CN103984233A - Four-rotor aircraft dual-granularity fault diagnosis method based on hybrid model - Google Patents

Abstract

The invention relates to a four-rotor aircraft dual-granularity fault diagnosis method based on a hybrid model. The method includes the steps that various physical effects affecting flight of four rotors are analyzed so that modeling categories of a prior model and a non-parameter model can be determined; for the non-parameter model, a nonlinearity degree is measured; according to the nonlinearity degree, an appropriate parameter identification method is selected, and the four-rotor aircraft hybrid model is established; according to the hybrid model, granularity levels of process variables are divided; according to the coarse granularity level, channels where faults occur are judged, components where faults occur are determined according to the fine granularity level, and then dual-granularity fault diagnosis is achieved. The structural features of the hybrid model are fully utilized, and the method is suitable for feasibility testing of process detection and diagnosis of four-rotor helicopter structure faults.

Description

A kind of quadrotor dual-granularity method for diagnosing faults based on mixture model

Technical field

The present invention relates to aviation aircraft fault diagnosis technology field, particularly relate to a kind of quadrotor dual-granularity method for diagnosing faults based on mixture model.

Background technology

Quadrotor is a kind of typical modern complication system, and than fixed wing aircraft, it has more complicated aerodynamic characteristic and state of flight more specifically, needs more high-precision mathematical model and more sane control law to guarantee flight quality and flight safety.

Traditional modeling method of quadrotor is roughly divided into modelling by mechanism and data-driven modeling.Modelling by mechanism parameter interpretation is strong, and model epitaxy is good, but for modern complication systems such as multivariate, non-linear and strong couplings, modeling difficulty is larger; Data-driven modeling does not need the priori of process object, but model accuracy and generalization ability height depend on modeling data.Therefore, only rely on single modeling means to be difficult to obtain high-quality object module.

Quadrotor, by three attitude angle signals of four actuators output, belongs to overdrive system, can effectively improve structural load ability and response speed, and it can inevitably receive external disturbance or break down etc. in operational process; This fault refers to that drive system has at least a characteristic or parameter to occur larger deviation, has exceeded acceptable scope.Now the performance of system is starkly lower than its normal level; The classification of this fault can be carried out from different aspects, and the position occurring from fault, can be divided into actuator failures, sensor fault and structure failure.

Mainly there is following problem in the method for diagnosing faults for quadrotor:

(1) aspect diagnostic method, the method for diagnosing faults of mainly take based on model is as main, and diagnostic procedure has been ignored data characteristic and the partial fault information of a large amount of process variable;

(2) aspect fault type, most of documents are considered the fault diagnosis of quadrotor under actuator failures, a small amount of document has been considered the fault diagnosis of sensor, and structure failure is due to complicated fault model and diversified fault type and seldom studied.

In sum, how to overcome the deficiencies in the prior art and become one of emphasis difficult problem urgently to be resolved hurrily in current aviation aircraft fault diagnosis technology field.

Summary of the invention

The object of the invention is provides a kind of quadrotor dual-granularity method for diagnosing faults based on mixture model for overcoming the existing deficiency of prior art, the present invention is to when the recurring structure fault, the mixture model that utilization is set up based on physical influence analysis and nonlinear metric, by the granularity rank of refining data, can improve the accuracy of fault diagnosis, be applicable to the feasibility checking of the process diagnosis and detection of four-rotor helicopter structure failure.

A kind of quadrotor dual-granularity method for diagnosing faults based on mixture model proposing according to the present invention, is characterized in that it comprises following concrete steps:

Steps A: apply all kinds of physical influences the influence degree of quadrotor is analyzed, according to influence degree, be divided into major influence factors and minor effect factor, determine prior model in quadrotor mixture model and the modeling category of nonparametric model, and set up prior model according to major influence factors;

Step B: for all kinds of nonlinear terms and the coupling terms in minor effect factor analysis nonparametric model, the nonlinear degree of tolerance quadrotor;

Step C: according to the influence degree of physical influence and every nonlinear degree, use indistinct logic computer to select suitable parameter identification, linearization technique, set up the mixture model of quadrotor;

Step D: according to the source of data and self physical significance, the granularity rank of quadrotor is divided;

Step e: adopt the process monitoring method based on pivot analysis, first for other data of coarseness level, determine the passage that fault occurs, then for other data of fine granularity level, orient the component that fault occurs, realize thus dual-granularity fault diagnosis.

