CN106844922A - The engine fire fault diagnosis method with manifold learning is estimated based on cylinder pressure - Google Patents
The engine fire fault diagnosis method with manifold learning is estimated based on cylinder pressure Download PDFInfo
- Publication number
- CN106844922A CN106844922A CN201710020324.5A CN201710020324A CN106844922A CN 106844922 A CN106844922 A CN 106844922A CN 201710020324 A CN201710020324 A CN 201710020324A CN 106844922 A CN106844922 A CN 106844922A
- Authority
- CN
- China
- Prior art keywords
- cylinder pressure
- engine
- cylinder
- matrix
- sample
- 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
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/17—Mechanical parametric or variational design
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Geometry (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Computer Hardware Design (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Combined Controls Of Internal Combustion Engines (AREA)
Abstract
The invention discloses a kind of engine fire fault diagnosis method estimated based on cylinder pressure with manifold learning, cylinder pressure of engines is estimated using imperial Burger sliding mode observer, popular study analysis are carried out by the cylinder pressure signal estimated, to realize engine fire fault diagnosis.Crankshaft dynamic system of the present invention based on engine, it is considered to parameter perturbation present in engine mockup and uncertain problem, establishes nonlinear state equation, devises imperial Burger sliding mode observer, and cylinder pressure is accurately estimated.It is sample with the cylinder pressure signal estimated, extracting cylinder presses the temporal signatures and frequency domain character of signal, the probability distribution of samples points of its farthest is analyzed while the Neighbor Points for considering sample are distributed, during the holding projection algorithm of nearest maximum distance applied into the emulation data and actual test of engine fire state, by recognizing error rate, it was demonstrated that the algorithm is effectively diagnosed to be engine fire state.
Description
Technical field
The invention belongs to automobile engine Misfire Fault Diagnosis method, and in particular to one kind is pressed based on cylinder and estimated and epidemiology
The engine fire fault diagnosis method of habit.
Background technology
The car networking for quickly growing in recent years, is felt comprehensively using advanced intelligent comprehensive technology to road and traffic
Know, realize between multiple systems on a large scale, the interaction of Large Volume Data, traffic whole-process control is carried out to each automobile, to each
Bar road carries out the full-time empty control of traffic, to provide network and the application based on traffic efficiency and traffic safety.Car networking passes through
Long range positioning, reads driving information, the mode such as simple fault message, gives automobile in traveling easily user's body
Test, be vehicle intellectualized important component.However, when failure of the automobile in the urgent complexity of burst in travelling way, at present
Car networking technology still can not well process this problem, need to by vehicle maintenance and repair enterprise carry out specialty automobile diagnosis and repair
Solve.
Vehicle remote monitoring and fault diagnosis system be support vehicles safe operation, improve vehicle use reliability it is important
Device.It is realized to running state of the vehicle real time on-line monitoring using new technology, new tool, grasps vehicle electrically controlling system and run number
According to, carry out in advance automobile operating state data collection and analysis work, it is ensured that automobile failure generation before obtain Rational Maintenance
And maintenance, all there is guidance well for automobile production manufacturer, service dealer, passenger traffic and Shipping enterprises and numerous car owners
Meaning.
The purpose of onboard diagnostic system (OBD) is monitoring automobile exhaust emission system, with the development of OBD, now
More and more apply in fault diagnosis, when automobile breaks down, OBD system can produce DTC, and remote diagnosis system
The key data of system comes from OBD system, and teledata exchange is just carried out only when automobile breaks down, and such case is assigned not
To vehicle remote monitoring.And being based on cylinder pressure most can directly recognize the related real-time status of automobile engine, in real time
The related real-time status of follow-up automobile engine, for the pre- judgement and diagnosis of automobile provide strong information.
The working condition and failure of gasoline engine greatly can be bent with the change of time (or crank angle) by cylinder pressure
Line is reflected.Typically manual type is taken to judge the working condition of gasoline engine, i.e., with the experience combination gasoline engine of people
Work phenomenon judges, such as smoke evacuation, temperature, power output, noise.And some reflection gasoline engine working conditions parameter then without
Method direct access, such as can most reflect the cylinder pressure of gasoline engine work and technology status, in the status monitoring and failure of gasoline engine
It is one of best index of sign gasoline engine running status in diagnosis.Gasoline engine cylinder pressure indicator card is that description gasoline is motor-driven
The basic means of power performance, its concentrated expression gasoline engine exports the heating power transfer process of mechanical work.The cylinder pressure of gasoline engine
Estimation and analysis for improve gasoline engine service behaviour, carry out the optimal IGNITION CONTROL of gasoline engine or oil injection time control, row
Putting control and condition monitoring and fault diagnosis all has very important effect.
