CN104850889A - Airplane rotation actuator drive unit adaptive fault detection, isolation and confidences assessment method - Google Patents
Airplane rotation actuator drive unit adaptive fault detection, isolation and confidences assessment method Download PDFInfo
- Publication number
- CN104850889A CN104850889A CN201410664864.3A CN201410664864A CN104850889A CN 104850889 A CN104850889 A CN 104850889A CN 201410664864 A CN201410664864 A CN 201410664864A CN 104850889 A CN104850889 A CN 104850889A
- Authority
- CN
- China
- Prior art keywords
- drive unit
- actuator drive
- rotating actuator
- fault
- isolation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 230000003044 adaptive effect Effects 0.000 title claims abstract description 44
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000002955 isolation Methods 0.000 title claims abstract description 39
- 238000001514 detection method Methods 0.000 title claims abstract description 9
- 238000012549 training Methods 0.000 claims abstract description 19
- 238000010606 normalization Methods 0.000 claims abstract description 7
- 238000006073 displacement reaction Methods 0.000 claims description 54
- 238000013528 artificial neural network Methods 0.000 claims description 47
- 230000001537 neural effect Effects 0.000 claims description 40
- 230000036541 health Effects 0.000 claims description 33
- 230000003862 health status Effects 0.000 claims description 7
- 238000004458 analytical method Methods 0.000 claims description 6
- 238000000926 separation method Methods 0.000 claims description 4
- 238000004088 simulation Methods 0.000 claims description 4
- 230000004927 fusion Effects 0.000 claims description 2
- 230000004807 localization Effects 0.000 claims description 2
- 238000013139 quantization Methods 0.000 claims description 2
- 238000013507 mapping Methods 0.000 abstract description 5
- 210000005036 nerve Anatomy 0.000 abstract 5
- 238000012360 testing method Methods 0.000 description 22
- 239000013598 vector Substances 0.000 description 16
- 241001503991 Consolida Species 0.000 description 12
- 238000011156 evaluation Methods 0.000 description 10
- 230000005540 biological transmission Effects 0.000 description 7
- 230000007257 malfunction Effects 0.000 description 5
- 230000008901 benefit Effects 0.000 description 4
- 230000008859 change Effects 0.000 description 4
- 230000006870 function Effects 0.000 description 3
- 238000013178 mathematical model Methods 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000004804 winding Methods 0.000 description 3
- 230000015556 catabolic process Effects 0.000 description 2
- 238000006731 degradation reaction Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000006698 induction Effects 0.000 description 2
- 238000002347 injection Methods 0.000 description 2
- 239000007924 injection Substances 0.000 description 2
- 239000007788 liquid Substances 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 230000000638 stimulation Effects 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000000739 chaotic effect Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 230000002860 competitive effect Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000004615 ingredient Substances 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 239000000243 solution Substances 0.000 description 1
Abstract
The invention discloses an airplane rotation actuator drive unit adaptive fault detection, isolation and confidences assessment method. Firstly, in dependence on the structure of a rotation actuator drive unit, establishing a two-stage observer based on a radial basis function (RBF) nerve network, wherein a first stage RBF nerve network observer is used for estimating system desired output values and generating residual errors, and a second stage RBF nerve network observer is used for generating adaptive threshold values; then performing fault isolation based on output information of the observer; training a self-organizing mapping (SOM) nerve network by utilizing residual errors under a normal state; and finally utilizing residual errors to be detected and assessed as input of the SOM nerve network and performing normalization processing to obtain confidences values (CVs) indexes between 0 and 1. The airplane rotation actuator drive unit adaptive fault detection, isolation and confidences assessment method has high practical engineering application values.
Description
Technical field
The present invention relates to rotary actuator fault detect, isolation and the technical field of health evaluating, be specifically related to a kind of airplane rotating actuator drive unit adaptive failure and detect, isolate and the method for health evaluating.
Background technology
Rotary actuator has the advantage of large rotating speed/mass ratio, simple compact structure, fast dynamic response, has been widely used in aircraft, boats and ships and tank.Drive unit is the important component part of rotary actuator, and airplane rotating actuator drive unit fault, by the safe and stable operation of the whole aircraft of impact, causes huge economic loss or even unpredictable consequence.Therefore, ensure that the normal work of airplane rotating actuator drive unit is significant in actual applications.
At present, almost not about the correlative study of the fault detect of airplane rotating actuator drive unit, isolation and health evaluating, only researchs often have ignored again the impact of various interference, are difficult to be applied in Practical Project.
Usually, the method for fault detect, isolation and health evaluating can be divided into the method for physically based deformation model and the method based on data-driven.The method of physically based deformation model requires to have the Structure Mechanism of research object to be very clearly familiar with, and needs to determine mathematical model accurately.Because the structure of airplane rotating actuator is very complicated, and often there is nonlinear characteristic, very difficult acquisition physical model accurately.Based on the method for data-driven without the need to accurate system model, only utilize Monitoring Data just can carry out fault detect and assessment, the application in recent years in complication system is very extensive.Zhang Changfan etc. have carried out fault modeling on the basis analyzing induction machine stator winding and machines under rotor winding faults characteristic, and utilize the space mathematical model of induction motor d-q coordinate system to propose a kind of method for designing of Robust Observers, the method has good robustness to Unknown worm disturbance, has higher sensitivity to winding failure.Lv Chen etc. propose a kind of method of the assessment bearing health status based on chaotic characteristic.Within 2006, Jayakumar and Das proposes a kind of aircraft control system fault detect based on observer, isolation and reconstructing method.Li Xiaoxiong etc., by setting up a set of sane Adaptive Observer, propose a kind of method of aircraft control system fault isolation and reconstruct.Visible, observer obtains good application in complication system/equipment.
