence. And slicing the actual command signal and the observation feedback signal by using a single-step time sliding window, so that each sample comprises the command signal and the observation feedback signal with the same sampling time number. And taking each slice sample as a training input, and superposing a threshold correction factor on a residual error numerical value corresponding to the last sampling moment of each sample as a training output to obtain a training sample of the secondary GRU network association observer. For the case where multiple redundant sensors are present, the process is the same.
Step four: establishing and training two-stage GRU network association observer
Constructing and initializing a gated cycle unit network with a multi-input single-output structure to serve as a secondary GRU network association observer; inputting the training sample set constructed in the third step, and training the network by using an optimization algorithm until the loss value is converged. At the moment, the secondary association observer can establish association mapping between the multistep command signal, the multistep observation feedback signal and the current residual signal threshold value. For the case where multiple redundant sensors are present, the process is the same.
Step five: aircraft cabin anomaly fault detection
And collecting an actual airplane cabin environment parameter instruction signal and feedback signals of left and right measurement. After carrying out the same delay slice processing as the first step, inputting the first-stage GRU network association observer model obtained by training in the second step to obtain an output association feedback signal; carrying out observation residual acquisition and slicing processing which are the same as the third step, and obtaining a secondary observer input matrix which consists of an actual residual sequence and a multi-step instruction signal-multi-step observation feedback signal; and inputting the input matrix of the secondary observer into the secondary GRU network association observer obtained by training in the fourth step to obtain an output self-adaptive residual error threshold sequence. And judging the occurrence of the abnormal faults of the aircraft cabin by comparing the relation between the actual residual sequence and the self-adaptive residual threshold sequence. And when the residual sequence measured by the left sensor and the right sensor continuously exceeds a certain number of sample points and exceeds a threshold value, an abnormal alarm is sent out.
The invention has the advantages and positive effects that:
(1) the dynamic residual threshold output by the observer is used for replacing a fixed threshold generated based on expert knowledge or data drive in the traditional aircraft cockpit fault detection algorithm, the dynamic characteristic in closed-loop feedback control is fully considered, the detection rate is effectively improved, and the false alarm rate is reduced;
(2) the method has the advantages that a gated loop network (GRU) with a loop structure is used for learning the relation between feedback and an instruction and the relation between residual error and the instruction and the feedback in closed-loop temperature control, the front and rear time sequence dependency relation included in a sequence is fully excavated, and the defect that a traditional observer can only establish single-point mapping in the aspect of processing a time sequence is overcome;
(3) the left and right temperature sensor signals of the aircraft cabin are used to jointly obtain a fault detection result, so that the accuracy and the credibility of fault detection are improved.
Drawings
FIG. 1 is a flow chart of an aircraft cabin abnormal fault detection method based on two-stage gated loop network associative memory according to the invention;
FIG. 2 is a schematic diagram of a GRU structure of the present invention;
FIG. 3 is a one-stage GRU network association observer (left) training penalty in an embodiment of the present invention;
FIG. 4 is a one-level GRU network associated observer (right) training penalty in an embodiment of the present invention;
FIG. 5 shows the residual error between the actual feedback (left) and the observed feedback under normal conditions in an embodiment of the present invention;
FIG. 6 shows the residual error between the actual feedback (right) and the observed feedback under normal conditions in an embodiment of the present invention;
FIG. 7 is a two-level GRU network association observer (left) training penalty in an embodiment of the present invention;
FIG. 8 is a two-level GRU network associated observer (right) training penalty in an embodiment of the present invention;
FIG. 9 is a cabin temperature (left) residual and adaptive threshold in an embodiment of the present invention;
FIG. 10 is a cabin temperature (right) residual and adaptive threshold in an embodiment of the present invention.
Detailed Description
The application discloses an aircraft cabin abnormal fault detection method based on two-stage gated loop network associative memory, which mainly comprises the following steps of:
the method comprises the following steps: time series data acquisition and delayed slicing processing
Acquiring a cockpit environmental parameter control instruction signal sequence X in a normal state of the systemh＝[x1，x2，...，xn]And a feedback signal sequence Yh＝[y1，y2，...，yn]The feedback signal sequence is delayed by one step on the time axis so that the feedback signal lags behind the control command signal by one sampling time and becomes Yh＝[y0，y1，...，yn-1]Wherein y is00. And slicing the processed command signal and the delayed feedback signal by using a single-step time sliding window with a preset length w, so that each sample comprises the command signal and the feedback signal with the same sampling time value. The ith sample matrix after slicing is [ x ]i，xi+1，...，xi+w-1；yi-1，yi，...，yi+w-2]. And taking each slice sample as a training input, taking a feedback signal value which is the same as the sampling time of the instruction signal in the last step in each slice sample as a training output, and finishing time sequence data acquisition and delayed slice processing to obtain a training sample of the primary GRU network association observer. For the ith sample matrix, the label output is yi+w-1. For the case where multiple redundant sensors are present, the process is the same.