The present invention compared with prior art its remarkable advantage is: the one, and the mixture model that the present invention sets up has taken into full account the influence degree of all kinds of physical influences to quadrotor, has avoided due to the model analysis that nonlinear terms and coupling terms cause, the difficulty in processing simultaneously; The 2nd, dual-granularity method for diagnosing faults makes full use of other data message of each particle size fraction, has improved accuracy and the reliability of diagnosis; The 3rd, utilize other data of coarseness level, be accurate to other fault diagnosis of channel level, for the modification of the fault-tolerant control law of different passages provides convenience; The 4th, utilize other data of fine granularity level, be accurate to the method for diagnosing faults of parts, effectively avoid structure failure modeling complexity in conventional fault diagnosis method, had a polynary difficult problem, expanded preferably the scope of fault diagnosis.

Accompanying drawing explanation

Fig. 1 is the step block diagram of a kind of quadrotor dual-granularity method for diagnosing faults based on mixture model of proposing of the present invention.

Fig. 2 is Z=X * Y type 3-D view schematic diagram.

Fig. 3 is type 3-D view schematic diagram.

Fig. 4 is the general process schematic diagram of hybrid modeling.

Fig. 5 is for other fault detect of coarseness level (attitude angle) schematic diagram.

Fig. 6 is for other fault detect of coarseness level (aerodynamic moment) schematic diagram.

Fig. 7 is the contribution rate schematic diagram of attitude angle to fault.

Fig. 8 is the total contribution rate schematic diagram of attitude angle to fault.

Fig. 9 is the contribution rate schematic diagram of aerodynamic moment to fault.

Figure 10 is the total contribution rate schematic diagram of aerodynamic moment to fault.

Figure 11 is the contribution rate schematic diagram of fuselage gyroscopic couple to fault.

Figure 12 is the total contribution rate schematic diagram of fuselage gyroscopic couple to fault.

Figure 13 is for other fault detect of fine granularity level (voltage) schematic diagram.

Figure 14 is for other fault detect of fine granularity level (rotor gyroscopic couple) schematic diagram.

Figure 15 is the contribution rate schematic diagram of rotor gyroscopic couple to fault.

Figure 16 is the total contribution rate schematic diagram of rotor gyroscopic couple to fault.

Embodiment

Below in conjunction with accompanying drawing, the specific embodiment of the present invention is done further and described in detail.

In conjunction with Fig. 1, a kind of quadrotor dual-granularity method for diagnosing faults based on mixture model that the present invention proposes, it comprises all kinds of physical influences of analyzing influence four rotor flyings, to determine the modeling category of prior model and nonparametric model; For nonparametric model, tolerance nonlinear degree; According to nonlinear degree, select suitable parameter identification method, set up quadrotor mixture model; According to the granularity rank of mixture model partition process variable; The passage occurring according to coarseness rank failure judgement, determines according to fine granularity rank the components and parts that fault occurs; Concrete implementation step is as follows:

Steps A: apply all kinds of physical influences the influence degree of quadrotor is analyzed, according to influence degree, be divided into major influence factors and minor effect factor, determine prior model in quadrotor mixture model and the modeling category of nonparametric model, and set up prior model according to major influence factors; Wherein: described all kinds of physical influences, main manifestations is the form of aerodynamic moment, this aerodynamic moment is that the pulling force and the resistance that by quadrotor rotor wing rotation, are produced cause, it is the topmost moment type that aircraft bears, comprise rolling moment, pitching moment and yawing, belong to the modeling category of nonparametric model; By described aerodynamic moment, be major influence factors, can set up prior model formula as follows:

In above formula, represent respectively the angle of pitch, roll angle and crab angle; L is the distance that motor center is put true origin; K _f, K _t,cit is the constant factor between electric moter voltage and torque; V ₁, V ₃, V ₂, V ₄the voltage that four of front, rear, left and right motor produces; J _xx, J _yy, J _zzrepresent respectively roll channel, pitch channel, jaw channel is about x, y, the moment of inertia of z axle.