Neural net method, as a kind of method based on data-driven, can realize the fault detect of non-linear and robustness, isolation and health evaluating.In neural net method, RBF neural has better approximation capability, faster learning ability, better robustness, local minimum can be avoided, for the change of accurate tracking control system model, and can adaptively modifying self neural network parameter, thus realize the fault detect of airplane rotating actuator drive unit, isolation and health evaluating.Simultaneously, because airplane rotating actuator drive unit working environment is changeable, fault threshold will be subject to the impact of system input, random disturbance, operating mode disturbance, parameter drift, modeling error, random noise etc., and traditional fault detect based on fixed threshold, health evaluating method cannot meet the demand of practical engineering application.Therefore, the learning ability utilizing RBF neural good and robustness, produce the adaptive threshold of fault.SOM neural network has self-organization, the adaptive ability of height, can carry out automatic cluster to sample, therefore applies SOM and carries out failure modes and health evaluating.Although be successfully applied to engineering reality based on the fault detect of neural network, isolation and health evaluating method, the application on airplane rotating actuator drive unit has not also been studied.
Summary of the invention
Based on above analysis, in order to solve the fault detect of airplane rotating actuator drive unit, isolation and health evaluating, the present invention proposes a set of adaptive failure based on bipolar RBF neural observer and SOM neural network and detecting, isolate and health evaluating method.
The technical solution used in the present invention is: a kind of airplane rotating actuator drive unit adaptive failure detects, the method for isolation and health evaluating, and concrete steps are as follows:
Step 1, structure by analysis of aircraft rotary actuator drive unit, set up the realistic model of airplane rotating actuator drive unit;
Step 2, fault detection method for airplane rotating actuator drive unit, utilize input instruction signal and angle displacement signal, set up two-stage RBF neural network Residual Generation device and adaptive threshold maker, realize the fault detect of airplane rotating actuator drive unit;
Step 3, failure separation method for airplane rotating actuator drive unit, utilize input instruction signal and angle displacement signal, and application two cover two-stage RBF neural networks set up Failure Observer, realize the localization of fault of airplane rotating actuator drive unit;
Step 4, health evaluating method for airplane rotating actuator drive unit, one-level RBF Residual Generation device is utilized to obtain fault residual value, utilize the residual values training SOM neural network under drive unit normal operating conditions, then utilize the health evaluating of the SOM neural fusion airplane rotating actuator drive unit trained.
Further, described airplane rotating actuator drive unit is closed loop feedback system, is substantially made up of control module, servo-valve, oil motor, reducing gear and topworks; In control loop, there are two angular displacement sensors to be used for feedback system information, help execution architecture to obtain optimal angle; In actual applications, the parameter of servo-valve and oil motor is difficult to obtain, and thus in simulations, the data of acquisition are the feedback signal of control signal and reducing gear and the output signal of displacement of drive unit previous moment.
Further, the described fault detection method for airplane rotating actuator drive unit, comprise two-stage RBF neural network, wherein, one-level RBF neural is as Residual Generation device, network is input as the displacement signal of steering order and the output of previous moment system, and export the estimated value into drive unit displacement signal, the difference of this estimated value and the actual output displacement signal of system is residual values; And secondary RBF neural is as adaptive threshold maker, be input as the estimated value of Systematical control instruction and drive unit displacement signal, export as adaptive fault threshold; By the relative size of the residual values and adaptive threshold that compare current time, judge the health status of current time equipment, realize the fault detect of drive unit.
Further, the described failure separation method for airplane rotating actuator drive unit, comprise two Failure Observers, each observer is made up of one group of two-stage RBF neural network, wherein one-level RBF neural is as Residual Generation device, network is input as the displacement signal of steering order and the output of previous moment drive unit, and export the estimated value into drive unit displacement signal, the difference of this estimated value and the actual output displacement signal of drive unit is residual values; And secondary RBF neural is as adaptive threshold maker, be input as the estimated value of Systematical control instruction and drive unit displacement signal, export as adaptive fault threshold; Two observers lay respectively in inside and outside two loops of drive unit closed-loop control system, according to the output information of two observers, judge inner-outer loop whether fault, and then can the concrete generation position of failure judgement, thus realize the fault isolation of drive unit.
Further, the described health evaluating method for airplane rotating actuator drive unit, comprise an one-level RBF neural and a SOM neural network, wherein one-level RBF neural is as Residual Generation device, network is input as the displacement signal of steering order and the output of previous moment drive unit, export the estimated value for drive unit displacement signal, the difference of this estimated value and the actual output displacement signal of drive unit is residual values; The residual values training SOM neural network utilizing drive unit normal operating conditions next stage RBF neural to produce; Afterwards, using from the residual values obtained during drive unit actual motion as the input of the SOM neural network trained, calculate minimum quantization error MQE value, and normalization obtains the CV value between 0-1, realizes the health evaluating of drive unit.
The present invention's advantage is compared with prior art:
(1) deficient for the fault detect of airplane rotating actuator drive unit, isolation and health evaluating correlative study, the present situation that achievement in research practical application effect has much room for improvement, proposes the airplane rotating actuator drive unit fault detect of complete set, isolation and the effective ways of health evaluating;
(2) utilize two-stage RBF neural network, obtain residual error and obtain adaptive fault threshold, compensate for the problem that fixed threshold brings, improve the accuracy rate of fault detect, reduce detection false alarm rate.
(3) by the structure of analysis of aircraft rotary actuator drive unit, utilize two groups of two-stage RBF neural networks to form two observers and fault detect is carried out to inner-outer loop, achieve the fault location and isolation of airplane rotating actuator drive unit.