Step two: establishing and training one-level GRU network association observer
A gated cyclic unit network of a multi-input single-output structure is constructed and initialized, as shown in fig. 2, to serve as a one-level GRU network association observer.
GRU is designed to solve the problems of long-term memory and back propagationThe cyclic neural network structure has the performance advantage of processing sequence data with time sequence dependency relationship before and after. At time step t, first use formula zt＝σ(W(z)xt+U(z)ht-1) Compute update gate zt(ii) a Wherein xtThe input vector for the t-th time step, i.e. the t-th component of the input sequence, is subjected to a linear transformation (with the weight matrix W)(z)Multiplication). h ist-1The information of the previous time step t-1 is stored, which is also subjected to a linear transformation. The updating gate adds the two parts of information and adds the two parts of information into the Sigmoid activation function sigma (t) which is 1/(1+ e)-t) Thus compressing the activation result to between 0 and 1.
By the formula rt＝σ(W(r)xt+U(r)ht-1) A reset gate is calculated. As described above for the refresh door, ht-1And xtFirstly, a linear transformation is carried out, and then a Sigmoid activation function is added to output an activation value. The new memory will use the reset gate to store the past related information, which is calculated by the expression: h't＝tanh(Wxt+rt⊙Uht-1). Input xtInformation h of last time stept-1First, a linear transformation is performed, i.e., right multiplication of the matrices W and U, respectively.
Finally, h is calculatedtThe vector will retain the information of the current unit and pass on to the next unit. This process can be expressed as:
ht＝zt⊙ht-1+(1-zt)⊙h′t。
and (4) inputting the training sample set constructed in the step one, and training the network by using an optimization algorithm until the loss value is converged. At the moment, the first-stage association observer can establish an association mapping between the multistep command signal, the multistep delay feedback signal and the current single-step feedback signal. For the case where multiple redundant sensors are present, the process is the same.
Step three: observation residual acquisition and slicing
Will train all inputs in the setThe sample matrix is input into the trained one-stage GRU network association observer according to the time sequence to obtain an observation feedback sequence output by the observer
Figure BDA0002424911950000041
Comparing the values of the observation feedback sequence and the actual feedback sequence at the corresponding sampling time to obtain a residual sequence between the observation feedback sequence and the actual feedback sequence
Figure BDA0002424911950000042
And slicing the actual command signal and the observation feedback signal by using a single-step time sliding window, so that each sample comprises the command signal and the observation feedback signal with the same sampling time number. The ith sample matrix after slicing is
Figure BDA0002424911950000043
And taking each slice sample as a training input, and superposing a threshold correction factor on a residual error numerical value corresponding to the last sampling moment of each sample as a training output to obtain a training sample of the secondary GRU network association observer. For the ith sample matrix, the label output is ri+2w-2+ β, where β is a set threshold correction factor the process is the same for the case where multiple redundant sensors are present.
Step four: establishing and training two-stage GRU network association observer
Constructing and initializing a gated cycle unit network with a multi-input single-output structure to serve as a secondary GRU network association observer; inputting the training sample set constructed in the third step, and training the network by using an optimization algorithm until the loss value is converged. At the moment, the secondary association observer can establish association mapping between the multistep command signal, the multistep observation feedback signal and the current residual signal threshold value. For the case where multiple redundant sensors are present, the process is the same.
Step five: aircraft cabin anomaly fault detection
Acquiring actual airplane cabin environment parameter command signal X ═ X1，x2，...，xm]Feedback with left and right measurementsSignal Yl＝[yl1，yl2，...，ylm]And Yr＝[yr1，yr2，...，yrm]Where m is the number of sample points. After the delay slice processing which is the same as that in the first step is carried out on the model, the model is input into a first-stage GRU network association observer model obtained by training in the second step to obtain an output association feedback signal
Figure BDA0002424911950000051
And
Figure BDA0002424911950000052
carrying out observation residual acquisition and slicing processing which are the same as the third step to obtain an actual residual sequence
Figure BDA0002424911950000053
And Rr＝[rrw，rrw+1，...，rrm]A secondary observer input matrix which is formed by the multi-step command signal and the multi-step observation feedback signal; inputting the input matrix of the secondary observer into the secondary GRU network associated observer obtained by the training in the fourth step to obtain an output self-adaptive residual error threshold sequence Tl＝[t2*lw，t2*lw+1，…，tlm]And Tr＝[t2*rw，t2*rw+1，...，trm]. And judging the occurrence of the abnormal faults of the aircraft cabin by comparing the relation between the actual residual sequence and the self-adaptive residual threshold sequence. And when the residual sequence measured by the left sensor and the right sensor continuously exceeds p points and exceeds the threshold value, an abnormal alarm is sent out.
Examples