From the nineties in 20th century, the research and development of automobile engine Misfire Fault Diagnosis method is rapid, has emerged and has been permitted
Many new methods.Cylinder is lighted a fire successfully each time, and engine can obtain power input, and then cause engine speed fluctuations.Such as
Fruit ignores inertia torque, load torque, friction torque and pumping moment of torsion, then each angular velocity fluctuation and combustion powered generation
Waveform will be directly related, and research velocity variations can provide a kind of method of detection of catching fire.Theoretically, any exception
Failure will reflect in variance causes the change of speed, acceleration or torque.Therefore, the variance or acceleration of research speed can be with
Cylinder pressure based on crankshaft speed is provided and estimates method for diagnosing accidental fire.Because crankshaft speed measuring method is simple, so bent axle
Instantaneous angular velocity is that current most widely used one kind is caught fire basis for estimation, for the instability problem of angular velocity of crankshaft.Compare
In bent axle instantaneous angular velocity, bent axle intermittent angle acceleration ratio angular velocity of crankshaft more can directly reflect the real work operation of engine
State, due to including the prompting message for catching fire in angular acceleration, preferably mistake is can obtain by suitable feature extracting method
Fiery signal.Because crank torque is unable to direct measurement, generally obtained using indirect estimation methods, some documents set up single cylinder respectively
With the engine mockup of multi-cylinder, using the method for parameter Estimation, estimating to the conjunction moment of torsion of cylinder torque and friction torque is realized
Meter, the positioning of caught fire to multicylinder engine judgement and the cylinder that catches fire.These method low costs, advantages of simple structure and simple.Some texts
Offer have studied and catch fire to waste gas, cylinder pressure variation track, indicate average effective moment of torsion, rate of heat release and combustion phase module
Influence, experiment discovery most vat presses the change of corresponding crank angle and burning initial parameters to the no effect of detection of catching fire, and
The cylinder buckling of different crank angles has very strong correlation with generation of catching fire at top dead centre, and then utilizes ANN
Network, for the detection of hcci engine misfire fault, with precision very high.Cylinder pressure can not only inline diagnosis engine
Catch fire and the abnormal combustion phenomenon such as pinking, engine transient air-fuel ratio, time of ignition, waste gas can also be realized by cylinder pressure
The control of the engines such as recycling (EGR).Above-mentioned method is summarized by comparing, based on the engine fire failure that cylinder pressure is estimated
Diagnosis is sensitiveer, more easily a kind of method.Due to not needing any optional equipment, it is allowed to have wide in misfire diagnosis
Prospect.
The content of the invention
Estimated based on cylinder pressure and the popular engine fire fault diagnosis for learning it is an object of the invention to provide a kind of,
Measurement according to engine speed provides a kind of accurate, inexpensive reliable method to obtain the state of engine, reaches hair
The purpose of motivation on-line fault diagnosis.
The present invention is adopted the following technical scheme that to achieve these goals：
Engine fire fault diagnosis method with manifold learning is estimated based on cylinder pressure, is comprised the following steps：
1) engine is analyzed by cylinder Mathematical Modeling, and engine is set up in the form of the differential equation by cylinder model,
It is expressed as：
Wherein, ApRepresent cylinder piston area, LtorThe effective arm of force of the cylinder pressure to axial line is represented, p represents cylinder pressure
Power, R represents crank throw, and ω represents angular velocity of crankshaft, TL(θ) represents total resistive torque, Tl(θ) represents the load of engine
Moment of torsion, J (θ) represents bent axle instant rotation inertia, and V represents cylinder total volume,Fuel combustion heat release rate is represented,Table
Show crankshaft accelerations,The derivative of pressure is represented, α represents scattering losses' proportionality coefficient, m1Represent piston mass and connecting rod quality it
With, γ specific heat ratios,Represent that cylinder controls the derivative of volume.WhereinL is
Length of connecting rod.
2) sliding mode theory is utilized, cylinder pressure dragon Burger sliding mode observer is set up by cylinder model with reference to engine, be expressed as：
Wherein,The observation of cylinder pressure is represented,It is observation error, imperial Burger gain parameter L=(l1
l2)T, sliding formwork gain parameter K=(k1 k2)T,RepresentObservation, TrIt is inertia torque, Tf
It is friction torque, TlIt is load torque.
It is suitable by using the opinion design of Lyapunov stability theorems in above-mentioned imperial Burger sliding mode observer
Luenberger gain parameters, make the Luenberger tracking velocity that can accelerate state estimation, make the sliding formwork can be fine
Related uncertainty and disturbance in ground processing system, while making designed observer error system stable convergence.In order that
Observer can overcome parameter perturbation present in model and uncertainty, and processing parameter disturbance is added in sliding mode observer
And probabilistic sliding formwork.
3) using imperial Burger sliding mode observer, the cylinder pressure to cylinder compression stroke and working stroke is observed, and surveys
Take cylinder pressure signal.Under the different rotating speeds of each misfire fault type of simulated engine, including, normal condition, a cylinder catch fire former
Barrier, a two cylinder misfire fault and one or four cylinder misfire faults, every kind of Status Type devise again 800r/min, 1200r/min,
Operating condition under 2000r/min rotating speeds.The cylinder pressure signal of each Status Type under every kind of rotating speed and idle condition is measured, altogether
12 groups of cylinder pressure signals are obtained, sampling number is 68400, and the time is 12s.
4) temporal signatures and frequency domain character of cylinder pressure signal are extracted, using the holding projection algorithm of nearest maximum distance
Fusion time domain index and frequency-domain index form the transition matrix that low-dimensional feature space training sample and higher-dimension map to lower dimensional space.
5) the current cylinder pressure signal of collection engine, builds corresponding higher-dimension test sample, by higher-dimension test sample
The transition matrix mapped with higher-dimension to lower dimensional space is combined obtains low-dimensional test sample space, then by recognizing clustering method (ginseng
See reference document [1]) the misfire fault state that determines belonging to test sample.