(4) apply one-level RBF neural observer and the combination of SOM neural network, utilize residual values to carry out Performance Evaluation, achieve the health evaluating of airplane rotating actuator drive unit.Meanwhile, application SOM neural network carries out health evaluating, data when only needing drive unit normally to run and without the need to fault data, effectively improve the engineering practicability of Performance Evaluation.
Accompanying drawing explanation
Fig. 1 is the overall flow figure of the inventive method;
Fig. 2 is airplane rotating actuator drive unit basic structure;
Fig. 3 is typical single hidden layer Architecture of Feed-forward Neural Network;
Fig. 4 is the foundation of Failure Observer;
Fig. 5 is the policy map of fault isolation;
Fig. 6 is the structural drawing of SOM neural network;
Fig. 7 is residual values and the fault threshold of lower No. 1 observer of normal condition;
Fig. 8 is residual values and the fault threshold of lower No. 2 observers of normal condition;
Fig. 9 is the testing result that servo-valve magnetic field intensity reduces lower No. 1 observer of state;
Figure 10 is the testing result that servo-valve magnetic field intensity reduces lower No. 2 observers of state;
Figure 11 is the testing result of lower No. 1 observer of oil motor inner ring leak condition;
Figure 12 is the testing result of lower No. 2 observers of oil motor inner ring leak condition;
Figure 13 is the testing result of lower No. 1 observer of Transmission Shaft Stiffness degenerate state;
Figure 14 is the testing result of lower No. 2 observers of Transmission Shaft Stiffness degenerate state;
Figure 15 is that 2 bugle displacement transducer output accuracies exceed standard the testing result of lower No. 1 observer of state;
Figure 16 is that 2 bugle displacement transducer output accuracies exceed standard the testing result of lower No. 2 observers of state;
Figure 17 is the performance evaluation result under normal condition;
Figure 18 is the performance evaluation result under servo-valve magnetic field intensity reduction state;
Figure 19 is the performance evaluation result under Transmission Shaft Stiffness degenerate state;
Embodiment
The present invention is further illustrated below in conjunction with accompanying drawing and specific embodiment.
A kind of airplane rotating actuator drive unit of the present invention adaptive failure detects, isolation and the method for health evaluating, and as shown in Figure 1, concrete steps are as follows for method flow:
Step one, set up the structure of airplane rotating actuator drive unit
Airplane rotating actuator drive unit is made up of control module, servo-valve, oil motor, reducing gear and topworks substantially, as shown in Figure 2.In control loop, there are two angular displacement sensors to be used for feedback system information, help execution architecture to obtain optimal angle.When topworks reaches the angle of expectation, this angle signal is fed back to control module by system, and then control module control system execution braking mode (as Suo Shi the line between " control module is to arrestment mechanism " and " arrestment mechanism is to oil motor " in Fig. 2) makes oil motor brake to keep topworks to be stabilized in this expection angle state.In actual applications, the parameter of servo-valve and oil motor is difficult to obtain, and thus in simulations, the data that we obtain are feedback signals of control signal and reducing gear.
The adaptive failure of step 2, airplane rotating actuator drive unit detects:
1, RBF neural
Neural network has the characteristic of Approximation of Arbitrary Nonlinear Function, may be used for the state recognition of nonlinear system, and sets up state recognition model without the need to the actual mathematical model based on system.
RBF neural is a kind of feedforward network, is made up of, as shown in Figure 3 input layer, hidden layer and output layer, X=[x1, x2 ... xn] be input layer, F=[f1, f2, fm] be hidden layer mapping function, W=[w1, w2,, wm] and be the weight of hidden layer to output layer.Hidden layer completes the Nonlinear Mapping of input layer to output layer.
Compare with traditional BP neural network, RBF neural has lot of advantages, as speed of convergence is very fast, network size is less, calculated amount is less, has robustness and without local minimum etc.In view of these features, RBF neural identifier can approach the model of controlled device soon, and the change of real-time accurate tracking model, the Neural Network Observer like this based on RBF neural identifier just can the change of tracking in time airplane rotating actuator drive unit.Consider the complex characteristics of airplane rotating actuator drive unit itself and the advantage of RBF neural, the present invention selects RBF neural as observer.
2, observer is set up
Residual error is the difference between rotary actuator drive unit real output value and desired output, can be defined as:
Wherein, γ
iresidual values, u
ireal output value,
it is desired output.
As can be seen from formula (1), residual error size reflects the difference between real output value and desired output, and when rotary actuator drive unit is abnormal, the difference between reality and desired output will become large, and namely residual values becomes large.When drive unit fault, residual error will exceed the threshold residual value of setting, that is, when residual error exceedes predetermined threshold value, decision-making system can occur fault.Therefore, can judge whether system there occurs fault by the relative size relation of the residual error and threshold residual value that compare current state.
In actual applications, the output of system, not only by the impact of system input instruction, often also can be subject to the impact of some uncertain factors such as parameter drift, modeling error, random noise and system state, working condition etc.If ignore the impact of these non-faulting factors, high false alarm rate or low fault detect rate will be caused.In order to address this problem, the present invention proposes the impact that the fault detection method based on adaptive threshold brings to residual error to eliminate non-faulting factor.
Each observer in the present invention comprises two neural networks, and one of them RBF neural is used for the desired output of estimating system to produce residual error, and another RBF neural is for generation of self-adaptation residual error.
3, residual sum threshold generator
As shown in Figure 4, two-stage RBF neural network observer is set up.