In this embodiment, the cabin temperature is taken as an example, and the method of the present invention is used to detect the abnormal cabin temperature, so as to explain the contents of the present invention and further explain the using process of the contents of the present invention.
According to the data recording condition, the abnormal cabin temperature fault occurs between the 13000 th sampling point and the 43000 th sampling point in all the sampling points. Therefore, the data is divided into fault data for model test; the rest part is divided into normal data for model training.
The method comprises the following steps: time series data acquisition and delayed slicing processing
After a cabin environment parameter control instruction signal and a feedback signal in a normal state of the system are obtained, the length of a sliding window is set to be 5, delayed slicing processing is carried out on data, and at the moment, each input sample matrix comprises parameter values of 5 sampling moments. The ith sample matrix after slicing is [ x ]i，xi+1，xi+2，xi+3，xi+4；yi-1，yi，yi+1，yi+2，yi+3]The output of the tag corresponding thereto is yi+4. And all the sample matrixes and the corresponding output labels form a training sample of the one-stage GRU network association observer.
Step two: establishing and training one-level GRU network association observer
The GRU networks with model structures as shown in table 1 were constructed for the left and right temperature sensors, respectively.
Table 1 one-stage GRU network association observer structure
Figure BDA0002424911950000054
Figure BDA0002424911950000061
The hyper-parameter settings for network training are as follows:
optimization algorithm: RMSprop
Loss function: mean square error
Update batch size: 128
Training iteration number: 50
And (3) respectively training the primary GRU network association observer (left) and the primary GRU network association observer (right) by using the training samples obtained in the step one, wherein the training loss change conditions are shown in fig. 3 and 4.
Step three: observation residual acquisition and slicing
And inputting all input sample matrixes in the training set into the trained primary GRU network association observer according to the time sequence to obtain an observation feedback sequence output by the observer. And comparing the values of the observation feedback sequence and the actual feedback sequence at the corresponding sampling time to obtain a residual sequence between the observation feedback sequence and the actual feedback sequence, as shown in fig. 5 and 6.
And slicing the actual command signal and the observation feedback signal by using a single-step time sliding window, wherein the length of the sliding window is also 5, so that each sample comprises the command signal and the observation feedback signal values at 5 sampling moments. The threshold correction factor is set to 0.02, which is empirically obtained. And taking the slice sample matrix as the training input of the secondary GRU network association observer, and taking the corrected residual error as the training output of the secondary GRU network association observer to obtain a training sample of the secondary GRU network association observer.
Step four: establishing and training two-stage GRU network association observer
And (4) constructing and initializing a gated cycle unit network with a multi-input single-output structure to serve as a secondary GRU network association observer, wherein the structural parameters and the training parameters of the gated cycle unit network are completely the same as those of the network in the step two. Training the two-stage GRU network association observer (left) and the two-stage GRU network association observer (right) respectively by using the training samples obtained in the third step, wherein the training loss variation is as shown in fig. 7 and 8.
Step five: aircraft cabin anomaly fault detection
Acquiring actual airplane cabin environment parameter command signal X ═ X1，x2，...，xm]Feedback signal Y measured with the left and the rightl＝[yl1，yl2，...，ylm]And Yr＝[yr1，yr2，...，yrm]Where m is the number of sample points. After the delay slice processing which is the same as that in the first step is carried out on the model, the model is input into a first-stage GRU network association observer model obtained by training in the second step to obtain an output association feedback signal
Figure BDA0002424911950000062
And
Figure BDA0002424911950000063
carrying out observation residual acquisition and slicing processing which are the same as the third step to obtain an actual residual sequence
Figure BDA0002424911950000064
And Rr＝[rrw，rrw+1，…，rrm]A secondary observer input matrix which is formed by the multi-step command signal and the multi-step observation feedback signal; inputting the input matrix of the secondary observer into the secondary GRU network associated observer obtained by the training in the fourth step to obtain an output self-adaptive residual error threshold sequence Tl＝[t2·lw，t2·lw+1，...，tlm]And Tr＝[t2*rw，t2*rw+1，...，trm]As shown in fig. 9 and 10.
And judging the occurrence of the abnormal faults of the aircraft cabin by comparing the relation between the actual residual sequence and the self-adaptive residual threshold sequence. And when the residual sequence measured by the left sensor and the right sensor continuously exceeds 10 points and the threshold value exceeds the limit, an abnormal alarm is sent out. As can be seen from the result graph, the residual error of the abnormal cabin temperature section obviously exceeds the adaptive threshold, which shows that the content of the invention can timely and accurately give an alarm when the abnormal cabin temperature occurs.
Unless defined otherwise, all technical and/or scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The materials, methods, and examples set forth in this application are illustrative only and not intended to be limiting.
Although the present invention has been described in conjunction with specific embodiments thereof, it is evident that many alternatives, modifications, and variations will be apparent to those skilled in the art in light of the teachings of this application and yet remain within the scope of this application.