Choosing quantity of state is controlled quentity controlled variable u=[V ₁v ₃v ₂v ₄] ^t, take state equation formal description quadrotor prior model formula as:

$\begin{array}{c}\stackrel{\·}{x}=\mathrm{Ax}+\mathrm{Bu}\\ y=\mathrm{Cx}\end{array}$

In above formula:

$A=\left[\begin{array}{cccccc}0& 0& 0& 1& 0& 0\\ 0& 0& 0& 0& 1& 0\\ 0& 0& 0& 0& 0& 1\\ 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0\\ 0& 0& 0& 0& 0& 0\end{array}\right];C=\left[\begin{array}{cccccc}1& 0& 0& 0& 0& 0\\ 0& 1& 0& 0& 0& 0\\ 0& 0& 1& 0& 0& 0\end{array}\right];$

$B=\left[\begin{array}{cccc}0& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 0& 0& 0\\ \frac{K_{i,t}}{J_{\mathrm{zz}}}& \frac{K_{i,t}}{J_{\mathrm{zz}}}& \frac{K_{i,t}}{J_{\mathrm{zz}}}& \frac{K_{i,t}}{J_{\mathrm{zz}}}\\ \frac{{\mathrm{lK}}_f}{J_{\mathrm{yy}}}& -\frac{{\mathrm{lK}}_f}{J_{\mathrm{yy}}}& 0& 0\\ 0& 0& \frac{{\mathrm{lK}}_f}{J_{\mathrm{xx}}}& -\frac{{\mathrm{lK}}_f}{J_{\mathrm{xx}}}\end{array}\right];D=0;$

Except this major influence factors of aerodynamic moment, although the suffered gyroscopic couple of system affects relatively faint very important also.Its existence has determined the partial approximation performance of system, belonging to minor effect factor, is the modeling category of nonparametric model, mainly comprises fuselage gyroscopic couple and rotor gyroscopic couple, the above analysis, can set up the more accurate mechanism model formula of quadrotor as follows:

After having considered minor effect factor, nonlinear terms and cross-couplings item in modular form, have been there is.

Step B: for all kinds of nonlinear terms and the coupling terms in minor effect factor analysis nonparametric model, the nonlinear degree of tolerance quadrotor; Wherein: described nonlinear degree refers to: for input signal , a stable causal system N:U _a→ Y, its nonlinear degree be defined as following non-losing side formula:

${\mathrm{\φ}}_N^U=\underset{G\∈Y}{\inf }\underset{u\∈U}{\sup }\frac{{\left|\right|G\left[u\right]-N\left[u\right]\left|\right|}_{\mathrm{VT}}}{{\left|\right|N\left[u\right]\left|\right|}_{\mathrm{VT}}}$

In above formula: G:U _a→ Y is linear operator, norm || || _vTbe defined as choose a series of sinusoidal signals and form input set U _lS:

$\begin{array}{c}{U}_{\mathrm{LB}}=\⟨u|u\left(t\right)=A\sin \left(\mathrm{\ωt}\right),A\∈A,\mathrm{\ω}\∈\mathrm{\Ω}\⟩\\ A=\⟨A\∈R^{+}|A_{\min }\≤A\≤A_{\max }\⟩\\ \mathrm{\Ω}=\⟨\mathrm{\ω}\∈R^{+}|{\mathrm{\ω}}_{\min }\≤\mathrm{\ω}\≤{\mathrm{\ω}}_{\max }\⟩\end{array}$ Stable state output can be expressed as following formula:

$y\left(t\right)=A_0+\underset{k=1}{\overset{\∞}{\mathrm{\Σ}}}{A}_k\sin (\mathrm{k\ωt}+{\mathrm{\φ}}_k)$

According to the definition of nonlinear degree, the computing formula of deriving nonlinear degree lower bound is:

${\mathrm{\φ}}_N^{U_{\mathrm{LB}}}\≥\underset{A\∈A:\mathrm{\ω}\∈\mathrm{\Ω}}{\sup }\frac{1}{A}\sqrt{A_0^2(\mathrm{\ω},A)+\underset{k=2}{\overset{\∞}{\mathrm{\Σ}}}\frac{A_k^2(\mathrm{\ω},A)}{2}}$

In the modeling process of quadrotor, this class minor effect factor of gyroscopic effect belongs to the modeling category of nonparametric model; As the above analysis, should first to the nonlinear degree of its each several part, measure.