First, one-level RBF neural Residual Generation device is set up.Because airplane rotating actuator drive unit is closed-loop control system, be difficult to the parameter value obtaining internal portions, therefore before RBF neural observer input end, increase Z
-1link, carrys out the sluggish link of Reality simulation Control Component lag output function with this; Gather the control signal c under actuator drive unit normal operating conditions
ithe output displacement signal u of (k), previous moment drive unit
iand current time output displacement signal u (k-1)
i(k), k=1,2,3 ..., n.By the drive unit control signal c obtained
ithe output signal u of (k) and previous moment drive unit
i(k-1) the training input amendment of a vector as RBF neural observer is synthesized, by the output signal u of current time drive unit
ik output sample that () trains as observer; Be normalized between [-1,1] for the input of network training, output sample, the basic parameter of RBF neural be set, network is trained.After having trained, preserve the RBF neural observer trained.Then, by control signal c
ithe output displacement signal u of (k) and previous moment drive unit
i(k-1) synthesize a vector and as the input vector of the above-mentioned RBF neural trained after normalization, then can obtain the estimated value of this moment drive unit output displacement
k=1,2,3 ..., n; By the Displacement Estimation value that RBF observer exports
output displacement value u true in the drive unit collected
ik () is poor, then can obtain residual signals r (k) in this moment of drive unit, k=1,2,3 ..., n.
Then, secondary RBF neural adaptive threshold generator is set up.Adaptive threshold refers to that fault threshold inputs the change of instruction and working condition along with system and changes, and synchronously can be obtained by the RBF neural trained.First, the control signal c under actuator drive unit normal operating conditions is gathered
ithe output displacement signal u of (k), previous moment drive unit
iand drive unit current time output displacement signal u (k-1)
i(k); By control signal c
ithe output displacement signal u of (k) and previous moment drive unit
i(k-1) synthesize a vector as the input vector of one-level RBF Residual Generation device after normalization, then can obtain the estimated value that system under normal condition exports
by RBF observer output estimation value
output displacement value u true in the drive unit collected
ik () is poor, obtain residual signals r (k) under drive unit normal condition, k=1,2,3 ..., n, residual signals r (k) obtained under definition normal condition is benchmark residual error th.Then, with drive unit control signal c
ik training input vector that the residual error r (k) under () and normal condition is secondary RBF adaptive threshold maker, using the adaptive threshold expected as the output vector of network.The adaptive threshold expected is defined as:
Wherein,
for expecting threshold value, r (k) is residual error, and b is correction factor.After having trained, preserve the RBF adaptive threshold generator trained.
After the training completing Residual Generation device RBF network, just complete the foundation of bipolar observer, may be used for fault detect.First, test data is input to RBF Residual Generation device, obtains residual values now; Then this residual values and steering order are input to secondary RBF adaptive threshold maker, obtain adaptive threshold value.Relatively the size of residual sum adaptive threshold, if residual values is larger than threshold value, illustrates this airplane rotating actuator fault.The whole process that adaptive failure detects is described in Fig. 5.
The fault isolation of step 3, airplane rotating actuator drive unit:
1, fault isolation
Fault isolation, under referring to real-time working environment, judges its normal operating conditions respectively to the subsystem of system or equipment or ingredient, narrows down to the last technical measures judging out of order subsystem or part.Fault isolation is extremely important for the maintenance management of airplane rotating actuator drive unit.The present invention, on the basis analyzing airplane rotating actuator driving device structure, proposes a kind of the Fault Isolation Strategy based on observer information.
2, the Fault Isolation Strategy
As shown in Figure 2, this control loop comprises two sub-loops, has two angular displacement sensor feedback angle information to control module.First loop contains servo-valve, oil motor and 1 bugle displacement transducer; Second loop contains servo-valve, oil motor and 2 bugle displacement transducers.Thus, twin-stage observer can be set up to monitor this two loops.Because the element comprised in two loops is different, so can carry out fault location according to the output of observer.Fig. 5 gives the strategy of fault isolation.
If it is all normal that testing result shows these two observer signals, just mean that this rotary actuator drive unit is normal.When No. 1 and No. 2 observers all detect fault, then show that servo-valve or oil motor there occurs fault.Only have when No. 1 observer display fault, and 1 bugle displacement transducer fault can be concluded when No. 2 observer displays are normal.When No. 2 observers show fault when No. 1 observer shows normally, 2 bugle displacement transducer faults or reducing gear fault can be judged.The result of algorithms for fault isolation is summarized in table 1.
Fault isolation result under the different testing result of table 1
The health evaluating of step 4, airplane rotating actuator drive unit:
1, CV value and SOM neural network
CV value is an index parameter of valuator device operating condition, can well represent the Evaluation results of rotary actuator drive unit.CV value is the numerical value performance of rotary actuator drive unit be normalized between 0 to 1, and CV value is larger, and equipment is more close to health status.By CV value size, can judge that system is in health status, sub-health state or malfunction.
Self-organizing map neural network SOM is a kind of special competitive type artificial neural network, can be used for mapping multivariate data, probability density and cluster.In SOM neural network, certain output node can be made a certain quasi-mode and being reacted to represent this Pattern Class especially, node adjacent on output layer can make special reflection to close Pattern Class in realistic model distribution, when certain class data pattern inputs, maximal stimulation (triumph node) is produced to a certain output node, larger stimulation is produced to some nodes around triumph node simultaneously.In the process of training, constantly the connection weights of triumph node are adjusted, the connection weights of knot neighborhood of a point node of winning are adjusted simultaneously; Along with the carrying out of training, this contiguous range constantly reduces, and to the last only carries out trickle connection weighed value adjusting to triumph node.The structure of SOM neural network is given in Fig. 6.