The time domain index includes square average pa, kurtosis pq, average paver, degree of skewness ppx, peak value pm, square amplitude prms、
Waveform index k, peak index k2, pulse index k3With margin index k4；Frequency-domain index includes 0.5 rank amplitude, 1 rank amplitude, 1.5
Rank amplitude and 2 rank amplitudes.
The calculating process of the holding projection algorithm of the nearest maximum distance is as follows：
(1) Neighbor Points and the solstics of the temporal signatures and frequency domain character extracted from cylinder pressure are calculated：Calculate each
Sample point piWith the Euclidean distance between remaining sample point, its nearest k is found outn(pi) individual sample point and farthest kl(pi) individual sample
This point, constructs neighbour's matrix samples and farthest matrix samples respectively；
(2) weighted value is selected according to following formula to neighbour's matrix samples, obtains the weight matrix W ∈ O of neighbour's matrix samplesN×N,
N is sample number, and O represents sample space set；T is thermonuclear parameter, WigRepresent
The weight matrix of nearest samples.
(3) weighted value is selected according to following formula to farthest matrix samples, obtains the weight matrix S ∈ O of farthest matrix samplesN×N；
T is thermonuclear parameter, SigRepresent the weight square of maximum distance sample
Battle array.
(4) characteristic vector of transition matrix A is calculated according to following formula
PZPTA=κ PDPTA
In formula, D is diagonal matrix, is metAnd Z=D-W+S. κ represent the characteristic value of transition matrix, P
Represent sample matrix.A is transition matrix, A=(a1,a2,…,ad)。
(5)a1,a2,…,adIt is characterized value κ1<…<κdCorresponding characteristic vector, d is the intrinsic dimensionality for extracting, d<M, low-dimensional
The sample in space is yi=ATpi。
In order to extract the status information of engine, intelligent diagnostics function is realized, randomly select the sample of total number of samples 60% use
Nearest maximum distance keeps projection algorithm to be trained, and remaining sample is used to test, and uses 1 arest neighbors 1NN[1](1-
Nearest Neighbor) the identification error rate that is calculated of sorting algorithm.
Based on above method, the present invention has advantages below：
(1) based on engine cylinder pressure kinetic model, for the convergence time problem that engine cylinder pressure is estimated,
The present invention devises Luenberger sliding mode observers, and designing observer by Lyapunov Theory of Stability and sliding mode theory increases
Beneficial parameter, realizes the accurate estimation to cylinder pressure.
(2) the NFDPP algorithms of projection are kept to consider the near of each sample point simultaneously based on arest neighbors and farthest sample distance
Adjacent matrix and farthest matrix, may be such that the sample after dimensionality reduction can preferably retain the structural information of initial data.
(3) engine information that is easily obtained by the indirect method of measurement estimates cylinder pressure, due to low cost, simple structure
The advantages of, the indirect method of measurement can write direct ECU, can be the assistance in terms of OBD sets up on-line monitoring or diagnosis.
Bibliography
[1] .2 editions Beijing of Bian Zhaoqi, Zhang Xuegong pattern-recognitions [M]：Publishing house of Tsing-Hua University, 2000.
Brief description of the drawings
Fig. 1 is crank-connecting rod mechanism for engine；V in figurecIt is combustion chamber volume, VsIt is displacement, VdIt is engine
Discharge capacity, D is piston diameter, αθIt is connecting rod angle, θ is crank angle, and l is length of connecting rod, and R is crank throw, and h is piston row
Journey, TDC (Top Dead Center) is top dead centre, and BDC (Bottom Dead Center) is lower dead center；
Fig. 2 is the schematic diagram that observer estimates cylinder pressure；
Fig. 3 is Luenberger sliding mode observer structure charts；
Fig. 4 is in engine failure identification process figure；
Fig. 5 is the estimation effect analogous diagram of present invention cylinder pressure under normal circumstances；
Fig. 6 is the estimation effect analogous diagram in the case of failure of the present invention.
Specific embodiment
It is of the present invention to be estimated based on cylinder pressure and the popular engine fire fault diagnosis method for learning, using imperial Burger
(Luenberger) sliding mode observer estimates that cylinder pressure of engines the measurement according to engine speed provides a kind of accurate,
The reliable method of low cost obtains the state of engine.In the lab, the cylinder pressure that is proposed estimate model be
Carried out in four cylinder SI engines, comprised the following steps that：(in the present inventionRepresent the derivative form of the variable, symbol ∧ tables
Show the observation implication of the variable, the error of symbol~the represent variable observations and actual value).
1) engine is by cylinder modelling
Angular velocity of crankshaft information is extremely important for engine interior condition monitoring, be engine control or diagnostic application most
Frequently system mode, crankshaft dynamic model such as Fig. 1 that the present invention sets up.Cylinder pressure observer needs cylinder pressure
The two systems state of power kinetic model, measurable transient state angular velocity of crankshaft and immesurable cylinder pressure, angular velocity of crankshaft
Information is highly useful for the monitoring of engine interior state, be engine control and the most frequent system mode of diagnostic application it
One, it is considered to friction torque Tf, load torque Tl, bent axle inertia non-linear dynamic model can be represented by the following formula
Wherein ω is angular velocity of crankshaft, and θ is crank angle, JeIt is effective total rotary inertia of engine, indication torque is
Tind, TrIt is inertia torque：
Tind=tPLtor=ApLtorp
Here p, ApAnd LtorThe respectively effective arm of force of cylinder pressure, piston area and pressure to axial line；tPRepresent gas
Body pressure value.