Now take roll channel as example, the computation process of nonlinear degree is described.For fuselage gyroscopic effect, its original expression is it is the coupling terms of two quantity of states.Therefore, can this class gyroscopic effect abstract be x * y type; For rotor gyroscopic couple, its original expression is Ω _ifor the rotating speed of rotor, it is proportional to controlled quentity controlled variable voltage V _isquare, therefore, rotor gyroscopic couple is the coupling terms of quantity of state and controlled quentity controlled variable in essence, can be this class gyroscopic effect is abstract type.Fig. 2 and Fig. 3 have described the 3-D view of two class abstract models, can reflect intuitively nonlinear degree separately; Use respectively the sinusoidal signal Asin (α t) of different amplitudes, phase angle and Bsin (β t) to represent x and y, x * y type is further write Asin (α t) * Bsin (β t) type, type is further write be that two class abstract models are chosen abundant representational amplitude and phase angle, can calculate nonlinear degree separately.Table 1 has provided take fuselage gyroscopic couple that roll channel is example and the nonlinear degree result of calculation of rotor gyroscopic couple, wherein, does the result of calculation of first, second row of table 1 represent respectively Taylor series expansion to for nonlinear terms? the nonlinear degree of one and second.

Table 1 gyroscopic effect nonlinear degree calculates

In order to determine suitable linearization technique, after drawing the nonlinear degree of two class gyroscopic effects, need to consider model accuracy and algorithm complexity.The representational linearization technique of four classes is: least-squares algorithm linear fitting (linear fit, LF), equilibrium point Taylor series expansion (series expansion, SE), T-S Fuzzy Inference Model (Takagi-sugeno, TS) and subspace System Discrimination (parameter identification, PI).

Although least-squares algorithm linear fitting algorithm identification efficiency is high, Identification Errors is larger, is a kind of rough data processing method, is applicable to the weak situation of nonlinear degree; Although the accuracy of subspace System Discrimination algorithm is higher, identification efficiency is low, and implementation procedure is complicated, is applicable to the situation that nonlinear degree is stronger; According to variable nonlinear degree separately and the applicable elements of linearization technique, can between nonlinear degree and linearization technique, set up fuzzy rule, as shown in table 2; Choose the mean square deviation of nonlinear degree and output quantity as input quantity, appropriate linearization technique, as output quantity, is set up mamdani Fuzzy inference pattern.

Table 2 fuzzy rule base

For the suffered gyroscopic effect of each passage of quadrotor, by situational variables relation, draw two range of influence of class gyroscopic effect to state equation; Through fuzzy reasoning, from fuzzy rule observation window, output linearization technique is separately as shown in table 3.

The principal linear optimization method of table 3 nonparametric model

According to table 3, nonparametric model each several part is carried out to linearization process, and in conjunction with prior model, draw the state equation of mixture model.In sum, Fig. 4 has provided the general process of hybrid modeling.

Step C: according to the influence degree of physical influence and every nonlinear degree, use indistinct logic computer to select suitable parameter identification, linearization technique, set up the mixture model of quadrotor; As utilize the method for fuzzy reasoning, and between nonlinear degree and parameter identification, linearization technique, set up mapping, the quadrotor mixture model formula finally obtaining based on physical influence analysis and nonlinear metric is as follows:

In above formula, f _sE<>:f _tS<>:f _pI<> represents respectively with SE, TS, and tri-kinds of parameter identification methods of PI are processed all kinds of gyroscopic effect expression formulas in " <> ", specifically as shown in table 6.

The expectation attitude angle of take is 1 °, and reference voltage is U _biasthe operating mode of=2V is example, and the state equation matrix formula of mixture model is as follows:

Step D: according to the source of data and self physical definition, the granularity rank of quadrotor is divided; Embodiment comprises:

Mixture model is compared with single model, except obtaining the routine informations such as output quantity in prior model, quantity of state, can also from nonparametric model, obtain some unknown parameter by intelligent algorithms such as parameter identification, fuzzy reasonings, in conjunction with the mechanism knowledge of prior model modeling process and controlled device, can obtain the physical significance of these unknown parameters; Only from the angle of quantity of information, consider, the data volume of mixture model output is more, means that acquisition failure message is more comprehensive; Yet, from fault diagnosis angle, consider, fault diagnosis need to be divided the granularity rank of data efficiently.