2, based on the health evaluating of residual analysis
Represent each neuron of SOM neural network with the weight vector of one dimension, map neural unit is connected with adjacent neuron by a kind of neighborhood relationships, defines mapping topological structure.For the process of training vector X, the distance between applications distances Likelihood Computation this vector and all SOM weight vectors, the nearest vector of distance X is just called best match unit (BMU).The weight vector of BMU and its neighborhood vector are increased by learning rules, and learning rules are as shown in formula (3).
w
i(t+1)＝w
i(t)+α(t)·h
BMU,i(t)(x(t)-w
i(t)) (3)
Wherein, w
it () is weight vector; α (t) is learning rate, and size is between 0 ~ 1; h
bMU, ibe neighborhood equation, determined by the distance between BMU and its neighborhood.After by the trained SOM neural network of the residual error under normal condition, by the residual error training SOM neural network of test data, obtain MQE value, MQE is defined as:
MQE＝||X
input-w
bmu|| (4)
Wherein, X
inputinput data, w
bmuit is the weight vector of BMU.Finally, between normalization MQE value to 0 ~ 1, normalizing equation is as follows:
Wherein, a is scale parameter, is determined by the MQE value under normal condition and the CV value that presets.
In actual applications, the residual values training SOM neural network first utilizing actuator drive unit normal operating conditions next stage RBF Residual Generation device to produce; After training SOM neural network, for a certain moment in real work, get the input of real time data as one-level RBF Residual Generation device, obtain the residual values of real time data, input again using this residual values as SOM neural network, by calculating the size of distance between current residue value and best match unit and MQE value, what quantitatively can draw real time data and normal data departs from situation, and namely equipment running status to be assessed and normal condition distinguish the drift rate in characteristic of correspondence space.Finally, the CV value before the MQE value obtained is normalized to 0-1, with the health status of this characterization device.
In order to verify the validity of put forward the methods of the present invention, instance analysis is as follows:
1, direct fault location
Establish a realistic model, and be filled with 4 kinds of typical faults, be i.e. leak in servo-valve magnetic field intensity reduction, oil motor, Transmission Shaft Stiffness is degenerated and 2 bugle displacement transducer output accuracies exceed standard.This fault mode of fault and concrete mode of direct fault location thereof in 4 is given in table 2.Wherein Q
0represent the flow of servo-valve; C
tmand eta-V
mcoefficient and volumetric efficiency are revealed in representative respectively; ρ represents the rigidity of transmission shaft; α represents the coefficient of 2 bugle displacement transducer precision.
The fault mode that table 2 realistic model injects and concrete direct fault location mode
2, neural network training
In normal state, using the output of control signal and previous moment as input Training RBF Neural Network, training effect as shown in Figure 7 and Figure 8, line above in figure represents adaptive threshold, line below represents residual error, can find out, when rotary actuator drive unit is under normal condition time, fault threshold is all the time higher than fault residual value.
3, inject servo-valve magnetic field intensity to reduce
The fault that servo-valve magnetic field intensity reduces is injected in realistic model.Fig. 9 and Figure 10 gives the testing result of two observers.Can obviously find out from figure, the testing result of two observers is all malfunction, then can according to table 1 localizing faults element to servo-valve or oil motor.
4, fault is leaked in injection liquid pressure motor
Fault is leaked in injection liquid pressure motor in realistic model.Figure 11 and Figure 12 gives the testing result of two observers.Can obviously find out from figure, the testing result of two observers is all malfunction, then can according to table 1 localizing faults element to servo-valve or oil motor.
5, Transmission Shaft Stiffness degradation filture is injected
Transmission Shaft Stiffness degradation filture is injected in realistic model.Figure 13 and Figure 14 gives the testing result of two observers.Can obviously find out from figure, the testing result of No. 1 observer is normal condition, and the testing result of No. 2 observers is malfunctions, thus can according to table 1 localizing faults element to 2 bugle displacement transducer or reducing gears.
6, inject 2 bugle displacement transducer output accuracies to exceed standard fault
In realistic model, inject 2 bugle displacement transducer output accuracies to exceed standard fault.Figure 15 and Figure 16 gives the testing result of two observers.As can be seen from the figure, No. 1 observer testing result is normal, and No. 2 observer testing results are fault, then according to table 1 can localizing faults element to 2 bugle displacement transducer or reducing gears.
Performance evaluation result of the present invention is: the residual values first under application normal condition is to train SOM neural network, then healthy degraded data is input in the neural network trained, then by the method that the present invention proposes, CV value curve as in figs. 17-19 can be obtained.CV value represents the health status of rotary actuator drive unit, here we setting when CV value lower than 0.6 time, equipment reaches malfunction.
To sum up, the invention provides the method for the airplane rotating actuator drive unit fault detect of complete set, isolation and Performance Evaluation.By setting up bipolar RBF observer, producing residual values and adaptive threshold, realizing fault detect and fault isolation according to the control loop of airplane rotating actuator drive unit.Using residual values as the parameter of Performance Evaluation, training SOM neural network, and between normalization performance index CV value to 0 ~ 1, to assess the performance of rotary actuator drive unit.By setting up the realistic model of airplane rotating actuator drive unit in the present invention, and inject 4 typical faults, verify the method proposed, result confirms that this cover method can carry out the fault detect of airplane rotating actuator drive unit, isolation and Performance Evaluation effectively.
Non-elaborated part of the present invention belongs to the known technology of those skilled in the art.