Ap=π D2/4
Wherein, l is length of connecting rod, and R is crank throw, and D is piston diameter.
TrIt is by rotating and moving back and forth the inertia torque for producing, i.e.,(m in formula1
Represent piston mass and connecting rod quality sum, trRepresent reciprocal inertia force,
To sum up to the analysis of engine, the crankshaft dynamic model for engine is arranged：
In formula, TL(θ) is to represent total resistive torque, and J (θ) is rotary inertia in bent axle equal set.
Assuming that each cylinder load is identical, the pressure, temperature between such cylinder and cylinder are consistent, cylinder pressure power
Learning model can be described by monomer thermodynamics burning equation, can be write as：
V represents volume of cylinder,The derivative of cylinder pressure is represented, γ represents specific heat ratio, V·Represent the derivative of volume of cylinder, p
Cylinder pressure is represented, α represents scattering losses' proportionality coefficient.
Fuel combustion heat release rateFor：
It is fuel firing rate, generally all uses empirical equation weber (wiebe) function representation, i.e.,
Wherein here mfRepresent the fuel oil quality of injection, QLHVFuel low heating value, the fuel low heating value of gasoline is QLHV=
43.448MJ/kg, in formula, n is efficiency parameters, and m is form factor, θ0It is burning starting point, △ θ are combustion duration.
Engine is done manual work the stage, and in the presence of cylinder pressure, piston is from top dead centre (TDC) to lower dead center (BDC)
It is mobile, and promote bent axle to rotate acting by connecting rod.In this process, the displacement y of pistonpThe angle, θ rotated with bent axle
Relation is as follows：yp=R (1-cos θ)+l (1-cos αθ), l represents length of connecting rod, αθRepresent link angle.
Cylinder total volume V is by displacement VsWith combustion volume VcComposition, i.e. V=Vs+Vc, wherein combustion volume is：
Vc=pD2S/ (4 (r-1)), s is piston stroke, and r is compression ratio, so
To sum up to the analysis of engine crankshaft dynamic model and engine cylinder pressure model, engine can be obtained
It is by cylinder model：
2) cylinder pressure based on Luenberger sliding mode observers is estimated
Can be deduced by cylinder model by 1) obtaining engine, we can obtain engine nonlinear state system：
Through analysis, from observability order criterion, system is observable.In order that during observer can overcome model
The parameter perturbation and uncertainty of presence, and the tracking velocity of state estimation is improved, the observer structure such as Fig. 3 is given to system
Thinking is made, Luenberger sliding mode observers are devised, Luenberger sliding mode observers include two parts：Accelerate tracking speed
Luenberger of degree and processing parameter disturbance and probabilistic sliding formwork.First to engine Nonlinear System Design
Luenberger sliding mode observers are as follows：
Wherein,It is state x observations, L=(l1 l2)TIt is Luenberger gain parameters.X=(ω p)T,C=(1 0), wherein A11=0,A21=0,
B1=0,
In order that the Luenberger tracking velocity that can accelerate state estimation, defining observation error isBy
The error that above formula can obtain system is：Known by Lyapunov stability theorems, if (A-LC) is
Hurwitz matrixes, the i.e. real part of the characteristic value of (A-LC) are negative, then systematic errorIt is asymptotically stability.Secondly, in order that seeing
Survey device can uncertain and parameter perturbation present in processing system, obtained for Luenberger observers addition sliding formwork
Luenberger sliding mode observers：
Wherein K=(k1 k2)TIt is sliding formwork gain parameter, cylinder pressure can be obtained by cylinder model with reference to engine
Luenberger sliding mode observers are as follows：
WhereinSystematic error can be obtained by contrasting us is：
Wherein, Crankshaft speed error derivative is represented,Cylinder pressure error derivative is represented,Table
Show cylinder pressure, L=(l1 l2)TIt is imperial Burger (Luenberger) gain parameter, sliding formwork gain parameter K=to be designed
(k1 k2)T。
Define sliding-mode surfaceBefore sliding-mode surface is reached, known by sliding mode theory：
Then
When engine speed estimate convergence actual value, that is, when reaching sliding-mode surface,ThenAndAs available from the above equation：
Above formula is updated in systematic error formula can be obtainedMakeAsymptotic convergence to zero, it is necessary toI.e.When piston moves to TDC (top dead centre) nearby, by its matrix
Differentiate that battle array understands that cylinder pressure is unobservable, should now close sliding formwork gain k, only Luenberger in cylinder pressure observer
Gain L portion is modified to pressure.