In mixture model, prior model is held the global property of system, low to the degree of refinement of mixture model, therefrom obtains the data message of coarseness level, as system output quantity and quantity of state; Nonparametric model has good partial approximation performance, high to the degree of refinement of mixture model, therefrom obtains the data message of fine granularity level, as the unknown parameter in prior model.For quadrotor, can choose the attitude angle of rolling, pitching and three passages of driftage and aerodynamic moment as other data of coarseness level, choose fuselage gyroscopic couple, rotor gyroscopic couple and electric moter voltage as other data of fine granularity level.

Step e: adopt the process monitoring method based on pivot analysis, first for other data of coarseness level, determine the passage that fault occurs, then for other data of fine granularity level, orient the component that fault occurs, realize thus dual-granularity fault diagnosis; Embodiment comprises:

The process monitoring method of utilization based on pivot analysis, adopts the method for diagnosing faults based on traditional contribution plot, and the square prediction error statistic (SPE) of take is evaluation index; First utilize other data of coarseness level such as angular velocity of quadrotor rolling, pitching and three attitude angle of driftage to determine the approximate range that fault occurs, determine the passage that fault occurs.

At reference voltage 3.5V, expectation attitude angle is under the operating mode of 0.3 °, gathers the data of 15 groups of nominal situations, and every group of 155 sampled points, are used for setting up Principal Component Analysis Model; Under fault condition, gather 12 groups of data, every group of 155 sampled points.

Take a kind of fault type as example, first with other data of coarseness level, carry out fault detect; From Fig. 5 and Fig. 6, three attitude angle and aerodynamic moment be having exceeded and controlled limit in various degree all, fault therefore detected and occur;

Secondly, the method for diagnosing faults of utilization based on traditional contribution plot, for these other data of three classes coarseness level of attitude angle, aerodynamic moment and fuselage gyroscopic couple, Fig. 7, Fig. 8 and Fig. 9 have provided the variation tendency to fault contribution rate at different sampled point variablees with the form of broken line graph; Figure 10, Figure 11 and Figure 12 have added up total contribution rate of each variable fault more intuitively with the form of histogram simultaneously.

The diagnostic result of synthesizing map 7 to Figure 12, roll angle, rolling moment and roll channel fuselage gyroscopic couple are maximum to the contribution rate of fault; Therefore, according to other fault diagnosis result of coarseness level, can infer that fault has occurred roll channel.

For roll channel, other physical quantity of fine granularity level (rotor gyroscopic couple and electric moter voltage) is repeated to above-mentioned troubleshooting step; As shown in Figure 13, voltage does not exceed controls limit, shows that motor does not break down.And while carrying out fault detect for rotor gyroscopic couple, SPE statistic has exceeded controls limit, as shown in figure 14, illustrate that the parts of fault generation are rotors.In sum, according to the Contribution Rate shown in Figure 15 and Figure 16, the parts that break down are front and back rotor.

On this basis, according to other data of fine granularity level such as fuselage gyroscopic effects, finally determine the parts that fault occurs, finally realize the quadrotor dual-granularity method for diagnosing faults based on mixture model.

The present invention, through validation trial, has obtained satisfied effect.

Claims (7)

1. the quadrotor dual-granularity method for diagnosing faults based on mixture model, is characterized in that it comprises following concrete steps:

Steps A: apply all kinds of physical influences the influence degree of quadrotor is analyzed, according to influence degree, be divided into major influence factors and minor effect factor, determine prior model in quadrotor mixture model and the modeling category of nonparametric model, and set up prior model according to major influence factors;

Step B: for all kinds of nonlinear terms and the coupling terms in minor effect factor analysis nonparametric model, the nonlinear degree of tolerance quadrotor;

Step C: according to the influence degree of physical influence and every nonlinear degree, use indistinct logic computer to select suitable parameter identification, linearization technique, set up the mixture model of quadrotor;

Step D: according to the source of data and self physical definition, the granularity rank of quadrotor is divided;

Step e: adopt the process monitoring method based on pivot analysis, first for other data of coarseness level, determine the passage that fault occurs, then for other data of fine granularity level, orient the component that fault occurs, realize thus dual-granularity fault diagnosis.