Claims (5)
1. airplane rotating actuator drive unit adaptive failure detect, isolation and the method for health evaluating, concrete steps are as follows:
Step 1, structure by analysis of aircraft rotary actuator drive unit, set up the realistic model of airplane rotating actuator drive unit;
Step 2, fault detection method for airplane rotating actuator drive unit, utilize input instruction signal and angle displacement signal, set up two-stage RBF neural network Residual Generation device and adaptive threshold maker, realize the fault detect of airplane rotating actuator drive unit;
Step 3, failure separation method for airplane rotating actuator drive unit, utilize input instruction signal and angle displacement signal, and application two cover two-stage RBF neural networks set up Failure Observer, realize the localization of fault of airplane rotating actuator drive unit;
Step 4, health evaluating method for airplane rotating actuator drive unit, one-level RBF Residual Generation device is utilized to obtain fault residual value, utilize the residual values training SOM neural network under drive unit normal operating conditions, then utilize the health evaluating of the SOM neural fusion airplane rotating actuator drive unit trained.
2. a kind of airplane rotating actuator drive unit adaptive failure according to claim 1 detect, the method for isolation and health evaluating, it is characterized in that: described airplane rotating actuator drive unit is closed loop feedback system, is substantially made up of control module, servo-valve, oil motor, reducing gear and topworks; In control loop, there are two angular displacement sensors to be used for feedback system information, help execution architecture to obtain optimal angle; In actual applications, the parameter of servo-valve and oil motor is difficult to obtain, and thus in simulations, the data of acquisition are the feedback signal of control signal and reducing gear and the output signal of displacement of drive unit previous moment.
3. a kind of airplane rotating actuator drive unit adaptive failure according to claim 1 and 2 detect, the method for isolation and health evaluating, it is characterized in that: the described fault detection method for airplane rotating actuator drive unit, comprise two-stage RBF neural network, wherein, one-level RBF neural is as Residual Generation device, network is input as the displacement signal of steering order and the output of previous moment system, export the estimated value for drive unit displacement signal, the difference of this estimated value and the actual output displacement signal of system is residual values; And secondary RBF neural is as adaptive threshold maker, be input as the estimated value of Systematical control instruction and drive unit displacement signal, export as adaptive fault threshold; By the relative size of the residual values and adaptive threshold that compare current time, judge the health status of current time equipment, realize the fault detect of drive unit.
4. a kind of airplane rotating actuator drive unit adaptive failure according to claim 1 and 2 detects, the method of isolation and health evaluating, it is characterized in that: the described failure separation method for airplane rotating actuator drive unit, comprise two Failure Observers, each observer is made up of one group of two-stage RBF neural network, wherein one-level RBF neural is as Residual Generation device, network is input as the displacement signal of steering order and the output of previous moment drive unit, export the estimated value for drive unit displacement signal, the difference of this estimated value and the actual output displacement signal of drive unit is residual values, and secondary RBF neural is as adaptive threshold maker, be input as the estimated value of Systematical control instruction and drive unit displacement signal, export as adaptive fault threshold, two observers lay respectively in inside and outside two loops of drive unit closed-loop control system, according to the output information of two observers, judge inner-outer loop whether fault, and then can the concrete generation position of failure judgement, thus realize the fault isolation of drive unit.
5. a kind of airplane rotating actuator drive unit adaptive failure according to claim 1 and 2 detects, the method of isolation and health evaluating, it is characterized in that: the described health evaluating method for airplane rotating actuator drive unit, comprise an one-level RBF neural and a SOM neural network, wherein one-level RBF neural is as Residual Generation device, network is input as the displacement signal of steering order and the output of previous moment drive unit, export the estimated value for drive unit displacement signal, the difference of this estimated value and the actual output displacement signal of drive unit is residual values, the residual values training SOM neural network utilizing drive unit normal operating conditions next stage RBF neural to produce, afterwards, using from the residual values obtained during drive unit actual motion as the input of the SOM neural network trained, calculate minimum quantization error MQE value, and normalization obtains the CV value between 0-1, realizes the health evaluating of drive unit.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410664864.3A CN104850889B (en) | 2014-11-19 | 2014-11-19 | Airplane rotation actuator drive unit adaptive fault detection, isolation and confidences assessment method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410664864.3A CN104850889B (en) | 2014-11-19 | 2014-11-19 | Airplane rotation actuator drive unit adaptive fault detection, isolation and confidences assessment method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104850889A true CN104850889A (en) | 2015-08-19 |
| CN104850889B CN104850889B (en) | 2017-02-22 |
Family
ID=53850521