Coupled what experiment porch was built with the engine of real vehicle based on Luenberger sliding mode observers, and be given
Observer estimates explanatory diagram directly perceived such as Fig. 2 of cylinder pressure, is taken using information such as the distributive values, crank angle obtained from real vehicle
Cylinder pressure state equation is built, the variable information design gas in the information such as distributive value, crank angle and state equation is recycled
Cylinder pressure observer, so as to obtain cylinder pressure observation.Such as Fig. 5 and Fig. 6, the cylinder pressure estimation respectively under normal condition
The cylinder pressure estimate for being worth and catching fire under state, it can be seen that the impact of misfire fault substantially with normal condition under
It is significantly different, its amplitude in the case of catching fire and under normal circumstances different.
3) projection algorithm (Nearest and farthest distance preserving are kept based on farthest recently
Projections NFDPP) engine fire state recognition, the cylinder pressure signal of engine includes abundant information,
Change, the impact of piston of rotating speed and cylinder pressure can effectively be reflected, engine condition prison is carried out using cylinder pressure signal
Survey and fault diagnosis has signal acquisition is easy, operation strategies are wide etc..Misfire fault directly causes engine operational cycle to occur
Change so that the exciting force frequency of engine becomes the other cycle.Such as, a cylinder catches fire can destroy the flat of engine operation
Weighing apparatus, the exciting force that remaining cylinders are produced constitutes the other cycle, also occurs accordingly from the frequency spectrum of the cylinder pressure signal estimated
Change, similarly, when two adjacent cylinder occurring or being separated by two cylinders and catches fire, the frequency spectrum of cylinder pressure signal will also change.As schemed
4, we study the cylinder pressure signal in the case of engine failure, extract corresponding feature, and each spy is merged using manifold learning
Levy index and form the transformation model that low-dimensional feature space training sample and higher-dimension map to lower dimensional space, then estimate present engine
Cylinder pressure signal, and build corresponding higher-dimension sample space, higher-dimension sample space and Mapping and Converting models coupling are obtained
Corresponding low-dimensional sample space, by recognizing clustering method[1]Determine the current state of sample estimates.Simulated engine each
Status Type, Status Type is：Normal condition, a cylinder misfire fault, a two cylinder catch fire and one or four cylinder misfire faults, every kind of state
Type devises the operating condition under 800r/min, 1200r/min, 2000r/min rotating speed again.Measure every kind of rotating speed and unloaded work
The cylinder pressure signal of each type fault under condition, is obtained 12 groups of cylinder pressure signals, and sampling number is 68400, and the time is
12s.For the data matrix for actually obtaining, the input sample matrix P of manifold learning is obtained by calculating various features index, removed
Calculate each sample point piNeighbour's sample relation outside, NFDPP algorithms also contemplate the sample away from the farthest of the sample point
This structural information, that is, algorithm realizes the calculating of maximum distance.Feature space after dimensionality reduction can more fully retain original
Beginning data matrix design feature.Here is that NFDPP algorithms realize step, sample P={ p1,p2,…,pN}∈Rm×NWherein m=14
Number is characterized, N=120 is sample number：
(1) Neighbor Points and solstics are calculated：Calculate each sample point piWith the Euclidean distance between remaining sample point, find out
Its nearest kn(pi) individual sample point and farthest kl(pi) individual sample point, construct neighbour's matrix samples and farthest matrix sample
This.Range formula is：
d(pi,pj)=| | pi-pj||
pi, pjThe sample point of different cylinder pressure characteristics is represented respectively
(2) weighted value is selected according to following formula to neighbour's matrix samples, obtains the weight matrix of neighbour's matrix samples
W∈ON×N。T is thermonuclear parameter, WigRepresent nearest
The weight matrix of adjacent sample
(3) because traditional LPP (Locality Preserving Projection) algorithm only studies its neighbour's sample
Structure and have ignored farthest sample information.For the problem, while NFDPP algorithm research nearest samples, it is considered to farthest sample
This structural information, weighted value is selected according to following formula, obtains the weight matrix S ∈ O of farthest matrix samplesN×N。
T is thermonuclear parameter, SigRepresent the weight square of maximum distance sample
Battle array
(4) characteristic vector of transition matrix A is calculated according to following formula
PZPTA=κ PDPTA
In formula, D is diagonal matrix, is metAnd Z=D-W+S.P represents cylinder pressure characteristics data square
Battle array.
(5) characteristic value and characteristic vector in calculation procedure (4), vector a1,a2,…,adAs characteristic value κ1<…<κdIt is right
The characteristic vector answered, d is the intrinsic dimensionality for extracting, d<m.Therefore, the sample of lower dimensional space is yi=ATpi, wherein A is conversion square
Battle array, A=(a1,a2,…,ad)。
In order to extract the status information of engine, intelligent diagnostics function is realized, randomly select the sample of total number of samples 60%
For training, remaining sample is used to test, and the identification mistake being calculated using 1NN (1-Nearest Neighbor) method
Rate.1NN has the advantages that algorithm is simple, parameter setting is few, is one of method of calculating identification error rate more conventional at present.