Claims (6)

1. An aircraft cabin abnormal fault detection method based on double-stage gated loop network associative memory is characterized by comprising the following steps:
the first step is as follows: acquiring time sequence data and delaying slicing processing;
the second step is that: establishing and training a primary GRU network association observer;
the third step: obtaining observation residual and processing slices;
the fourth step: establishing and training a secondary GRU network association observer;
the fifth step: and detecting abnormal faults of the airplane cockpit.
2. The aircraft cabin abnormal fault detection method of claim 1, wherein:
acquiring a cabin environmental parameter control command signal and a feedback signal sequence in a normal state of a system, and delaying the feedback signal sequence by one step on a time axis to enable the feedback signal to lag behind the control command signal by a sampling moment; slicing the processed command signal and the delayed feedback signal by using a single-step time sliding window with a preset length, so that each sample comprises the command signal and the feedback signal with the same sampling time number; taking each slice sample as a training input, taking a feedback signal value in each slice sample, which is the same as the sampling time of the last step of instruction signal, as a training output, and completing time sequence data acquisition and delayed slice processing to obtain a training sample of the primary GRU network association observer; for the case where multiple redundant sensors are present, the process is the same.
3. The aircraft cabin abnormal fault detection method of claim 2, wherein:
constructing and initializing a gated cycle unit network with a multi-input single-output structure to serve as a primary GRU network association observer; inputting the training sample set constructed in the step one, and training the network by using an optimization algorithm until the loss value is converged; at the moment, the first-stage association observer can establish association mapping between the multistep instruction signal, the multistep delay feedback signal and the current single-step feedback signal; for the case where multiple redundant sensors are present, the process is the same.
4. The aircraft cabin abnormal fault detection method of claim 3, wherein:
inputting all input sample matrixes in the training set into the trained primary GRU network association observer according to a time sequence to obtain an observation feedback sequence output by the observer; and comparing the values of the observation feedback sequence and the actual feedback sequence at the corresponding sampling time to obtain a residual sequence between the observation feedback sequence and the actual feedback sequence. Slicing the actual instruction signal and the observation feedback signal by using a single-step time sliding window, so that each sample comprises the instruction signal and the observation feedback signal with the same sampling time number; taking each slice sample as a training input, and superposing a threshold correction factor on a residual error numerical value corresponding to the last sampling moment of each sample as a training output to obtain a training sample of the secondary GRU network association observer; for the case where multiple redundant sensors are present, the process is the same.
5. The aircraft cabin abnormal fault detection method of claim 4, wherein:
constructing and initializing a gated cycle unit network with a multi-input single-output structure to serve as a secondary GRU network association observer; inputting the training sample set constructed in the third step, and training the network by using an optimization algorithm until the loss value is converged; at the moment, the secondary association observer can establish association mapping between the multi-step instruction signal, the multi-step observation feedback signal and the current residual signal threshold; for the case where multiple redundant sensors are present, the process is the same.
6. The aircraft cabin abnormal fault detection method of claim 5, wherein:
collecting an actual airplane cabin environment parameter instruction signal and feedback signals measured in a left path and a right path; after carrying out the same delay slice processing as the first step, inputting the first-stage GRU network association observer model obtained by training in the second step to obtain an output association feedback signal; carrying out observation residual acquisition and slicing processing which are the same as the third step, and obtaining a secondary observer input matrix which consists of an actual residual sequence and a multi-step instruction signal-multi-step observation feedback signal; inputting the input matrix of the secondary observer into the secondary GRU network associated observer obtained by training in the fourth step to obtain an output self-adaptive residual error threshold sequence; judging the occurrence of abnormal faults of the aircraft cockpit by comparing the relation between the actual residual sequence and the self-adaptive residual threshold sequence; and when the residual sequence measured by the left sensor and the right sensor continuously exceeds a certain number of sample points and exceeds a threshold value, an abnormal alarm is sent out.
CN202010217602.8A 2020-03-25 2020-03-25 Aircraft cabin abnormal fault detection method based on two-stage gated loop network associative memory Pending CN111381584A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010217602.8A CN111381584A (en) 2020-03-25 2020-03-25 Aircraft cabin abnormal fault detection method based on two-stage gated loop network associative memory