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410664864.3A Expired - Fee Related CN104850889B (en) | 2014-11-19 | 2014-11-19 | Airplane rotation actuator drive unit adaptive fault detection, isolation and confidences assessment method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104850889B (en) |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105129109A (en) * | 2015-09-30 | 2015-12-09 | 北京航空航天大学 | Method for evaluating health of aircraft aileron actuator system based on multi-fractal theory and self-organizing map (SOM) network |
| CN105335698A (en) * | 2015-09-15 | 2016-02-17 | 河南理工大学 | Gear failure diagnosis method based on adaptive genetic algorithm and SOM (Self-Organizing Map) network |
| CN105354399A (en) * | 2015-12-14 | 2016-02-24 | 北京航空航天大学 | Multidisciplinary and reliable modeling method of hydraulic servo mechanism based on failure mechanism |
| CN105717912A (en) * | 2016-01-27 | 2016-06-29 | 西北工业大学 | Reliability analysis method for electromechanical actuator based on fuzzy dynamic fault tree |
| CN106301610A (en) * | 2016-08-29 | 2017-01-04 | 北京航空航天大学 | The adaptive failure detection of a kind of superhet and diagnostic method and device |
| CN106571882A (en) * | 2016-09-06 | 2017-04-19 | 北京航空航天大学 | Superheterodyne receiver performance degradation assessment and prediction method and system thereof |
| CN108427400A (en) * | 2018-03-27 | 2018-08-21 | 西北工业大学 | A kind of aircraft airspeed pipe method for diagnosing faults based on neural network Analysis design |
| CN108803317A (en) * | 2018-05-08 | 2018-11-13 | 天津大学 | Adaptive multivariable quadrotor drone finite time fault tolerant control method |
| CN109033532A (en) * | 2018-06-29 | 2018-12-18 | 北京航空航天大学 | A kind of zero voltage switch phase-shifting full-bridge power supply health state evaluation method |
| CN111474919A (en) * | 2020-04-27 | 2020-07-31 | 西北工业大学 | Aeroengine control system sensor fault diagnosis method based on AANN network group |
| CN111680356A (en) * | 2020-05-06 | 2020-09-18 | 北京航空航天大学 | Method for evaluating real-time safety performance of unmanned aerial vehicle actuator based on double-stage observer |
| CN112597637A (en) * | 2020-12-11 | 2021-04-02 | 北京自动化控制设备研究所 | Rudder system health state assessment method |
| CN112648140A (en) * | 2020-12-21 | 2021-04-13 | 北京华能新锐控制技术有限公司 | Fault tolerance method for wind turbine generator pitch angle encoder based on signal reconstruction |
| CN112668415A (en) * | 2020-12-17 | 2021-04-16 | 震兑工业智能科技有限公司 | Aircraft engine fault prediction method |
| CN113221441A (en) * | 2020-12-24 | 2021-08-06 | 山东鲁能软件技术有限公司 | Method and device for health assessment of power plant equipment |
| CN113295043A (en) * | 2021-05-21 | 2021-08-24 | 江苏科技大学 | Tank vertical stabilizer neural self-adaptive control method based on BP 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 |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2010125080A2 (en) * | 2009-04-28 | 2010-11-04 | Airbus Operations Gmbh | Device and method for the residual analysis of a residuum to detect system errors in the system behavior of an aircraft |
| CN102606557A (en) * | 2012-01-16 | 2012-07-25 | 北京航空航天大学 | Health evaluation method for hydraulic system based on fault observer and SOM (self-organized mapping) |
| CN102705303A (en) * | 2012-05-16 | 2012-10-03 | 北京航空航天大学 | Fault location method based on residual and double-stage Elman neural network for hydraulic servo system |
| CN103353752A (en) * | 2013-04-12 | 2013-10-16 | 北京航空航天大学 | Aircraft environment control system control component fault diagnosis method based on four-level RBF neural network |
-
2014
- 2014-11-19 CN CN201410664864.3A patent/CN104850889B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2010125080A2 (en) * | 2009-04-28 | 2010-11-04 | Airbus Operations Gmbh | Device and method for the residual analysis of a residuum to detect system errors in the system behavior of an aircraft |
| CN102606557A (en) * | 2012-01-16 | 2012-07-25 | 北京航空航天大学 | Health evaluation method for hydraulic system based on fault observer and SOM (self-organized mapping) |
| CN102705303A (en) * | 2012-05-16 | 2012-10-03 | 北京航空航天大学 | Fault location method based on residual and double-stage Elman neural network for hydraulic servo system |
| CN103353752A (en) * | 2013-04-12 | 2013-10-16 | 北京航空航天大学 | Aircraft environment control system control component fault diagnosis method based on four-level RBF neural network |
Cited By (28)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105335698A (en) * | 2015-09-15 | 2016-02-17 | 河南理工大学 | Gear failure diagnosis method based on adaptive genetic algorithm and SOM (Self-Organizing Map) network |
| CN105129109B (en) * | 2015-09-30 | 2017-05-17 | 北京航空航天大学 | Method for evaluating health of aircraft aileron actuator system based on multi-fractal theory and self-organizing map (SOM) network |
| CN105129109A (en) * | 2015-09-30 | 2015-12-09 | 北京航空航天大学 | Method for evaluating health of aircraft aileron actuator system based on multi-fractal theory and self-organizing map (SOM) network |
| CN105354399A (en) * | 2015-12-14 | 2016-02-24 | 北京航空航天大学 | Multidisciplinary and reliable modeling method of hydraulic servo mechanism based on failure mechanism |
| CN105354399B (en) * | 2015-12-14 | 2018-07-13 | 北京航空航天大学 | A kind of multidisciplinary Reliability Modeling of hydraulic servomechanism based on failure mechanism |
| CN105717912B (en) * | 2016-01-27 | 2018-08-24 | 西北工业大学 | A kind of electromechanical actuator analysis method for reliability based on fuzzy dynamic fault tree |
| CN105717912A (en) * | 2016-01-27 | 2016-06-29 | 西北工业大学 | Reliability analysis method for electromechanical actuator based on fuzzy dynamic fault tree |
| CN106301610A (en) * | 2016-08-29 | 2017-01-04 | 北京航空航天大学 | The adaptive failure detection of a kind of superhet and diagnostic method and device |
| CN106571882A (en) * | 2016-09-06 | 2017-04-19 | 北京航空航天大学 | Superheterodyne receiver performance degradation assessment and prediction method and system thereof |
| CN108427400A (en) * | 2018-03-27 | 2018-08-21 | 西北工业大学 | A kind of aircraft airspeed pipe method for diagnosing faults based on neural network Analysis design |