We are in the time domain beamformer of analog result as can be seen that for same state, the amplitude of its time domain is with the increasing of rotating speed
Plus and increase, the impact of misfire fault is substantially different from normal condition, the cylinder pressure that cylinder that a cylinder catches fire pressure signal and two cylinders catch fire
The amplitude of signal is different, and two cylinders of different order catch fire, and its cylinder pressure signal is also what is differed.From corresponding spectrogram we
It can also be seen that engine is under each state, the amplitude that engine turns at 2 times of frequency f is maximum, for misfire fault, goes out
The inapparent frequency 0.5f of normal condition, the amplitude of 1.5f are showed.During due to certain cylinder misfire of engine, destruction engine fortune
Capable balance, makes engine remaining cylinders gas blow-through exciting force constitute another period of change, so 0.5f is occurred in that, f, 1.5f
Frequency content.So the cylinder pressure signal estimated by more than is uniformly divided into 10 sections, and the analytical calculation time domain index according to more than：
Square average pa, kurtosis pq, average paver, degree of skewness ppx, peak value pm, square amplitude prms, waveform index k, peak index k2, pulse
Index k3, margin index k4, frequency-domain index：0.5 rank amplitude, 1 rank amplitude, 1.5 rank amplitudes, 2 rank amplitudes, totally 14 indexs.Finally
Obtain 14 × 120 sample matrix.By to real vehicle Platform Analysis, in time domain waveform, the impact amplitude of misfire fault is obvious
Under more than normal condition, in the spectrogram of frequency domain, it can be seen that appear in what normal condition did not had in the case of misfire fault
The frequency content of 0.5f, f, 1.5f.By experimental analysis, when our training samples are 72 (equivalent to the 60% of total sample), low
When dimension dimension is 2~10, farthest recently to keep projection algorithm to obtain preferable recognition effect, minimum identification error rate is 0.
NFDPP algorithms are due to while consider the nearest and farthest structural information of sample data so that the sample after dimensionality reduction can be more
Ground retains the distributed intelligence of initial data, obtains more preferable recognition result.Due to have on engine engine speed sensor and
Camshaft-signal sensor, the then engine fire fault diagnosis algorithm estimated based on cylinder pressure can write direct ECU, by estimating gas
Cylinder pressure is analyzed, and the assistance in terms of on-line monitoring or diagnosis can be set up for OBD.
Claims (6)
1. the engine fire fault diagnosis method with manifold learning is estimated based on cylinder pressure, it is characterised in that：Comprise the following steps：
1) engine is analyzed by cylinder Mathematical Modeling, and engine is set up in the form of the differential equation by cylinder model, represented
For：
Wherein, ApRepresent cylinder piston area, LtorThe effective arm of force of the cylinder pressure to axial line is represented, p represents cylinder pressure, R
Crank throw is represented, ω represents angular velocity of crankshaft, TL(θ) represents total resistive torque, and J (θ) represents bent axle instant rotation inertia,
V represents cylinder total volume,Fuel combustion heat release rate is represented,Represent crankshaft accelerations,Represent the derivative of pressure, α
Represent scattering losses' proportionality coefficient, m1Expression piston mass and connecting rod quality sum, γ specific heat ratios,Represent cylinder control volume
Derivative；WhereinL is length of connecting rod；
2) sliding mode theory is utilized, cylinder pressure dragon Burger sliding mode observer is set up by cylinder model with reference to engine, be expressed as：
Wherein,The observation of cylinder pressure is represented,It is observation error, imperial Burger gain parameter L=(l1 l2)T, it is sliding
Mould gain parameter K=(k1 k2)T；
3) using imperial Burger sliding mode observer, the cylinder pressure to cylinder compression stroke and working stroke is observed, and measures gas
Cylinder pressure signal；
4) temporal signatures and frequency domain character of cylinder pressure signal are extracted, is merged using the holding projection algorithm of nearest maximum distance
Time domain index and frequency-domain index form the transition matrix that low-dimensional feature space training sample and higher-dimension map to lower dimensional space；
5) the current cylinder pressure signal of collection engine, builds corresponding higher-dimension test sample, by higher-dimension test sample and height
The transition matrix combination for tieing up lower dimensional space mapping obtains low-dimensional test sample space, then by recognizing that clustering method determines test
Misfire fault state belonging to sample.
2. the engine fire fault diagnosis method with manifold learning, its feature are estimated based on cylinder pressure according to claim 1
It is：The time domain index includes square average pa, kurtosis pq, average paver, degree of skewness ppx, peak value pm, square amplitude prms, waveform
Index k, peak index k2, pulse index k3With margin index k4；Frequency-domain index includes 0.5 rank amplitude, 1 rank amplitude, 1.5 rank width
Value and 2 rank amplitudes.
3. the engine fire fault diagnosis method with manifold learning, its feature are estimated based on cylinder pressure according to claim 1
It is：The calculating process of the holding projection algorithm of the nearest maximum distance is as follows：
(1) Neighbor Points and solstics are calculated：Calculate each sample point piWith the Euclidean distance between remaining sample point, it is found out most
Near kn(pi) individual sample point and farthest kl(pi) individual sample point, neighbour's matrix samples and farthest matrix samples are constructed respectively；
(2) weighted value is selected according to following formula to neighbour's matrix samples, obtains the weight matrix of neighbour's matrix samplesT is thermonuclear parameter, WigRepresent the weight of nearest samples
Matrix；
(3) weighted value is selected according to following formula to farthest matrix samples, obtains the weight matrix S ∈ O of farthest matrix samplesN×N；
T is thermonuclear parameter, SigRepresent the weight matrix of maximum distance sample；
(4) characteristic vector of transition matrix A is calculated according to following formula
PZPTA=κ PDPTA
In formula, D is diagonal matrix, is metAnd Z=D-W+S. κ represent the characteristic value of transition matrix, P is represented
Cylinder pressure characteristics data matrix；A is transition matrix, A=(a1,a2,…,ad)；
(5)a1,a2,…,adIt is characterized value κ1<…<κdCorresponding characteristic vector, d is the intrinsic dimensionality for extracting, d<M, lower dimensional space
Sample be yi=ATpi。
4. the engine fire fault diagnosis method with manifold learning is estimated based on cylinder pressure according to claim 1 or 2 or 3,
It is characterized in that：Step 3) also under the different rotating speeds including each misfire fault type of simulated engine, estimate every kind of rotating speed and sky
The cylinder pressure signal of each Status Type under load operating mode.