2. a kind of quadrotor dual-granularity method for diagnosing faults based on mixture model according to claim 1, it is characterized in that all kinds of physical influences described in steps A, refer to the form that shows as aerodynamic moment, this aerodynamic moment is that the pulling force and the resistance that by quadrotor rotor wing rotation, are produced cause, it is the topmost moment type that aircraft bears, the form that comprises rolling moment, pitching moment and yawing, belongs to the modeling category of nonparametric model.

3. a kind of quadrotor dual-granularity method for diagnosing faults based on mixture model according to claim 2, is characterized in that described aerodynamic moment is major influence factors, can set up prior model formula as follows:

In above formula, represent respectively the angle of pitch, roll angle and crab angle; L is the distance that motor center is put true origin; K _f, K _t,cit is the constant factor between electric moter voltage and torque; V ₁, V ₃, V ₂, V ₄the voltage that four of front, rear, left and right motor produces; J _xx, J _yy, J _zzrepresent respectively roll channel, pitch channel, jaw channel is about x, y, the moment of inertia of z axle.

4. a kind of quadrotor dual-granularity method for diagnosing faults based on mixture model according to claim 1, is characterized in that the nonlinear degree described in step B refers to: for input signal a stable causal system N:U _a→ Y, its nonlinear degree be defined as following non-losing side formula:

${\mathrm{\&phi;}}_N^U=\underset{G\&Element;Y}{\inf }\underset{u\&Element;U}{\sup }\frac{{\left|\right|G\left[u\right]-N\left[u\right]\left|\right|}_{\mathrm{VT}}}{{\left|\right|N\left[u\right]\left|\right|}_{\mathrm{VT}}}$

G:U in above formula _a→ Y is linear operator, norm || || _vTbe defined as to after the measuring nonlinearity of the nonlinear terms in nonparametric model and coupling terms, can select suitable linearization, parameter identification method again.

5. a kind of quadrotor dual-granularity method for diagnosing faults based on mixture model according to claim 1, it is characterized in that the mixture model described in step C refers to: the method for utilizing fuzzy reasoning, between nonlinear degree and parameter identification, linearization technique, set up mapping, the quadrotor mixture model formula finally obtaining based on physical influence analysis and nonlinear metric is as follows:

In above formula, the crab angle that represents respectively aircraft, the angle of pitch and roll angle, F<g (x) > represents to process with corresponding parameter identification or linearizing method F the result obtaining after the expression formula g (x) in bracket " <> "; F can select SE, TS, and tri-kinds of methods of PI, SE, TS, PI represents respectively equilibrium point Taylor series expansion, T-S Fuzzy Inference Model and subspace System Discrimination, g (x) is all kinds of expression formulas that represent gyroscopic effect.

6. a kind of quadrotor dual-granularity method for diagnosing faults based on mixture model according to claim 1, it is characterized in that granularity rank described in step D is divided refers to: in mixture model, prior model is held the global property of system, degree of refinement to mixture model is low, therefrom obtain the data message of coarseness level, as system output quantity and quantity of state; Nonparametric model has good partial approximation performance, high to the degree of refinement of mixture model, therefrom obtains the data message of fine granularity level, as the unknown parameter in prior model.

7. a kind of quadrotor dual-granularity method for diagnosing faults based on mixture model according to claim 1, it is characterized in that the dual-granularity method for diagnosing faults described in step e, refer to: utilize the process monitoring method based on pivot analysis, the square prediction error statistic (SPE) of take is carried out fault detect as evaluation index, utilize the method based on traditional contribution plot to carry out fault diagnosis simultaneously, think that the larger variable of contribution rate is the causal variable that causes fault to occur, and specifically comprises:

Step e 1: utilize other data of coarseness level such as angular velocity of quadrotor rolling, pitching and three attitude angle of driftage to determine the approximate range that fault occurs, determine the passage that fault occurs;

Step e 2: determining on the basis of faulty channel, finally determining according to other data of fine granularity level such as fuselage gyroscopic effects the component that fault occurs.