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010217602.8A CN111381584A (en) 2020-03-25 2020-03-25 Aircraft cabin abnormal fault detection method based on two-stage gated loop network associative memory

Publications (1)

Publication Number Publication Date
CN111381584A true CN111381584A (en) 2020-07-07

Family

ID=71220049

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010217602.8A Pending CN111381584A (en) 2020-03-25 2020-03-25 Aircraft cabin abnormal fault detection method based on two-stage gated loop network associative memory

Country Status (1)

Country Link
CN (1) CN111381584A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113219942A (en) * 2021-04-23 2021-08-06 浙江大学 Blast furnace fault diagnosis method based on weighted joint distribution adaptive neural network
CN113341927A (en) * 2021-06-11 2021-09-03 江西洪都航空工业集团有限责任公司 Flight control system servo actuator BIT fault detection method and device

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113219942A (en) * 2021-04-23 2021-08-06 浙江大学 Blast furnace fault diagnosis method based on weighted joint distribution adaptive neural network
CN113219942B (en) * 2021-04-23 2022-10-25 浙江大学 Blast furnace fault diagnosis method based on weighted joint distribution adaptive neural network
CN113341927A (en) * 2021-06-11 2021-09-03 江西洪都航空工业集团有限责任公司 Flight control system servo actuator BIT fault detection method and device
CN113341927B (en) * 2021-06-11 2022-12-02 江西洪都航空工业集团有限责任公司 Flight control system servo actuator BIT fault detection method and device

Similar Documents

Publication Publication Date Title
CN108594788B (en) Airplane actuator fault detection and diagnosis method based on deep random forest algorithm
CN109459993B (en) Online adaptive fault monitoring and diagnosing method for process industrial process
CN109376413B (en) Online neural network fault diagnosis method based on data driving
EP2228545B1 (en) Methods and apparatus to arbitrate valve position sensor redundancy
CN111381584A (en) Aircraft cabin abnormal fault detection method based on two-stage gated loop network associative memory
CN110909822B (en) Satellite anomaly detection method based on improved Gaussian process regression model
CN114564000A (en) Active fault tolerance method and system based on fault diagnosis of intelligent aircraft actuator
CN113033011B (en) Ship mechanical health state assessment method and system
CN111823273B (en) Operation monitoring system and monitoring method for industrial robot online track prediction
CN112947385B (en) Aircraft fault diagnosis method and system based on improved Transformer model
CN111459142A (en) Aircraft liquid cooling failure fault diagnosis method based on stacked sparse noise reduction self-encoder
Balzano et al. Air data sensor fault detection with an augmented floating limiter
CN116661307A (en) Nonlinear system actuator fault PPB-SIADP fault-tolerant control method
Chen et al. A fault detection method based on enhanced GRU
Zhang et al. Dynamic learning-based fault tolerant control for robotic manipulators with actuator faults
CN112631258B (en) Fault early warning method for key indexes of industrial process
CN113496261A (en) On-orbit anomaly monitoring method for satellite telemetry data
CN112734131A (en) Fan blade icing state prediction method based on deep learning algorithm
CN111887004B (en) Control method of rod-shaped crop harvesting robot
CN114397809A (en) Intelligent control system for material weighing big data detection and packaging
CN114326686A (en) Unmanned aerial vehicle multiplicative fault diagnosis and fault-tolerant control method based on residual entropy minimization
CN113268919A (en) Design method of linear active disturbance rejection controller based on neural network prediction
CN112116003B (en) Self-adaptive dynamic fault diagnosis method and system for complex sensor network
Gu et al. A Baseline Modeling Method for UAV Condition Monitoring based on Multiple Flight Data
CN117572779B (en) Control method of electric aeroengine under blade damage

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