| CN108803317A (en) * | 2018-05-08 | 2018-11-13 | 天津大学 | Adaptive multivariable quadrotor drone finite time fault tolerant control method |
| CN108803317B (en) * | 2018-05-08 | 2021-08-03 | 天津大学 | Self-adaptive multivariable four-rotor unmanned aerial vehicle finite time fault-tolerant control method |
| CN109033532A (en) * | 2018-06-29 | 2018-12-18 | 北京航空航天大学 | A kind of zero voltage switch phase-shifting full-bridge power supply health state evaluation method |
| CN109033532B (en) * | 2018-06-29 | 2022-09-02 | 北京航空航天大学 | Method for evaluating health state of zero-voltage switching phase-shifted full-bridge power supply |
| CN111474919A (en) * | 2020-04-27 | 2020-07-31 | 西北工业大学 | Aeroengine control system sensor fault diagnosis method based on AANN network group |
| CN111474919B (en) * | 2020-04-27 | 2022-03-25 | 西北工业大学 | Aeroengine control system sensor fault diagnosis method based on AANN network group |
| CN111680356A (en) * | 2020-05-06 | 2020-09-18 | 北京航空航天大学 | Method for evaluating real-time safety performance of unmanned aerial vehicle actuator based on double-stage observer |
| CN111680356B (en) * | 2020-05-06 | 2023-10-10 | 北京航空航天大学 | Unmanned aerial vehicle actuator real-time safety performance evaluation method based on two-stage observer |
| CN112597637A (en) * | 2020-12-11 | 2021-04-02 | 北京自动化控制设备研究所 | Rudder system health state assessment method |
| CN112668415A (en) * | 2020-12-17 | 2021-04-16 | 震兑工业智能科技有限公司 | Aircraft engine fault prediction method |
| CN112648140B (en) * | 2020-12-21 | 2022-03-18 | 北京华能新锐控制技术有限公司 | Fault tolerance method for wind turbine generator pitch angle encoder based on signal reconstruction |
| CN112648140A (en) * | 2020-12-21 | 2021-04-13 | 北京华能新锐控制技术有限公司 | Fault tolerance method for wind turbine generator pitch angle encoder based on signal reconstruction |
| CN113221441A (en) * | 2020-12-24 | 2021-08-06 | 山东鲁能软件技术有限公司 | Method and device for health assessment of power plant equipment |
| CN113221441B (en) * | 2020-12-24 | 2022-12-02 | 山东鲁能软件技术有限公司 | Method and device for health assessment of power plant equipment |
| CN113295043A (en) * | 2021-05-21 | 2021-08-24 | 江苏科技大学 | Tank vertical stabilizer neural self-adaptive control method based on BP neural network |
| CN113295043B (en) * | 2021-05-21 | 2022-10-14 | 江苏科技大学 | Tank vertical stabilizer neural self-adaptive control method based on BP 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 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104850889B (en) | 2017-02-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104850889A (en) | Airplane rotation actuator drive unit adaptive fault detection, isolation and confidences assessment method | |
| Cho et al. | Fault detection and diagnosis of a blade pitch system in a floating wind turbine based on Kalman filters and artificial neural networks | |
| CN107941537B (en) | A kind of mechanical equipment health state evaluation method | |
| CN103983453B (en) | A kind of executing agency of aero-engine and the differentiating method of sensor fault diagnosis | |
| CN104390776B (en) | A kind of redundance aileron actuator fault detect, diagnosis and performance estimating method | |
| CN102705303B (en) | Fault location method based on residual and double-stage Elman neural network for hydraulic servo system | |
| CN101299004B (en) | Vibrating failure diagnosis method based on determined learning theory | |
| Blesa et al. | An interval NLPV parity equations approach for fault detection and isolation of a wind farm | |
| CN108227676A (en) | The online fault detect of valve-controlled cylinder electrohydraulic servo system, estimation and localization method | |
| CN106006344A (en) | Online escalator fault early warning system and fault diagnosis method | |
| CN105129109A (en) | Method for evaluating health of aircraft aileron actuator system based on multi-fractal theory and self-organizing map (SOM) network | |
| CN113092115A (en) | Digital twin model construction method of digital-analog combined drive full-life rolling bearing | |
| CN102589890A (en) | Integrated fault diagnostic method of steam turbine based on CPN (counter-propagation network) and D-S (dempster-shafer) evidences | |
| CN113188794B (en) | Gearbox fault diagnosis method and device based on improved PSO-BP neural network | |
| CN106873568A (en) | Sensor fault diagnosis method based on H infinity robust Unknown Input Observer | |
| CN104018988A (en) | Wind generating set monitoring system based on physical component model and real-time data | |
| Pérez-Pérez et al. | Fault diagnosis in wind turbines based on ANFIS and Takagi–Sugeno interval observers | |
| Zhou et al. | Structural health monitoring of offshore wind power structures based on genetic algorithm optimization and uncertain analytic hierarchy process | |
| CN102749584B (en) | Prediction method for residual service life of turbine generator based on ESN (echo state network) of Kalman filtering | |
| CN117212077B (en) | Wind wheel fault monitoring method, device and equipment of wind turbine and storage medium | |
| CN103245491A (en) | Rotor system fault diagnosis method based on confirmed learning theory | |
| CN105373112A (en) | A steering engine fault detection and diagnosis method based on multi-model parameter estimation | |
| Simani et al. | Fault diagnosis of a wind turbine simulated model via neural networks | |
| CN102253663B (en) | Leakage fault diagnose method and system thereof for networked three-container water tank system | |
| CN102749199A (en) | Method for predicting residual service lives of turbine engines on basis of ESN (echo state network) |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| EXSB | Decision made by sipo to initiate substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20200615 Address after: No. 214, xiaoxinzhuang village, Beifang Town, Huairou District, Beijing Patentee after: Beijing Hengxing Yikang Technology Co.,Ltd. Address before: 100191 Haidian District, Xueyuan Road, No. 37, Patentee before: BEIHANG University |
|
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170222 |