5. the engine fire fault diagnosis method with manifold learning, its feature are estimated based on cylinder pressure according to claim 4
It is：The engine fire fault type includes, normal condition, a cylinder misfire fault, a two cylinder misfire fault and one or four cylinders
Misfire fault.
6. the engine fire fault diagnosis side of estimation and manifold learning is pressed according to claim 1 or 2 or 3 or 5 based on cylinder
Method, it is characterised in that：Also include calculating identification error rate using 1NN methods.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710020324.5A CN106844922A (en) | 2017-01-11 | 2017-01-11 | The engine fire fault diagnosis method with manifold learning is estimated based on cylinder pressure |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710020324.5A CN106844922A (en) | 2017-01-11 | 2017-01-11 | The engine fire fault diagnosis method with manifold learning is estimated based on cylinder pressure |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106844922A true CN106844922A (en) | 2017-06-13 |
Family
ID=59123073
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710020324.5A Pending CN106844922A (en) | 2017-01-11 | 2017-01-11 | The engine fire fault diagnosis method with manifold learning is estimated based on cylinder pressure |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106844922A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107989708A (en) * | 2017-12-08 | 2018-05-04 | 天津大学 | Self study engine torque control system and its method based on disturbance observation |
| CN108448983A (en) * | 2018-02-01 | 2018-08-24 | 西安理工大学 | A method of inhibiting the extremely low fast non-linear hour interference of servo-drive system |
| CN109447236A (en) * | 2018-09-28 | 2019-03-08 | 重庆邮电大学 | A kind of method for diagnosing faults of hybrid vehicle heat management system |
| CN110635686A (en) * | 2019-11-14 | 2019-12-31 | 东北电力大学 | Control and fault detection method of boost circuit based on switching system |
| CN110823474A (en) * | 2019-09-27 | 2020-02-21 | 一汽解放汽车有限公司 | Fuel system leakage degree evaluation method and storage medium |
| CN111413098A (en) * | 2020-01-13 | 2020-07-14 | 哈尔滨工程大学 | Natural gas engine combustion system nonlinear feature identification method based on IMEP time sequence |
| CN112257747A (en) * | 2020-09-16 | 2021-01-22 | 北京工业大学 | Diagnostic method based on compressed data and supervised global-local/non-local analysis |
| WO2021195948A1 (en) * | 2020-03-31 | 2021-10-07 | 潍柴动力股份有限公司 | Engine misfire detection method and apparatus, device, storage medium, and program |
| CN115467752A (en) * | 2021-06-11 | 2022-12-13 | 广州汽车集团股份有限公司 | Method, system and computer storage medium for diagnosing and analyzing fire of automobile engine |
| CN114441179B (en) * | 2021-12-28 | 2023-11-17 | 潍柴动力股份有限公司 | Engine combustion non-uniformity detection system and detection method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102116242A (en) * | 2010-12-30 | 2011-07-06 | 天津锐意泰克汽车电子有限公司 | Method for diagnosing engine misfire |
| JP2012207538A (en) * | 2011-03-29 | 2012-10-25 | Honda Motor Co Ltd | Fault diagnosis method, fault diagnosis system and fault diagnosis device for engine |
| US20150192498A1 (en) * | 2014-01-08 | 2015-07-09 | Fuji Jukogyo Kabushiki Kaisha | Engine Misfiring Diagnosis Device |
| CN105760839A (en) * | 2016-02-22 | 2016-07-13 | 重庆大学 | Bearing fault diagnosis method based on multi-feature manifold learning and support vector machine |
| CN105865794A (en) * | 2016-05-12 | 2016-08-17 | 长安大学 | Engine misfire fault diagnosis method based on short-time Fourier transformation and principal component analysis |
-
2017
- 2017-01-11 CN CN201710020324.5A patent/CN106844922A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102116242A (en) * | 2010-12-30 | 2011-07-06 | 天津锐意泰克汽车电子有限公司 | Method for diagnosing engine misfire |
| JP2012207538A (en) * | 2011-03-29 | 2012-10-25 | Honda Motor Co Ltd | Fault diagnosis method, fault diagnosis system and fault diagnosis device for engine |
| US20150192498A1 (en) * | 2014-01-08 | 2015-07-09 | Fuji Jukogyo Kabushiki Kaisha | Engine Misfiring Diagnosis Device |
| CN105760839A (en) * | 2016-02-22 | 2016-07-13 | 重庆大学 | Bearing fault diagnosis method based on multi-feature manifold learning and support vector machine |
| CN105865794A (en) * | 2016-05-12 | 2016-08-17 | 长安大学 | Engine misfire fault diagnosis method based on short-time Fourier transformation and principal component analysis |
Non-Patent Citations (2)
| Title |
|---|
| 李巍华 等: "基于最近最远距离保持投影算法的发动机失火状态识别", 《机械工程学报》 * |
| 郑太雄 等: "基于Luenberger滑模观测器的气缸压力估计", 《内燃机工程》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107989708A (en) * | 2017-12-08 | 2018-05-04 | 天津大学 | Self study engine torque control system and its method based on disturbance observation |
| CN108448983A (en) * | 2018-02-01 | 2018-08-24 | 西安理工大学 | A method of inhibiting the extremely low fast non-linear hour interference of servo-drive system |
| CN108448983B (en) * | 2018-02-01 | 2021-08-06 | 西安理工大学 | Method for inhibiting nonlinear interference of servo system at extremely low speed |
| CN109447236A (en) * | 2018-09-28 | 2019-03-08 | 重庆邮电大学 | A kind of method for diagnosing faults of hybrid vehicle heat management system |
| CN110823474A (en) * | 2019-09-27 | 2020-02-21 | 一汽解放汽车有限公司 | Fuel system leakage degree evaluation method and storage medium |
| CN110635686A (en) * | 2019-11-14 | 2019-12-31 | 东北电力大学 | Control and fault detection method of boost circuit based on switching system |
| CN111413098A (en) * | 2020-01-13 | 2020-07-14 | 哈尔滨工程大学 | Natural gas engine combustion system nonlinear feature identification method based on IMEP time sequence |
| WO2021195948A1 (en) * | 2020-03-31 | 2021-10-07 | 潍柴动力股份有限公司 | Engine misfire detection method and apparatus, device, storage medium, and program |
| CN112257747A (en) * | 2020-09-16 | 2021-01-22 | 北京工业大学 | Diagnostic method based on compressed data and supervised global-local/non-local analysis |
| CN112257747B (en) * | 2020-09-16 | 2024-05-28 | 北京工业大学 | Diagnostic method based on compressed data and supervised global-local/non-local analysis |
| CN115467752A (en) * | 2021-06-11 | 2022-12-13 | 广州汽车集团股份有限公司 | Method, system and computer storage medium for diagnosing and analyzing fire of automobile engine |
| CN115467752B (en) * | 2021-06-11 | 2024-05-28 | 广州汽车集团股份有限公司 | Method, system and computer storage medium for diagnosing and analyzing fire of automobile engine |
| CN114441179B (en) * | 2021-12-28 | 2023-11-17 | 潍柴动力股份有限公司 | Engine combustion non-uniformity detection system and detection method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106844922A (en) | The engine fire fault diagnosis method with manifold learning is estimated based on cylinder pressure | |
| CN105736140B (en) | A kind of diesel engine flash speed measures and cylinder stops working trouble-shooter and method | |
| US11454567B2 (en) | Fault diagnosis method of reciprocating machinery based on keyphasor-free complete-cycle signal | |
| CN106338406A (en) | On-line monitoring and fault early-warning system and method for traction electric transmission system of train | |
| CN108407816B (en) | Method and system for evaluating driving of automobile driver | |
| Wang et al. | Research on the fault monitoring method of marine diesel engines based on the manifold learning and isolation forest | |
| CN203083829U (en) | Automobile fault diagnosis system based on tail gas discharge detection adopting simple driving mode | |
| CN103813943A (en) | Method and apparatus for estimating fuel consumption of vehicle | |
| Shrivastava et al. | Application of soft computing in the field of internal combustion engines: a review | |
| CN102518521B (en) | Torque estimation method for vehicle engine | |
| CN103091112A (en) | Method and device of car emission fault detection and diagnosis based on fuzzy reasoning and self-learning | |
| Zheng et al. | Real-time combustion torque estimation and dynamic misfire fault diagnosis in gasoline engine | |
| CN101892918A (en) | The fuel that utilizes cylinder pressure sensor to carry out sprays and the combustion failure diagnosis | |
| CN109613905A (en) | A kind of method and apparatus of the high oil consumption bad working environments of Dynamic Recognition heavy-duty commercial vehicle actual motion | |
| CN106872174A (en) | Automobile transmission rack Knock test engine transient cycle moment of torsion analogy method | |
| CN107798472A (en) | Vehicle power flow distribution and the analysis method of oil consumption factor of influence evaluation | |
| CN108319739A (en) | A kind of Chaos-RBF Neural Network petrol engine air admission method for predicting | |
| EP1365129B1 (en) | Method and apparatus for controlling an internal combustion engine | |
| CN104677629A (en) | Fault detection method for vehicle transmission | |
| US6873902B2 (en) | Non-intrusive method of measuring engine operating parameters | |
| CN101550879A (en) | Drive chain | |
| Du et al. | Research on the application of artificial intelligence method in automobile engine fault diagnosis | |
| CN106339720A (en) | Automobile engine failure detection method | |
| CN111911307A (en) | Two-stage adjustable supercharging system of variable altitude diesel engine, oil injection system and cooling system optimal matching method | |
| Leonhardt et al. | Real time supervision of diesel engine injection with RBF-based neural networks |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170613 |
|
| RJ01 | Rejection of invention patent application after publication |